Philosophy of Psychology: An Introduction

• Cover
• Dedication
• Title Page
• Copyright Page
• Detailed Contents
• Acknowledgements
• Introduction
• An Overview
• Philosophy of Psychology
• What is Philosophy of Psychology?
• Philosophy of Psychology and Philosophy of Mind
• Philosophy of Psychology and Philosophy of Science
• Foundational and Implicational
• Why Do We Need Philosophy of Psychology?
• Evaluating Psychological Studies
• Replication
• Research Participants
• Ecological Validity
• Further Resources
• Articles and Books
• Online Resources
• Questions
• 1 Rationality
  • 1.1 Introduction
  • 1.2 Clarifying Rationality
  • 1.3 Systematic Biases and Errors
  • Wason Selection Task
  • Conjunction Fallacy
  • Base-Rate Neglect
  • Preference Reversal
  • 1.4 Pessimism about Rationality
  • Making Sense of the Results
  • Argument for Pessimism
  • 1.5 Objections to Pessimism
  • The Feasibility Objection
  • The Meaninglessness Objection
  • The Ecological Rationality Objection
  • 1.6 The Aim of Cognition
  • Aiming at Truth
  • Positive Illusions
  • 1.7 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 2 Self-Knowledge
  • 2.1 Introduction
  • 2.2 Clarifying Self-Knowledge
  • Self-Knowledge and Its Targets
  • Self-Knowledge: Privileged and Peculiar
  • 2.3 Challenges to Peculiarity
  • Peculiarity and Parity
  • Dissonance Studies
  • Confabulation Studies
  • Success and Failure of Self-Knowledge
  • 2.4 The Moderate View
  • 2.5 The Extreme Parity View
  • Extreme Parity and Parsimony
  • The Interpretive Sensory-Access Theory
  • 2.6 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 3 Duality
  • 3.1 Introduction
  • 3.2 The Dual-Process Theory
  • Basic Ideas
  • Empirical Reasons: Reasoning Biases
  • Wason Selection Task
  • Belief Bias
  • Philosophical Reasons: The Rationality Paradox
  • 3.3 Processes and Interactions
  • Type-1 and Type-2 Processes
  • Interaction between Type-1 and Type-2 Processes
  • 3.4 The Dual-System Theory
  • Exactly Two Systems?
  • Two Agents?
  • Sloman on Criterion S
  • Davidson on Mental Partitioning
  • 3.5 The Dual-State Theory
  • Two States
  • Non-Doxastic Dual-State Theory
  • Gendler on Aliefs and Beliefs
  • Doxastic Dual-State Theory
  • Frankish on Type-1 and Type-2 Beliefs
  • Schwitzgebel on In-Between Beliefs
  • 3.6 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 4 Moral Judgment
  • 4.1 Introduction
  • 4.2 Harm and Emotion
  • Moral and Conventional
  • Moral–Conventional Tasks
  • Morality and Harm
  • The Dog Eating Study
  • Morality and Affect
  • 4.3 Interaction between Affective Processes and Reasoning Processes
  • Models of Interaction
  • More Than a Post Hoc Rationalization
  • The Trolley/Footbridge Study
  • 4.4 Affective Processes and Reasoning Processes
  • The Mapping Thesis
  • Content Interpretation
  • Commitment Interpretation
  • 4.5 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 5 Moral Motivation and Behaviour
  • 5.1 Introduction
  • 5.2 The Empathy–Altruism Hypothesis
  • Empathy
  • Altruism
  • 5.3 Altruistic Motivation or Aversive-Arousal Reduction?
  • The Katie Banks Experiment
  • The Elaine Experiment
  • 5.4 The Empathy–Benefit Hypothesis
  • Empathy and Its Consequences
  • Empathy and Its Biases
  • 5.5 Responding to the Challenge
  • Revising the Hypothesis
  • Revising Response 1 (Correcting Empathy)
  • Revising Response 2 (Full Empathy)
  • Biting the Bullet
  • Biting the Bullet Response 1 (Maximizing Local Happiness)
  • Biting the Bullet Response 2 (Partial Obligation)
  • 5.6 Is Empathy Particularly Problematic?
  • 5.7 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 6 Free Will and Responsibility
  • 6.1 Introduction
  • 6.2 Varieties of Free Will Scepticism
  • Materialism and Determinism
  • Epiphenomenalism
  • 6.3 Empirical Evidence for Epiphenomenalism?
  • Libet-Style Studies
  • The Libet Experiment
  • Situationist Studies
  • The Good Samaritan Study
  • Wegner-Style Studies
  • The I Spy Experiment
  • 6.4 Implicit Bias and Responsibility
  • Implicit Bias: A Case Study
  • The Implicit Association Test
  • Awareness and Control
  • 6.5 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 7 Delusion and Confabulation
  • 7.1 Introduction
  • 7.2 Delusion
  • What Are Delusions?
  • How Are Delusions Formed?
  • Multiple Factors
  • Prediction Errors
  • Are Delusions Beliefs?
  • 7.3 Delusion and Irrationality
  • Is Delusional Reasoning Irrational?
  • The Jumping-to-Conclusion Bias
  • Are Delusions Irrational in a Distinctive Way?
  • 7.4 Confabulation
  • What Is Confabulation?
  • Does Confabulation Distort Reality in a Distinctive Way?
  • How Do We Remember?
  • 7.5 Summary
  • Further Resources
  • Articles and Books
  • Online Resources
  • Questions
• 8 Autism and Psychopathy
  • 8.1 Introduction
  • 8.2 Autism and Mindreading
  • Autism
  • The Sally–Anne Experiment
  • Simulation and Theory
  • Do Problems with Mindreading Explain Autism?
  • The Eye-Tracking Mindreading Study
  • 8.3 Psychopathy and Empathy
  • Psychopathy
  • Empathy Impairments
  • Do Empathy Impairments Explain Psychopathy?
  • 8.4 Psychopathy and Moral Judgment
  • Psychopathy and the Moral–Conventional Distinction
  • Psychopathy and Sacrificial Dilemmas
  • 8.5 Summary
  • Further Resources
  • Articles and Books
  • Questions
• Conclusion
• References
  • Introduction
  • Chapter 1
  • Chapter 2
  • Chapter 3
  • Chapter 4
  • Chapter 5
  • Chapter 6
  • Chapter 7
  • Chapter 8
  • Conclusion
• Index
• End User License Agreement

List of Tables

• Chapter 3
  • Table 1. Type-1 and Type-2 properties (Evans 2008)
• Chapter 4
  • Table 2. The mapping thesis

List of Illustrations

• Chapter 1
  • Figure 1. Wason selection task with abstract options
  • Figure 2. Wason selection task with concrete options

For Keiko Miyazono and Kenichi Miyazono

For Ennia Scarduelli and Adalberto Bortolotti

Copyright © Kengo Miyazono and Lisa Bortolotti 2021

The right of Kengo Miyazono and Lisa Bortolotti to be identified as Authors of this Work has been asserted in accordance with the UK Copyright, Designs and Patents Act 1988.

First published in 2021 by Polity Press

Polity Press 65 Bridge Street Cambridge CB2 1UR, UK

Polity Press 101 Station Landing Suite 300 Medford, MA 02155, USA

All rights reserved. Except for the quotation of short passages for the purpose of criticism and review, no part of this publication may be reproduced, stored in a retrieval system or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the publisher.

ISBN-13: 978-1-5095-1547-9 ISBN-13: 978-1-5095-1548-6 (pb)

A catalogue record for this book is available from the British Library.

by Fakenham Prepress Solutions, Fakenham, Norfolk NR21 8NL

The publisher has used its best endeavours to ensure that the URLs for external websites referred to in this book are correct and active at the time of going to press. However, the publisher has no responsibility for the websites and can make no guarantee that a site will remain live or that the content is or will remain appropriate.

Every effort has been made to trace all copyright holders, but if any have been overlooked the publisher will be pleased to include any necessary credits in any subsequent reprint or edition.

For further information on Polity, visit our website: politybooks.com

• Acknowledgements
• Introduction
  • An Overview
  • Philosophy of Psychology
  • Evaluating Psychological Studies
  • Further Resources
  • Questions
• 1 Rationality
  • 1.1 Introduction
  • 1.2 Clarifying Rationality
  • 1.3 Systematic Biases and Errors
  • 1.4 Pessimism about Rationality
  • 1.5 Objections to Pessimism
  • 1.6 The Aim of Cognition
  • 1.7 Summary
  • Further Resources
  • Questions
• 2 Self-Knowledge
  • 2.1 Introduction
  • 2.2 Clarifying Self-Knowledge
  • 2.3 Challenges to Peculiarity
  • 2.4 The Moderate View
  • 2.5 The Extreme Parity View
  • 2.6 Summary
  • Further Resources
  • Questions
• 3 Duality
  • 3.1 Introduction
  • 3.2 The Dual-Process Theory
  • 3.3 Processes and Interactions
  • 3.4 The Dual-System Theory
  • 3.5 The Dual-State Theory
  • 3.6 Summary
  • Further Resources
  • Questions
• 4 Moral Judgment
  • 4.1 Introduction
  • 4.2 Harm and Emotion
  • 4.3 Interaction between Affective Processes and Reasoning Processes
  • 4.4 Affective Processes and Reasoning Processes
  • 4.5 Summary
  • Further Resources
  • Questions
• 5 Moral Motivation and Behaviour
  • 5.1 Introduction
  • 5.2 The Empathy–Altruism Hypothesis
  • 5.3 Altruistic Motivation or Aversive-Arousal Reduction?
  • 5.4 The Empathy–Benefit Hypothesis
  • 5.5 Responding to the Challenge
  • 5.6 Is Empathy Particularly Problematic?
  • 5.7 Summary
  • Further Resources
  • Questions
• 6 Free Will and Responsibility
  • 6.1 Introduction
  • 6.2 Varieties of Free Will Scepticism
  • 6.3 Empirical Evidence for Epiphenomenalism?
  • 6.4 Implicit Bias and Responsibility
  • 6.5 Summary
  • Further Resources
  • Questions
• 7 Delusion and Confabulation
  • 7.1 Introduction
  • 7.2 Delusion
  • 7.3 Delusion and Irrationality
  • 7.4 Confabulation
  • 7.5 Summary
  • Further Resources
  • Questions
• 8 Autism and Psychopathy
  • 8.1 Introduction
  • 8.2 Autism and Mindreading
  • 8.3 Psychopathy and Empathy
  • 8.4 Psychopathy and Moral Judgment
  • 8.5 Summary
  • Further Resources
  • Questions
• Conclusion
• References
• Index

Our greatest debt is to Yukihiro Nobuhara: if he had not invited Lisa to present her work on delusions at the University of Tokyo in October 2010, then Kengo and Lisa wouldn’t have met and discovered common interests and potential areas of collaboration.

Next, we are grateful to the Japan Society for the Promotion of Science: Kengo was funded by one of their international fellowships to visit Lisa at the University of Birmingham. There, he completed his research on delusions and actively participated in Lisa’s projects. Being colleagues enabled Lisa and Kengo to talk about belief, delusion, and rationality together, and also become friends. Lisa’s European Research Council project, PERFECT, made it possible for the collaboration to continue, at a distance. Kengo made significant contributions to the project outputs, and in particular to the notion of epistemic innocence that was developed as part of that project.

We thank anonymous reviewers for helpful and insightful suggestions. We are also grateful to Uku Tooming for detailed comments on the final manuscript, and to Benjamin Costello for his excellent proofreading. This work was supported by JSPS KAKENHI (16H06998, 18H00605).

This book project started in 2015, and it took us five years to complete the manuscript. We thank Pascal Porcheron, Ellen MacDonald-Kramer, and Stephanie Homer at Polity Press for their extraordinary patience and encouragement during this time.

Our project took time because we aimed to do something novel with it. Although there are good introductions to the philosophy of psychology or to the philosophy of the cognitive sciences, they tend to focus on what we call ‘foundational philosophy of psychology’, where the foundational issues in psychology (such as the nature of representation, or the language of thought hypothesis) are prioritized. In contrast, this book focuses on ‘implicational philosophy of psychology’, where the philosophical implications of psychological studies (such as the study of reasoning biases by Daniel Kahneman and Amos Tversky [discussed in Chapter 1], or the study of altruistic behaviour by Daniel Batson [discussed in Chapter 5]) take centre-stage.

The present book is an attempt to put together the issues we find interesting and stimulating at the intersection of philosophy and psychology, and share them with our students in an accessible and engaging way. We have had excellent role models as lecturers of Philosophy of Psychology over the years, including John Campbell, Martin Davies, David Papineau, Kim Sterelny, Helen Steward, … to mention only a few.

We hope we have learnt something from them.

An Overview

This book introduces and explains some central issues in the philosophy of psychology. It is important to note that it is not a neutral introduction to the topic. It is an opinionated introduction to the philosophy of psychology in which we defend a particular view of human cognition and agency. The view that we defend is that human cognition and agency are imperfect, in the sense that humans fail to meet some ideal standards of cognition and agency, such as the ideal of rationality, self-knowledge, and free and responsible agency.

Imagine that Lucy wants to pass her ballet exam tomorrow and she knows that she needs to rehearse the opening routine to make sure she remembers all the steps. Passing the exam is important to her, and she would be very disappointed if she failed. Yet she sits on the sofa to watch a movie instead of practising her routine. Given Lucy’s goal (passing the ballet exam tomorrow), her behaviour (watching a movie instead of practising) may be considered irrational since she does not take the appropriate steps to fulfil her goal. Chapter 1 discusses various notions of rationality and some ways in which human agents might fall short of the standards of rationality.

Now imagine that Giorgio describes himself as an exceptionally generous child, always willing to support his friends. However, that is not how his friends describe him. On a number of occasions, they have observed that Giorgio has put his own needs before theirs and has refrained from sharing snacks or helping them with their homework. They think of him as mean and selfish. If Giorgio’s friends are right, then we can say that Giorgio lacks self-knowledge. In other words, he does not have an accurate representation of himself. In particular, he attributes to himself qualities (such as generosity) that he does not have. Chapter 2 discusses several ways in which self-knowledge can fail.

Finally, imagine that Elsa, who is a white US citizen, thinks of herself as a very egalitarian person who respects everybody and considers them all equal, independently of ethnic origin or skin colour. However, when she goes to work, she avoids sitting next to Asian passengers on the bus and, when she talks to her colleagues at work, she tends to ignore her Nigerian secretary. Elsa does not have explicit beliefs that are prejudiced against people who look different from her but behaves in ways that are inconsistent with her explicit beliefs and values. She is vulnerable to implicit biases for which she may or may not be responsible. Chapter 3 deals with the duality of states and attitudes and Chapter 6 with the question whether people are responsible for their implicit biases.

You may have experienced similar failures of rationality and self-knowledge and have encountered biases in yourselves and others – you can come up with your own examples of how human agents are imperfect. Indeed, the view that human cognition and agency are imperfect is not surprising. Not many philosophers are convinced that humans are perfect agents who are ideally rational, have complete self-knowledge, and behave in unbiased ways. However, this book does not just claim that human cognition and agency are limited; on the basis of relevant psychological studies, it also makes some potentially surprising and controversial claims about the extent to which and the reasons why human cognition and agency are limited.

In this context, ‘imperfect’ does not mean ‘stupid’ or ‘dumb’. It means that performance falls short of the ideal standards that apply to it. This book does not advocate an overly pessimistic conception of human cognition and agency. For instance, we do not endorse what John Kihlstrom (2004) calls the ‘People are Stupid’ school of psychology, which says that humans are fundamentally irrational, that their behaviour is automatic and inflexible, and that they are ignorant of what they are doing. Rather, we advocate a realistic conception that is informed by psychological findings and is defended by philosophical arguments. We resist two extreme views of human cognition and agency. On the one hand, we resist an overly optimistic view according to which human cognition and agency are perfect (or near-perfect) with respect to ideal standards such as rationality, self-knowledge, and free and responsible agency. On the other hand, we resist an overly pessimistic view according to which human cognition and agency are hopeless with respect to these ideal standards. Our cognition and agency are certainly not perfect. However, by learning about the nature of, extent of, and reasons for their limitations, human agents can improve their performance. That is one of the reasons why philosophy of psychology is important. It does not simply provide information about what human agents can or cannot do. It offers human agents the resources to enhance their performance (in reasoning, decision-making, problem-solving, etc.) so that they can get closer to the ideal standards.

There is another sense of ‘imperfection’ that is addressed in this book: human cognition and agency can be ‘imperfect’ in the sense that they are vulnerable to disorders of the mind (broadly construed, including psychiatric disorders, behavioural anomalies caused by brain damage, developmental disorders, etc.). Disorders of the mind are usually identified by behaviours that appear unusual. While it is common to be weak-willed like Lucy, blind to one’s own character flaws like Giorgio, or unconsciously biased like Elsa, endorsing wildly implausible beliefs or seeing things that other people cannot see is less common and might lead to the person being diagnosed with a mental disorder. For instance, the mathematician John Nash experienced hallucinations and delusions, coming to believe that aliens were publishing messages in newspapers (which was described in the biographical movie A Beautiful Mind). This book will discuss ‘abnormalities’ of the mind, including delusion and confabulation (in Chapter 7) and autism and psychopathy (in Chapter 8). Thinking about unusual cognitions and behaviours helps us understand how the mind works and how vulnerability to disorders of the mind relates to the failure to meet ideal standards.

Here are the central questions discussed in the following chapters:

• Rationality (Chapter 1): Are we rational or irrational? How good are we at reasoning in accordance with the rules of logic and mathematics? How do we reason or make judgments? Do we rely on unreliable heuristics or shortcuts?
• Self-Knowledge (Chapter 2): How well do we know our own mind? How do we know our mental states? Is the process of knowing our own mental states different in kind from the process of knowing somebody else’s mental states?
• Duality (Chapter 3): Is human reasoning determined by the interaction between two distinct kinds of processes? If so, how are these processes different and how do they interact? Why are there only two kinds of processes, rather than three or four?
• Moral Judgment (Chapter 4): Are moral judgments based on reasoning or emotion? Or perhaps both? How do moral reasoning and moral emotion interact with one another to produce moral judgments?
• Moral Motivation and Behaviour (Chapter 5): Can we be genuinely altruistic? Or are we inevitably selfish? Does empathy cause genuinely altruistic behaviour? Does empathy-induced behaviour have morally desirable consequences?
• Free Will and Responsibility (Chapter 6): Are we free and responsible agents? Is the idea of free will compatible with the empirical findings in psychology and neuroscience? Are we responsible for our implicit biases?
• Delusion and Confabulation (Chapter 7): How are clinical delusions and confabulations different from everyday irrational beliefs? Are delusions more irrational than everyday irrational beliefs? Is the irrationality of delusions different in kind from the irrationality of everyday beliefs?
• Autism and Psychopathy (Chapter 8): What does autism tell us about mind-reading capacity and its role in social and communicative activities? What does psychopathy tell us about empathy and its role in moral behaviour?

Philosophy of Psychology

What is Philosophy of Psychology?

We use the term ‘psychology’ broadly. ‘Psychology’ in our sense of the word includes both psychological disciplines in the narrow sense – such as cognitive psychology, social psychology, developmental psychology, and evolutionary psychology – and research areas in neuroscience and psychiatry. (We also use phrases such as ‘psychology and neuroscience’ or ‘psychology and psychiatry’ in some contexts to emphasize the relevance of neuroscience or psychiatry.) ‘Psychology’ in our sense of the word is somewhat similar to ‘cognitive science’, but the latter typically includes not only psychology and neuroscience but also artificial intelligence research, linguistics, and anthropology, which will not be discussed in this book.

What is philosophy of psychology? Providing a clear definition of philosophy of psychology is challenging and perhaps not very helpful. One thing we can do at this stage, and perhaps a useful thing to do, is to introduce the field of philosophy of psychology with some examples of the questions it asks. Philosophy of psychology is the discipline in philosophy that asks the questions we mentioned above concerning rationality, self-knowledge, free will and responsibility, moral judgment, moral motivation and behaviour, and so on.

Below, we clarify what philosophy of psychology is by contrasting philosophy of psychology with two related areas of philosophy, philosophy of mind and philosophy of science, and then distinguishing two approaches to the philosophy of psychology: the first approach is mainly concerned with the theoretical and conceptual foundations of psychology and the second approach is mainly concerned with the implications of psychological studies for philosophically relevant issues.

Philosophy of Psychology and Philosophy of Mind

Philosophy of psychology can be thought of as a sub-area of philosophy of mind, where ‘philosophy of mind’ is understood very broadly as the philosophical exploration of the mind. The focus of mainstream philosophy of mind, which is sometimes called ‘metaphysics of mind’, has been on metaphysical issues of the mind, including the mind–body problem, mental causation, physicalism, and the nature of consciousness. Philosophy of psychology, in contrast, focuses on the kind of issues that are closely related to empirical studies of the mind or the kind of issues on which we cannot make progress without relevant empirical input. When philosophy of psychology addresses human rationality, for example, it examines the available empirical data to find out whether humans actually reason in accordance with the norms of rationality.

Philosophy of Psychology and Philosophy of Science

Philosophy of psychology can also be thought of as a sub-area of philosophy of science, where ‘philosophy of science’ is understood very broadly as the philosophical exploration of science. Philosophers of science study what a theory is, investigate applications of scientific method, and ask whether scientists make progress. Central issues in philosophy of science, or ‘general philosophy of science’, are not essentially connected to a particular scientific field (although many of the traditional issues in philosophy of science are explicitly or implicitly associated with the physical sciences). A recent trend in philosophy of science, however, is the growth and development of those sub-areas that deal with issues that are specific to a particular scientific field. For example, philosophy of biology discusses philosophical issues that emerge from biological studies (e.g., the issues related to adaptationism, the concept of ‘function’ in biology, the level of selection, etc.). Philosophy of psychology is another sub-area that discusses philosophical issues that emerge from psychological studies.

Foundational and Implicational

We find it useful to divide philosophy of psychology into two sub-areas: ‘foundational philosophy of psychology’ and ‘implicational philosophy of psychology’.

Foundational philosophy of psychology is concerned with core concepts in psychology (e.g., ‘representation’, ‘computation’, or ‘concept’) and general hypotheses in psychology that can make sense of a series of relevant studies as interconnected, including: the massive modularity hypothesis, which roughly says that our mind is composed of ‘modular’ systems that are dedicated to some specific tasks (Fodor 1983); the language of thought hypothesis, which roughly says that our cognition is based upon processing language-like representations (Fodor 1975); and the adaptive unconsciousness hypothesis, which roughly says that a part of our mind has evolved via natural selection and we do not and cannot have direct access to it (Wilson 2002). Questions in foundational philosophy of psychology include: ‘What are representations?’, ‘How does computation work?’, ‘Is the massive modularity hypothesis plausible?’, ‘Is the language of thought hypothesis needed to explain cognition?’, and so on.

In contrast, implicational philosophy of psychology focuses on the results of particular psychological studies (rather than their theoretical foundations) and investigates their implications for issues that are philosophically relevant. For example, Chapter 1 will focus on a series of influential studies on reasoning biases (Kahneman, Slovic, & Tversky 1982; Tversky & Kahneman 1974) exploring the implications of such studies for questions concerning human rationality. It will ask, for example, whether these studies show that humans are irrational after all. Chapter 7 will discuss studies on the limitations of memory (Loftus 2003; Schacter & Addis 2007), which show that human agents can easily misremember even important details of past events and thus might have important implications for eye-witness testimony in forensic settings. One might subsequently ask whether these studies show that eye-witness testimony is fundamentally unreliable.

Although we think that the distinction between foundations and implications is useful, we do not assume that it is clear-cut. The difference between them is probably a matter of degree: for example, some issues are more foundational and less implicational than others. There will be some borderline cases. For example, the dual-process theory (the theory according to which there are two distinct types of cognitive processes: Type-1 processes that are fast, unconscious, and automatic, and Type-2 processes that are slow, conscious, and controlled), which is discussed in Chapter 3, might be a borderline case.

Both dimensions of philosophy of psychology are equally important, but this book is more focused on implicational philosophy of psychology than foundational philosophy of psychology. There are two reasons for this. First, implicational issues have recently stimulated very lively discussions in philosophy of psychology research (which used to be dominated by foundational issues). We wanted this book to reflect this recent trend. Second, existing philosophy of psychology textbooks mainly focus on foundational issues, leaving the implications largely unexplored. This book is motivated by our frustration with the lack of accessible resources for discussing implicational issues. Hereafter, ‘philosophy of psychology’ means implicational philosophy of psychology.

This book will focus on implicational issues rather than foundational issues; however, strictly speaking, its focus is even narrower. There are so many interesting implicational issues and one book is not able to cover them all. As hinted previously, we will prioritize topics related to the ‘imperfection’ of human cognition and agency. Our central topics will include reasoning, judgment, belief, emotion, behaviour, and agency. For this reason, unfortunately, we will largely ignore the issues concerning consciousness, perception, attention, and so on. (See Further Resources for some material on the issues that we do not discuss in this book.)

Why Do We Need Philosophy of Psychology?

Why do we need philosophy of psychology, in addition to metaphysical philosophy of mind and psychology as distinct disciplines? Why can’t philosophers confine themselves to purely metaphysical issues? Why can’t psychologists confine themselves to purely empirical investigations?

There has been no sharp distinction between philosophy and psychology in Western intellectual history. Many philosophers who talk about mind and cognition in the history of philosophy go beyond the field of philosophy of mind in a narrow sense and touch on issues that now belong to the field of psychology, for example: Plato’s tripartite theory of soul (according to which the soul is divided into reason, spirit, and appetite) in The Republic; Descartes’ physiological analysis of various emotions in Passions of the Soul; Hume’s associationist psychology in A Treatise of Human Nature; Kant’s classification of mental disorders in Essay on the Maladies of the Head, and so on. The sharp distinction between philosophy and psychology is a modern development, mainly due to the specialization of each field. The advancement of research in each field obliged some researchers to focus on the metaphysical issues of the mind, and other researchers to dedicate themselves to empirical investigations instead. This is an inevitable and perhaps good development overall, but it invites a potential problem: many of the fundamental questions about human nature require philosophical argumentation as well as empirical inputs.

Suppose that you are interested in whether humans have some property X, such as the property of being rational or the property of being altruistic. To answer this, you need to investigate two sets of questions: (1) ‘What does it mean to have X?’, ‘What is necessary for a person to have X?’, and ‘What is sufficient for a person to have X?’; and (2) ‘Do humans satisfy a sufficient condition for having X?’, ‘Do humans fail to satisfy a necessary condition for having X?’, and ‘What do empirical studies say about these issues?’ Let us call the former the ‘philosophical questions’ about X, and the latter the ‘psychological questions’ about X.

Addressing the philosophical questions requires the articulation and justification of certain (controversial) claims, such as ‘being rational’ means reasoning in accordance with the rules of logic and mathematics, or ‘being altruistic’ means being motivated to increase somebody else’s well-being for its own sake. Addressing the psychological questions requires setting up laboratory experiments and observations in the wild, where human behaviour is assessed on the basis of the criteria for rationality or altruism, such as measuring how many people out of 100 commit a logical fallacy in a simple test or how many people out of 100 act in a selfless way given the chance.

The reason you can’t just let philosophers do what they do is that although a purely philosophical investigation is useful for addressing the philosophical questions, it does not say much about the psychological questions. For instance, a philosophical argument does not say anything about whether as a matter of fact humans can be motivated to increase somebody else’s well-being for its own sake. In contrast, the reason you can’t just let psychologists do what they do is that although a purely psychological investigation is useful for addressing the psychological questions, it does not say much about the philosophical questions. For example, a psychological experiment does not say anything about whether ‘rationality’ should be regarded as the capacity for reasoning in accordance with the rules of logic and mathematics.

Let us think about a further example (which we discuss in detail in Chapter 6). Suppose that you want to know whether humans have free will and responsibility, according to relevant psychological findings. To know this, you need to investigate both the philosophical and the psychological questions: (1) ‘What does it mean to have free will?’, ‘What is necessary for a person to have free will?’, ‘What is sufficient for a person to have free will?’; (2) ‘Do humans satisfy a sufficient condition for having free will?’, ‘Or, do humans fail to satisfy a necessary condition for having free will?’, ‘What do empirical studies say about these issues?’

A purely philosophical investigation can address the philosophical questions, but not the psychological questions. For example, one might argue, for some philosophical reasons, that it is necessary for person A to have free will that A’s conscious mental states and processes play some significant causal role in producing or controlling A’s behaviour. This is certainly a possible answer to a philosophical question, but we need to know more to ascertain whether humans have free will. In particular, the psychological questions have not been answered yet. It is still unclear whether, according to relevant empirical studies, conscious states and processes in human cognition play the right kind of causal roles. As we shall see in Chapter 6, this is the topic of a recent interdisciplinary debate on free will.

A purely psychological investigation can address the psychological questions, but not the philosophical questions. For example, psychological studies might suggest that Elsa’s discriminatory behaviour is generated by her implicit biases, which can only be detected by psychological tests. Elsa is not introspectively aware of her biases. This is certainly an interesting finding, but we need to know more to judge whether she is responsible for her discriminatory behaviour. In particular, the philosophical questions have not yet been addressed. It is still unclear whether Elsa’s lack of introspective awareness of her implicit biases is incompatible with her being responsible for her discriminatory behaviour. As we shall see in Chapter 6, this is the topic of a recent discussion about whether people are responsible for their implicit biases.

Evaluating Psychological Studies

This book focuses on implicational philosophy of psychology, investigating the implications of some particular psychological studies. Before getting into our main discussions, we would like to mention three factors that are relevant in the context of evaluating psychological studies and their implications: replication, research participants, and ecological validity.

Replication

Psychological studies, just like other scientific studies, need to be replicated in order to be credible. It is possible that some interesting results fail to be replicated. In fact, psychology, social psychology in particular, is now facing the so-called ‘replication crisis’ (e.g., Earp & Trafimow 2015): the replicability of social psychological studies, including some famous ones, seems to be remarkably low.

A recent controversy, for example, is about a series of studies of priming effects by John Bargh and colleagues. Priming effects are the unconscious effects that the exposure to a stimulus has on the responses to subsequent stimuli. One famous study (Bargh, Chen, & Burrows 1996, which has over 5,500 citations on Google Scholar as of October 2020) reports the surprising result that the participants who had completed the simple task of making a sentence out of some given words, including words related to stereotypes of the elderly (e.g., ‘grey’, ‘wise’, or ‘wrinkle’), walked away more slowly than other participants who had completed the same task without elderly-related words. However, several researchers (e.g., Doyen et al. 2012) reported their failure to replicate the elderly stereotype study and suggested that the popularity of priming studies by Bargh and colleagues is disproportional to their scientific credibility. Daniel Kahneman, who had once mentioned the elderly stereotype study in his best-selling book Thinking, Fast and Slow (2011), admitted that he had ‘placed too much faith in underpowered studies’ in response to a blog post on the controversy (Schimmack, Heene, & Kesavan 2017).

The problem is compounded by the fact that replication failure tends not to be reported in journals (unless, just like in the case of priming studies by Bargh and colleagues, a serious controversy arises). Paul Bloom says that when the project of replicating a study turns out to be unsuccessful, ‘[u]sually, the project is just abandoned, though sometimes the word gets out in an informal way – in seminars, lab meetings, conferences – that some findings are vaporware (“Oh, nobody can replicate that one”)’ and ‘[m]any psychologists now have an attitude that if a finding seems really implausible, just wait a while and it will go away’ (Bloom 2017, 224).

We do not necessarily endorse radical pessimism about social psychology; after all, many important studies have been replicated. But, as a general rule, replication should always be kept in mind when evaluating psychological studies, and perhaps some extra care is needed when evaluating studies in social psychology with surprising results.

Replication really matters, but what about the observations of rare conditions about which we cannot expect statistical analysis of data or replication? Despite the lack of statistical analysis or replication, the study of some unusual behaviours can be informative and can help us to understand how the mind works. In the book Phantoms in the Brain, V. S. Ramachandran defends the usefulness of observing rare cases as opposed to the statistical study of normal individuals:

[I]n neurology, most of the major discoveries that have withstood the test of time were, in fact, based initially on single studies and demonstrations. More was learned about memory from a few days of studying a patient called H.M. than was gleaned from previous decades of research averaging data on many subjects. The same thing can be said about hemispheric specialization (organization of the brain into a left brain and a right brain, which are specialized for different functions) and the experiments carried out on two patients with so-called split brains (in whom the left and right hemispheres were disconnected by cutting the fibers between them). (Ramachandran & Blakeslee 1998, xiii)

We agree with Ramachandran, especially with his idea that we can gain useful insights by observing some rare cases. This is why we will include the split-brain cases (which Ramachandran mentions in this passage) in our discussion of self-knowledge in Chapter 2, and Capgras syndrome (which is a rare condition that Ramachandran discusses in his book) in our exploration of delusion in Chapter 7. Ramachandran’s idea is also perfectly compatible with the importance of statistical analysis and replication when it comes to the phase in which we test the insights that we initially gain in observing rare cases. In fact, he only says that major discoveries in neurology were ‘initially’ based on single studies and demonstrations; his recommendation is ‘to begin with experiments on single cases and then to confirm the findings through studies of additional patients’ (Ramachandran & Blakeslee 1998, xiii).

Research Participants

Psychological experiments need human participants. Typically, the participants of psychological studies are university students because these are the most easily accessible kind of people for psychology researchers. But this can be a problem. In psychology, we are not just interested in finding out how the mind of a university student works; rather we are interested in finding out how the mind works in general. Can we learn about how the mind works in general from psychological studies involving only university students?

Uncritical optimism on this issue has been questioned in recent years. It has been suggested (Henrich, Heine, & Norenzaya 2010a, 2010b) that most psychological studies published in major journals rely on the participants in Western, educated, industrialized, rich, and democratic (WEIRD) societies, who are in fact highly unusual in many psychological aspects:

[P]eople from Western, educated, industrialized, rich and democratic (WEIRD) societies – and particularly American undergraduates – are some of the most psychologically unusual people on Earth. So the fact that the vast majority of studies use WEIRD participants presents a challenge to the understanding of human psychology and behaviour. […] Strange, then, that research articles routinely assume that their results are broadly representative, rarely adding even a cautionary footnote on how far their findings can be generalized. (Henrich, Heine, & Norenzaya 2010b, 29)

We will come back to this problem in Chapter 4 when discussing cross-cultural studies of moral judgments (e.g., Haidt, Koller, & Dias 1993; Shweder, Mahapatra, & Miller 1987).

Ecological Validity

Many psychological studies are undertaken in some artificial situations that are not representative of everyday settings, and the tasks proposed in the laboratory may not have the same relevant features as everyday tasks. This raises a methodological problem. In psychology, we are not just interested in understanding how people think and behave in artificial tasks and settings; rather we are interested in understanding how people think and behave in everyday tasks and settings. Can we learn about how people think and behave in everyday tasks and settings from the psychological studies involving only artificial tasks and settings?

Bloom warns us against the tendency to conclude that human agents are irrational on the basis of the experimental findings of irrational biases and errors in artificial tasks and settings. Perhaps biases and errors are exaggerated in artificial settings; they might not be very significant in real-life cases.

Statistically significant doesn’t mean actually significant. Just because something has an effect in a controlled situation doesn’t mean that it’s important in real life. Your impression of a résumé might be subtly affected by its being presented to you on a heavy clipboard, and this tells us something about how we draw inferences from physical experience when making social evaluations. Very interesting stuff. But this doesn’t imply that your real-world judgments of job candidates have much to do with what you’re holding when you make those judgments. What will actually matter much more are such boringly relevant considerations as the candidate’s experience and qualification. (Bloom 2017, 224–225)

This is something we should keep in mind, especially when we explore the limitations of human cognition; it is conceivable that the limitation that is found in a controlled experiment does not have much impact in real life.

Further Resources

Articles and Books

Tony Stone and Martin Davies’s article on cognitive neuropsychology and philosophy of mind (1993) includes a very interesting discussion of the aim and scope of philosophy and psychology. Alvin Goldman’s Philosophical Applications of Cognitive Science (2018) includes many examples of how philosophical discussions can be fruitfully informed by psychology and cognitive science.

As we noted, philosophy of psychology contains many topics and issues that we cannot cover in this book. See Philosophy of Psychology: A Contemporary Introduction (2004) and Cognitive Science: An Introduction to the Science of the Mind (2020) by José Luis Bermúdez; An Introduction to the Philosophy of Psychology (2015) by Daniel Weiskopf and Fred Adams; Mindware: An Introduction to the Philosophy of Cognitive Science (2000) by Andy Clark; and The Philosophy of Cognitive Science (2015) by Mark Cain. We also recommend The Phenomenological Mind (2012) by Shaun Gallagher and Dan Zahavi, and The Disordered Mind: An Introduction to Philosophy of Mind and Mental Illness (2013) by George Graham.

Online Resources

Paul Thagard’s Stanford Encyclopedia of Philosophy entry on cognitive science (2019) is helpful as it summarizes some of the key areas of philosophy of psychology.

Tamar Gendler’s lecture Philosophy and the Science of Human Nature (2011), available at Open Yale Courses, discusses a wide range of issues in philosophy of psychology by combining ancient philosophical wisdom with cutting-edge psychological studies.

Questions

1. How can philosophy and psychology profitably interact in understanding how the human mind works?
2. Can you think of some particular examples in which philosophers and psychologists profitably interact with one another?
3. What can philosophers learn from psychological studies?
4. What can psychologists learn from philosophical arguments and discussions?
5. Is there a clear boundary between philosophical discussion of the mind and psychological discussion of it? Or is the boundary inevitably vague?

1.1 Introduction

It is a long tradition in Western philosophy to characterize humans as rational animals and to argue that rationality is one of the features that distinguishes them from other animals. It is not just Aristotle who describes the human as the rational animal in his Metaphysics (1984). In Discourse on the Method (1985), Descartes also characterizes humans in terms of their distinctive reason or understanding; non-human animals do not have reason at all. This trend has continued: Donald Davidson says that rationality distinguishes ‘between the infant and the snail on one hand, and the normal adult person on the other’ (Davidson 1982, 318).

This view, however, can be (and has been) challenged. One challenge is to deny the sharp contrast between human cognition, which is rational, and non-human cognition, which is not. Humans are rational, but so are non-human animals. For example, Hume writes in his A Treatise of Human Nature that ‘no truth appears to me more evident than that beasts are endowed with thought and reason as well as man’ (Hume 1739/2007, 118).

Another challenge, which is the focus of this chapter, is to deny optimism about human rationality. Non-human animals are not rational, but humans are not rational either. In the memorable beginning of his essay ‘An Outline of Intellectual Rubbish’, Russell wrote:

Man is a rational animal – so at least we have been told. Throughout a long life I have searched diligently for evidence in favour of this statement. So far, I have not had the good fortune to come across it, though I have searched in many countries spread over three continents. (Russell 1961, 45)

Russellian pessimism about human rationality is echoed in a particularly influential psychological research programme in the 20th century, the heuristics and biases research programme, led by Amos Tversky and Daniel Kahneman. The studies in this programme revealed ‘systematic and severe errors’ (Tversky & Kahneman 1974, 1124) in human reasoning, which seem to have a ‘bleak implication for human rationality’ (Nisbett & Borgida 1975, 935). These studies caused heated debates on human rationality, sometimes dubbed ‘rationality wars’ (Samuels, Stich, & Bishop 2002), both at the theoretical and conceptual level and at the empirical and experimental level.

The main aim of this chapter is to examine the relevant psychological studies to see if they really do have bleak or pessimistic implications for human rationality.

We start by presenting a definition of rationality (Section 1.2) before turning to the relevant psychological studies, in particular the ones from the heuristics and biases programme, which reveal a systematic failure to reason according to the rules of logic, probability, and decision-making (Section 1.3). The results of these studies support a pessimistic view of human rationality (Section 1.4). However, objections to the pessimistic interpretation were raised by Gerd Gigerenzer and his ecological rationality research programme (Section 1.5). The objections from this programme are significant because they bring to the fore important ideas that enhance our understanding of human rationality/irrationality. However, we shall argue that these objections do not refute the pessimistic interpretation. Although Gigerenzer’s objections do not refute pessimism, they lead us to examine an important issue: whether the difference between optimism and pessimism about rationality hangs on different accounts of the aim of cognition (Section 1.6). In the end, we defend a moderate form of pessimism, according to which humans are not as rational as we might have thought, before appreciating the results of the psychological studies on reasoning.

1.2 Clarifying Rationality

Are humans rational? To answer this question, we need to investigate two sets of issues: (1) ‘What does it mean to be rational?’, ‘What is necessary for a person to be rational?’, and ‘What is sufficient for a person to be rational?’ (the philosophical questions about rationality); and (2) ‘Do humans satisfy a sufficient condition for being rational?’, ‘Do humans fail to satisfy a necessary condition for being rational?’, and ‘What do empirical studies say about these issues?’ (the psychological questions about rationality).

We discuss the philosophical questions in this section and the psychological questions in the next section.

‘Rationality’ means different things in different contexts (Bortolotti 2014). We will focus on a narrow sense of rationality here: that is, rationality in the context of reasoning. In particular, we (tentatively) accept what Edward Stein (1996) calls ‘the standard picture of rationality’, according to which rationality consists in reasoning in accordance with the rules of logic, probability, and decision-making (see Box 1A).

BOX 1A: The Standard Picture of Rationality

Definition:

‘According to this picture, to be rational is to reason in accordance with principles of reasoning that are based on rules of logic, probability theory, and so forth. If the standard picture of reasoning is right, principles of reasoning that are based on such rules are normative principles of reasoning, namely they are the principles we ought to reason in accordance with.’ (Stein 1996, 4)

Example of deductive reasoning:

Jessie believes that either Paula got an A in geometry or Vanessa did, and also that Vanessa did not get an A but got a D instead. From these premises, Jessie concludes that it is Paula who got an A. Jessie is rational: her logical reasoning is in accordance with a rule of logic that one can infer P from (P or Q and not Q) (disjunctive syllogism).

Example of probabilistic reasoning:

Felix assigns the probability 0.8 to the idea that it will rain tomorrow based on the televised weather forecast that there is an 80% chance of rain tomorrow. Felix also assigns the probability 0.9 to the idea that it will rain tomorrow and so he will go to the gym for a workout. Felix is irrational: his probabilistic judgment violates what is known as the ‘conjunction rule’; the probability of an event A occurring, P(A), cannot be less than the probability of A and another event B occurring at the same time, P(A&B).

But how often should human agents reason in accordance with these rules to be considered rational agents? If the answer is ‘Always’, then a rational person wouldn’t be allowed to make any mistakes; this requirement is far too demanding. It is obvious that human agents are not rational according to such a stringent requirement. A realistic view would take into account the fact that rational agents can make reasoning errors, such as the occasional performance error that is attributed to some interfering factor (e.g., lack of concentration) (see Box 1B), but they do not make systematic reasoning errors. Thus, rationality is compatible with occasional reasoning errors, but it is not compatible with systematic reasoning errors.

BOX 1B: Competence vs Performance

When you attempt a task and fail, one of two things may be happening: either you do not have the capacity to accurately perform the task due to some knowledge-gap, or you have the capacity to accurately perform the task but fail to perform it accurately on a particular occasion due to external factors (such as a distraction).

Example: Gina and Tommaso were asked to calculate the square of four by their teacher. Gina did not answer at all because she did not know that the square of a number is the number multiplied by itself. Gina lacked the competence to answer the question correctly. Tommaso answered ‘twelve’. He knew how to make the square of a number and he was familiar with the four times table, but he was distracted by a sudden noise outside the classroom and gave the wrong answer. Tommaso made a performance error.

When Cohen says that ‘ordinary human reasoning […] cannot be held to be faultily programmed’, he means that human reasoning competence is intact (human agents can apply the rules of good reasoning in ideal conditions), although reasoning performance may be imperfect (human agents may make errors in applying the rules of good reasoning due to external factors).

The standard picture of rationality is accepted by many, perhaps most, philosophers, and reasonably so. After all, there is a key sense of ‘rational’ that describes people who follow the rules of logic, probability, and decision-making when they reason and solve problems. There are other senses of ‘rational’, of course, such as those that describe people who are not overwhelmed by their emotions when they make decisions, or those that describe people who support their arguments with evidence instead of merely stamping their foot in a debate (Bortolotti 2014). But rationality as logicality, as we might call it, is widely accepted in philosophy, economics, and psychology. For instance, Phil Gerrans says that ‘a rational subject is one whose reasoning conforms to procedures, such as logical rules, or Bayesian decision theory, which produce inferentially consistent sets of propositions’ (Gerrans 2001, 161), and Richard Nisbett and Paul Thagard define rational behaviour as ‘what people should do given an optimal set of inferential rules’ (Thagard & Nisbett 1983, 251).

Having said that, not everybody agrees that the standard picture is the best understanding of human rationality. This controversy lies at the heart of the rationality wars between pessimists about human rationality (often appealing to the heuristics and biases programme) and optimists about human rationality (often associated with the ecological rationality programme). We will come back to this in Section 1.5.

There are some technical issues about the standard picture of rationality that we would like to mention briefly here.

First, the standard picture seems to presuppose that there is just one system of logic, one theory of probability, and one set of principles for decision-making. However, there are different formal systems of logic and different interpretations of probability; even the principles of decision-making can be disputed. Some rules of inference that are valid in standard logic (often called classical logic) are not valid in some non-classical logical systems. This raises a question: which system of logic should be adopted in evaluating the reasoning performance of agents? This is especially tricky if reasoning performance is consistent with one system of logic but not with another. Should we adopt the former and say that an agent’s performance is rational? Or should we adopt the latter and say that it is irrational? A similar issue arises when considering interpretations of probability. There are different interpretations of what probabilistic statements (e.g., there is a 80% chance that it will rain tomorrow) actually mean. There are also some probabilistic statements that make sense in some interpretations but not in others. This issue is relevant to the debate between pessimists and optimists. Gigerenzer, the most notable optimist, argues that some probabilistic questions in the heuristics and biases experiments are meaningless in light of his favourite interpretation of probability (which is known as the frequency interpretation). This issue will be discussed in Section 1.5.

Second, the standard picture assumes that our reasoning should be evaluated against the standards of logic, probability, and decision-making. This implies that, if our intuitive answer to a reasoning task is incompatible with a rule of logic, we should conclude that our intuition is at fault. But why can’t we say that it is logic, not intuition, that is at fault? In fact, the development of non-classical logic is sometimes at least partially motivated by some counter-intuitive features of classic logic.

This issue raises a further question: what should we do when facing an apparent discrepancy between logic and intuitive judgment? Should we trust logic and dismiss intuition as irrational? Or should we trust intuition and dismiss logic instead? Jonathan Cohen, another notable optimist, raises a similar issue. If the normative rules, against which our intuitive judgments are evaluated, are themselves evaluated on the basis of our intuitive judgments, then our intuition ‘sets its own standards’ (Cohen 1981, 317). But then how can our intuitive judgment be irrational? How can our intuitive judgment deviate from the standards that are set by itself? Cohen adopts a radical conclusion that human irrationality cannot be proven in principle no matter what psychology shows; ‘ordinary human reasoning – by which I mean the reasoning of adults who have not been systematically educated in any branch of logic or probability theory – cannot be held to be faultily programmed’ (Cohen 1981, 317). In making this claim, Cohen relies on the distinction between reasoning competence and reasoning performance (see Box 1B).

Third, the standard picture presupposes that what matters to rational and successful reasoning is the conformity to the rules of logic, probability, and decision-making. Not everybody agrees. For example, Keith Stanovich (1999) makes a distinction between rationality and intelligence, where rationality is a broader notion than in the standard picture, and intelligence is what the standard picture captures. In common discourse, intelligence and rationality are often conflated, or it is assumed that intelligence comprises rationality. But IQ tests do not measure the capacity for making good judgments and good choices that we commonly regard as a mark of rationality. If we take intelligence to stand for whatever IQ tests measure, then it does not tell us which behaviours are more likely to be conducive to the agent’s well-being in real life. Whereas IQ tests measure the capacity to process and manipulate information quickly and efficiently, they are not sensitive to whether the agent forms beliefs that are well supported by the evidence or whether she can critically evaluate the information she receives.

To illustrate this distinction, Stanovich describes famous cases of smart people who acted foolishly, by which he means that people who have high intelligence in some domain made bad judgments and bad choices, thereby behaving irrationally. This does not mean that intelligence is not worth studying, just that there are other things that we value. Intelligence and rationality can be seen as having different domains of applications: Stanovich, for instance, suggests that intelligence maps the efficiency of cognitive functioning at an algorithmic level, whereas his more comprehensive notion of rationality tracks thinking dispositions at a higher level, governs decision-making, and takes into account the agent’s goals and values. Some notions of rationality like Stanovich’s are distinct from the standard picture, where rationality is associated with behaviour that conforms to the rules of logic, probability, and decision-making.

1.3 Systematic Biases and Errors

Here is a tentative answer to the philosophical question (with the qualifications we made at the end of the previous section): ‘rationality’ consists in reasoning in accordance with the rules of logic, probability, and decision-making.

Now, let us turn to the psychological questions. The crucial question is: ‘Do human agents actually reason in accordance with the rules of logic, probability, and decision-making?’ We will now review relevant studies, mainly from the heuristics and biases programme, which offer ample evidence that human reasoning systematically deviates from the rules of logic, probability, and decision-making. They include some of the most famous results in 20th-century psychological research.

Wason Selection Task

According to deductive logic, for a conditional statement of the form ‘If P then Q’ to be falsified, the antecedent (P) must be true and the consequent (Q) must be false. So, the statement ‘If you want to go to Brighton, then you need to catch the next train’ is false if you do want to go to Brighton but you don’t need to catch the next train.

The purpose of the selection task is to establish whether people can recognize when conditional statements are false. In the classic version of the task (Wason 1966), there is a deck of cards and each of them has a number on one side and a letter on the other. Participants can see four cards on the table, the first has a vowel (A) on the visible side, the second an odd number (7), the third a consonant (K), and the last an even number (4) (Figure 1). Participants have to say which cards they need to turn to test the following rule: ‘If a card has a vowel on one side, then it has an even number on the other side.’ Most participants in the classic version of the task said that the cards to be turned are the card with A on the visible side and the card with 4 on the visible side, or just the card with A on the visible side. However, the correct way to test the rule is to turn the card with A on the visible side and the card with 7 on the visible side, because a conditional statement is falsified when the antecedent (‘If a card has a vowel on one side’) is true and the consequent (‘then it has an even number on the other side’) is false. Only 5% of the participants solved the selection task in this version.

Figure 1. Wason selection task with abstract options

Conjunction Fallacy

The next study involves the so-called ‘conjunction fallacy’ (also known as the conjunction effect), which is a mistake people make when they assume that a statement describing the conjunction of two states of affairs (e.g., ‘Tomorrow it will be raining and it will be cold’) is more probable than a statement describing one of those states of affairs alone (e.g., ‘Tomorrow it will be raining’).

In one experiment by Tversky and Kahneman (1983), participants first read personality sketches of hypothetical people, and then answered questions about them. Participants were divided into three groups according to their background in probability and statistics: the naïve group (those with no background), the intermediate group (those with basic knowledge of probability and statistics), and the sophisticated group (those with advanced knowledge of probability and statistics). Here is the personality sketch of a hypothetical person, Linda.

Linda is 31 years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and participated in anti-nuclear demonstrations.

Please rank the following statements by their probability, using 1 for the most probable and 8 for the least probable:

1. Linda is a teacher in elementary school.
2. Linda works in a bookstore and takes Yoga classes.
3. Linda is active in the feminist movement.
4. Linda is a psychiatric social worker.
5. Linda is a member of the League of Women Voters.
6. Linda is a bank teller.
7. Linda is an insurance salesperson.
8. Linda is a bank teller and is active in the feminist movement.

The result revealed that statements 3 and 7 were regarded, respectively, as the most probable and the least probable, which is explained by the similarity of Linda’s personality sketch with the stereotypical image of someone politically active, and the dissimilarity between her personality sketch and the stereotypical image of an insurance salesperson. The crucial statements to consider in this experiment are 6 and 8. The mean rank of the former was lower than the mean rank of the latter, which means that Linda being a bank teller and active in the feminist movement was regarded as more probable than her being a bank teller. This, however, is a mistake. Linda being a bank teller and active in the feminist movement is the conjunction of her being a bank teller and of her being active in the feminist movement. And the conjunction cannot be more probable than one of the conjuncts.

Thus, this mistake is an obvious violation of the conjunction rule. This fallacy was found in all three groups of participants and background knowledge in statistics and probability did not have a significant effect on the participant’s performance.

Base-Rate Neglect

In another famous experiment by Kahneman and Tversky (1973), half of the participants read a story, ‘cover story’, which said that psychologists prepared 100 personality descriptions on the basis of interviewing and testing 30 engineers and 70 lawyers. The other half of the participants read almost the same cover story, except that the number of engineers and the number of lawyers were switched; 70 engineers and 30 lawyers. Then participants were presented with personality descriptions – supposedly randomly selected from the 100 personality descriptions – and were asked to judge the probability of that person being an engineer. Here is one such description:

Jack is a 45-year-old man. He is married and has four children. He is generally conservative, careful, and ambitious. He shows no interest in political and social issues and spends most of his free time on his many hobbies which include home carpentry, sailing, and mathematical puzzles.

One finding was that participants made a judgment on the basis of the stereotypes associated with the two occupations – engineer and lawyer – which is consistent with what we saw in the Linda experiment. Linda was regarded as most likely to be active in the feminist movement, which nicely fits the stereotype provided in her personal description. Similarly, in this study, Jack was most likely judged to be an engineer rather than a lawyer because his personality description nicely fits the stereotype of engineers.

The crucial finding was that the base-rate information was largely neglected; the judgment was independent of the base rates provided in the cover stories.

This error constitutes a violation of Bayes’ rule, which says that the probability of an hypothesis H given an observation O (‘posterior probability’, P(H/O)) is determined by both how likely O is if H is true (‘likelihood’, P(O/H)), and how probable H is without the observation (‘prior probability’, P(H)). For example, the probability of Jack being an engineer given the personality description is determined by both how likely the personal description (being generally conservative, careful, ambitious, etc.) is if Jack is really an engineer, and how probable it is that Jack is an engineer without the personal description. What happened in the experimental results was that the participants largely ignored the prior probability (i.e., how probable it is that Jack is an engineer without the personal description) which is determined by the base-rate. Thus, Kahneman and Tversky summarize their finding as follows:

One of the basic principles of statistical prediction is that prior probability, which summarizes what we knew about the problem before receiving independent specific evidence, remains relevant even after such evidence is obtained. Bayes’ rule translates this qualitative principle into a multiplicative relation between prior odds and the likelihood ratio. Our subjects, however, fail to integrate prior probability with specific evidence. […] The failure to appreciate the relevance of prior probability in the presence of specific evidence is perhaps one of the most significant departures of intuition from the normative theory of prediction. (Kahneman & Tversky 1973, 243)

Preference Reversal

Another area of weakness in human reasoning can be found in the psychology and economics literature on preference reversals. The principle of procedure invariance tells us that, given two options, if one prefers A to B, then this preference should not change when the method for eliciting the preference changes. Yet participants often state a preference for A over B when they are asked to make a direct choice, but are prepared to pay more to obtain B than they are to obtain A.

Take two lotteries: a relatively safe lottery, where one has a 10% chance of winning nothing and a 90% chance of winning £10; and a relatively risky lottery, where one has a 10% chance of winning £90 and a 90% chance of winning nothing. If asked to choose, people usually prefer to buy a ticket for the safer lottery. In contrast, if asked at what price they would sell their ticket, they set a higher selling price for the ticket of the risky lottery. This phenomenon is observed in different contexts of choice and matching too (Stalmeier, Wakker, & Bezembinder 1997).

The classic example of the violation of procedure invariance in the literature is the Traffic Problem (Tversky & Thaler 1990, 201–202):

(1) The Minister of Transportation is considering which of the following two programs would make the electorate happier:

Program A is expected to reduce the yearly number of casualties in traffic accidents to 570 and its annual cost is estimated at $12 million.

Program B is expected to reduce the yearly number of casualties in traffic accidents to 500 and its annual cost is estimated at $55 million.

Which program would you like better?

(2) The Minister of Transportation is considering which of the following two programs would make the electorate happier:

Program A is expected to reduce the yearly number of casualties in traffic accidents to 570 and its annual cost is estimated at $12 million.

Program B is expected to reduce the yearly number of casualties in traffic accidents to 500.

At what cost would program B be as attractive as program A?

Options 1 and 2 represent two different ways of eliciting people’s preferences for one of the two life-saving programmes. In option 1, participants are given all necessary information about the two programmes: how many lives they would save and at what cost. When preferences are elicited in this way (direct choice), two-thirds of participants express a preference for programme B (which allows more lives to be saved at a higher cost). In option 2, participants are told how many lives would be saved and the cost of programme A, but they are not told the cost of programme B. Rather, they are asked at what cost programme B would become as attractive as programme A. When the preference is elicited this way (price matching), 90% of participants provide values smaller than $55 million for programme B, thereby indicating a preference for programme A.

If we take the evidence concerning people’s responses to the Traffic Problem as ecologically valid and reliable, it tells us something interesting: people have inconsistent attitudes about what the Minister of Transportation should do concerning the Traffic Problem. They believe that the Minister should implement programme B to save the lives of 70 more people a year, even if the programme costs $43 million more than programme A. They also believe that the Minister should implement programme A, which would save fewer lives, unless programme B cost considerably less than $55 million. Depending on the method by which the preference is elicited, participants seem to attribute different monetary value to human lives.

1.4 Pessimism about Rationality

Making Sense of the Results

Overall, these studies reveal ‘systematic and severe errors’ (Tversky & Kahneman 1974, 1124) in human reasoning that have a ‘bleak implication for human rationality’ (Nisbett & Borgida 1975, 935). They provide powerful empirical support for the pessimistic conclusion about human rationality that human agents systematically fail to reason in accordance with the rules of deductive reasoning, such as in the Wason selection task, or in accordance with the rules of probability theory, such as the conjunction rule (the Linda experiment) or Bayes’ rule (the Jack experiment). Moreover, participants violate basic principles of decision-making (procedure invariance and description invariance), and change their preferences depending on the methods by which their preferences are elicited and the way in which options are presented.

According to Kahneman and Tversky, systematic failures in human reasoning are due to the fact that human agents do not rely on the rules of logic, probability, and decision-making that would guarantee accuracy, but rather rely on heuristics. Heuristics are cognitive shortcuts, or cognitive rules of thumb, that ‘reduce the complex tasks […] to simpler judgmental operations’ (Tversky & Kahneman 1974, 1124). Heuristics are reliable in many cases, especially in those cases where heuristics and the rules of logic, probability, and decision-making deliver the same answer. However, in other cases, heuristics can lead to systematic errors and deliver different answers from those one would arrive at by applying the rules of logic, probability, and decision-making.

In making a heuristic judgment, you replace a difficult question with an easier one and answer that question instead. For example, the question ‘How far is the mountain over there from here?’ is a relatively difficult question, which, if you want to answer it in a canonical way, requires you to find a map, identify your place on the map, identify the mountain on the map, measure the distance between them on the map, and then calculate the actual distance while taking into account the scale of the map. Instead of answering the difficult question, you can substitute it with another question, ‘How clear does the mountain look to me?’, which is a lot easier to answer. When it looks very clear to you, for example, you can conclude that the mountain is very close to you. This substitution strategy works in many cases, but it inevitably leads to systematic errors in other cases: for example, distances are often overestimated when the contours of objects are blurry and are underestimated when the contours of objects are sharp.

Similarly, it is not easy to answer probabilistic judgments such as ‘What is the probability of Linda being a bank teller?’ or ‘What is the probability of Jack being an engineer?’ However, instead of answering these questions, one can substitute them with questions of similarity, which are a lot easier: ‘How is Linda’s description similar to that of a stereotypical bank teller?’ or ‘How is Jack’s description similar to that of a stereotypical engineer?’ The substitution in this case is known as an application of the ‘representativeness heuristic’, in which ‘probabilities are evaluated by the degree to which A is representative of B, that is, by the degree to which A resembles B’ (Tversky & Kahneman 1974, 1124).

The representativeness heuristic works in many cases, but it inevitably leads to systematic errors in other cases. For instance, it leads to the violation of the conjunction rule when participants are asked to compare the probability of Linda being a bank teller with the probability of Linda being a feminist bank teller. When participants rely on the representativeness heuristic, they compare the similarity between Linda and a stereotypical bank teller and the similarity between Linda and a stereotypical feminist bank teller. Since Linda is not similar to a stereotypical bank teller at all, participants come to the conclusion that Linda is more likely to be a feminist bank teller than a bank teller, which is mathematically fallacious.

Kahneman and Tversky claim that reasoning errors are systematic and severe, but this does not mean that they are unavoidable. A finding is that the conjunction effect is reduced by asking about the frequency of an event (how often the event occurs) rather than its probability (how probable it is that the event will occur). A study by Klaus Fiedler (1988) compared two versions of the Linda experiment: the original version (probability), which asks the ‘probability’ of Linda being a bank teller and of her being a feminist bank teller, and a new version (frequency), which instead asks how many people out of 100 who fit Linda’s description are bank tellers and how many are feminist bank tellers. Participants are much more likely to give the correct answer to the new (frequency) version of the task (78% correct answer) than to the original (probability) version (9% correct answer).

Similarly, other versions of the Wason selection task were devised based on the hypothesis that performance improves when the rule that is tested is more concrete and refers to situations that participants experience in everyday life. Indeed, there was significant improvement in the results. In one version (Wason & Shapiro 1971), the following statement was tested: ‘Every time I go to Manchester, I travel by car.’ Participants were presented with four cards that showed on their open sides one of the following: ‘Manchester’, ‘Leeds’, ‘car’, and ‘train’ (Figure 2). On this occasion, two-thirds of participants were able to choose the right pair of cards to turn to test the rule (the correct pair being the cards with ‘Manchester’ and ‘train’ on the visible sides). These performance improvements will be an important issue in our later discussion.

Figure 2. Wason selection task with concrete options

Argument for Pessimism

The psychological experiments on reasoning seem to support the pessimistic conclusion that humans are vulnerable to systematic and widespread irrationality. According to the experimental results, people exhibit poor reasoning performance in a number of tasks, including logical, probabilistic, and decision-making tasks; the conclusion drawn is that people do not reason rationally in those circumstances.

The evidence indicates that people make inferential errors. The errors seem to be due to lack of knowledge of certain inductive rules or an inability to apply them. If so, then people are not fully rational in that their inferences fall short of the best available normative standards. (Thagard & Nisbett 1983, 257)

We argue that the deviations of actual behavior from the normative model are too widespread to be ignored, too systematic to be dismissed as random error and too fundamental to be accommodated by the theory of rational choice and then show that the most basic rules of the theory are commonly violated by decision makers. (Tversky & Kahneman 1986, 252)

The argument for pessimism goes as follows. We provisionally answered the philosophical questions by adopting the standard picture of rationality according to which to be rational is to reason in accordance with the rules of logic, probability, and decision-making (Section 1.3). Then, we moved onto the psychological questions and saw that human reasoning systematically deviates from the rules of logic, probability, and decision-making (Section 1.4). Putting the philosophical and psychological discussions together, we reach a pessimistic conclusion about human rationality.

We now turn to the objections to pessimism, in particular the objections from the ecological rationality programme led by Gigerenzer. As Kahneman and Tversky note, Gigerenzer’s objections include two claims: ‘a conceptual argument against our use of the term “bias”’ and ‘an empirical claim about the “disappearance” of the patterns of judgment that we had documented’ (Kahneman & Tversky 1996, 582).

Gigerenzer’s ‘empirical claim’ is related to the psychological part of the argument for pessimism. Gigerenzer interprets the psychological studies differently, emphasizing the fact that reasoning performance can be improved when the problems are formulated in a different way. As we saw above, in the frequency version of the experiment run by Fiedler (1988), participants’ performances improved significantly. Notice that the proponents of the heuristics and biases programme and the proponents of the ecological rationality programme both recognize the possible performance improvements. Gigerenzer tends to stress the fragility of biases; he says that the biases ‘disappear’ (Gigerenzer 1991). Kahneman and Tversky, in contrast, tend to stress the robustness of the biases; they say that biases can be ‘reduced by targeted interventions but cannot be made to disappear’ (Kahneman & Tversky 1996, 589). But it is far from obvious that this is more than a difference in emphasis or rhetoric. Indeed, Kahneman and Tversky write: ‘There is less psychological substance to [Gigerenzer’s] disagreement with our position than meets the eye’ (Kahneman & Tversky 1996, 589; see also Samuels, Stich, & Bishop 2002).

A more substantial and philosophically interesting disagreement concerns Gigerenzer’s ‘conceptual argument’, which is related to the philosophical part of the argument for pessimism. He rejects the standard picture of rationality, or at least the way in which the standard picture is used in the argument for pessimism. He does not think that reasoning performance should be evaluated in terms of the rules of logic, probability, and decision-making. Among Gigerenzer’s conceptual or philosophical objections, we focus on three main objections, which we discuss in turn in the next section.

1.5 Objections to Pessimism

The Feasibility Objection

According to the first objection, which we call the ‘feasibility objection’, systematic failures to reason in accordance with the rules of logic, probability, and decision-making do not necessarily imply irrationality, because it is unfair and unrealistic to evaluate human reasoning performance in terms of such rules in the first place.

As we noted, the argument for pessimism is based on the standard picture according to which rationality consists in reasoning in accordance with the rules of logic, probability, and decision-making, such as the conjunction rule, Bayes’ rule, and the principle of descriptive invariance. But it would be unfairly simplistic and demanding to evaluate human reasoning in terms of the rules of logic, probability, and decision-making. As Gigerenzer notes, such an unfair and unrealistic evaluation ‘ignores the constraints imposed on human beings. A constraint refers to a limited mental or environmental resource. Limited memory span is a constraint of the mind, and information cost is a constraint on the environment’ (Gigerenzer 2008, 5).

Indeed, not everybody agrees with the claim that standards of good reasoning should be derived from the rules of logic, probability, and decision-making. Some philosophers (e.g., Harman 1999) have argued that human thought should have independent normative standards that reflect human cognitive capacities and limitations. In their accounts, normative standards of rationality are not modelled on formal principles of logic, probability, and decision theory.

Let us consider an example. The standard picture would demand logical consistency among a person’s beliefs, which means that a person should not have beliefs that are logically inconsistent with one another. However, the task of maintaining a logically consistent belief system is extremely demanding given realistic computational constraints. The job requires greater computational resources than those available to human cognitive systems. As Stephen Stich puts it, perhaps evaluating human reasoning in terms of the standard picture might be committed to the ‘perverse’ judgment that ‘subjects are doing a bad job of reasoning because they are not using a strategy that requires a brain the size of a blimp’ (Stich 1990, 27).

The feasibility objection certainly raises a fair worry about the standard picture in its idealized form: that is, rationality requires reasoning perfectly in accordance with the rules of logic, probability, and decision-making. The standard picture needs to be weakened in light of computational and other relevant constraints (although it is not easy to distinguish relevant constraints from irrelevant ones; see Stich 1990). In this vein, Christopher Cherniak argues against upholding logical consistency as an ideal of rationality for human belief systems and endorses a less demanding version of the standard picture of rationality that includes feasibility considerations. He calls it minimal rationality (cf., Cherniak 1990). According to the dictates of minimal rationality, for a belief-like state to be minimally rational it must conform to standards of correct reasoning that are feasible in light of the limitations of human cognitive capacities.

For our purposes, the normative theory is this: The person must make all (and only) feasible sound inferences from his beliefs that, according to his beliefs, would tend to satisfy his desires. (Cherniak 1990, 23)

However, the pessimistic argument cannot be avoided even when computational constraints are taken into account. It would not be fair to expect that rational agents maintain a perfectly consistent belief system when computational constraints are taken into account. However, even when the constraints are taken into account, it would be fair to expect that rational agents do not make particular errors and mistakes, such as the error of assigning a higher probability to Linda being a feminist bank teller than to Linda being a bank teller. After all, in all of the experiments discussed above, some participants do offer the logically or mathematically correct answer, and they do not have ‘a brain the size of a blimp’.

Stanovich (1999) empirically examined this issue by studying the relation between computational capacity (measured by SAT scores) and reasoning performance in reasoning tasks. He found a moderate positive correlation between computational capacity and reasoning performance but concluded that the computational limitation is responsible only for part of the reasoning failures. If correct, this supports the idea that reasoning errors that are found in the heuristics and biases research have little to do with computational constraints; thus, computational constraints cannot be an excuse for their reasoning errors.

The Meaninglessness Objection

The Linda experiment and other probabilistic tasks in the heuristics and biases programme ask participants to assess the probability of a single event (e.g., the probability of Linda being a feminist bank teller). However, according to a frequentist interpretation of probability, it is hard to make sense of the probability of a single event.

According to the (simple) frequency interpretation, a claim about the probability of an event (e.g., the probability of winning a coin toss is 0.5) is understood as talking about the frequency of the event relative to the relevant reference class (e.g., the frequency of winning relative to the total number of tosses of that particular coin). The Linda experiment asks the probability of Linda being a feminist bank teller. On a frequentist interpretation, this amounts to asking the frequency of Linda being a feminist bank teller. What does that mean? The frequency of Linda being a feminist bank teller relative to the total cases of her being a feminist bank teller in her life would be 1. Does that mean, then, that the probability of Linda being a feminist bank teller is 1?

Gigerenzer’s objection is that asking the probability of a single, non-repetitive event – like Linda being a feminist bank teller – is an ill-formed question. Since the question is ill formed and meaningless, there is no correct answer to it. Thus, it is not a mistake to assign a higher probability to Linda being a feminist bank teller than to Linda being a bank teller.

The philosophical and statistical distinction between single events and frequencies clarifies that judgments hitherto labeled instances of the ‘conjunction fallacy’ cannot be properly called reasoning errors in the sense of violations of the laws of probability. (Gigerenzer 1994, 144)

Moreover, as we have already seen (Fiedler 1988), when questions are explicitly framed in the format of frequency with a clearly specified reference class (‘Among the 100 people who fit Linda’s description, how many are bank tellers and how many are feminist bank tellers?’), participants’ performance improves. This reveals that people make apparent ‘mistakes’ when the question is ill formed (asking the probability of a single event with no specified reference class), but do not make mistakes when the question is well formed (asking the frequency of something being the case among a specified reference class). It would seem, then, that people are far from being irrational.

What we call the ‘meaninglessness objection’ can be developed in at least two ways. In other words, there are at least two versions of it. The first version, the ‘factual version’, states that the frequency interpretation of probability is true as a matter of fact and, hence, the probabilistic questions in the heuristics and biases experiments are nonsense as a matter of fact. The second version, the ‘psychological version’, states that frequentism is psychologically true in the sense that the information about probability is represented in the frequentist format in the mind of human agents. Hence, the probabilistic questions in the heuristics and biases experiments are nonsense psychologically (i.e., the participants cannot make sense of the questions).

The factual version of the meaninglessness objection is philosophically bold; it presupposes the truth of the frequency interpretation of probability. The interpretation of probability is a highly contested topic in philosophy, which goes beyond the scope of this book. But it is worth pointing out that the factual version involves a controversial argumentative strategy. After all, (the simplistic version of) frequentism has been criticized exactly because people think that they can meaningfully talk about the probability of a single event (cf., Hajek 2019). It is true that the probability of a single event does not make sense according to frequentism, but it only shows that frequentism is problematic.

The psychological version of the meaninglessness objection is less bold than the factual version. Unlike the factual version, the psychological version is not committed to a particular claim about the interpretation of probability. The problem with the psychological version, however, is that it is not bold enough; if frequentism is not true as a matter of fact but is merely true psychologically, then the probabilistic questions do make sense as a matter of fact and, thus, reasoning biases are real errors as a matter of fact after all.

The psychological version is psychologically problematic as well. It is unlikely that the probabilistic questions in the Linda experiment are psychological nonsense. Apparently, participants do not regard the questions as nonsense; after all, they provided meaningful answers to the questions in the experiments, rather than refusing to answer or demanding clarifications (Samuels, Stich, & Bishop 2002).

The Ecological Rationality Objection

The next objection directly confronts the standard picture of rationality, which defines rationality in terms of the rules of logic, probability, and decision-making. This seems to be where Gigerenzer’s fundamental disagreement with pessimists lies. As Gigerenzer notes, his main disagreement with the pessimism in the heuristics and biases programme is that it ‘does not question the norms [of logic and probability] themselves’ and ‘it retains the norms and interprets deviations from these norms as cognitive illusions’ (Gigerenzer 2008, 6).

The standard picture is problematic, according to Gigerenzer:

Humans have evolved in natural environments, both social and physical. To survive, reproduce, and evolve, the task is to adapt to these environments, or else to change them. […] The structure of natural environments, however, is ecological rather than logical. (Gigerenzer 2008, 7)

The problem identified by Gigerenzer is that the standard picture neglects the role of environment, which is a crucial factor for biological success. As an alternative to the standard picture of rationality, Gigerenzer offers the ecological picture of rationality, which characterizes rationality in terms of cognitive success in the relevant environment. According to the standard picture, rationality requires a fit between the mind and the rules of logic, probability, and decision-making, while ecological rationality requires a ‘fit between structures of information-processing mechanisms in the mind and structures of information in the world’ (Todd & Gigerenzer 2007, 170).

When we evaluate human reasoning performance in light of the ecological picture rather than the standard picture, the pessimistic interpretation is no longer warranted. For example, the information about probability available in the ancient environment was represented in the frequency format (e.g., ‘3 rainy days in 10 days’ rather than ‘30% chance of rain’). As we have seen, human reasoning performance is relatively good when the questions are represented in the frequency format. Thus, human probabilistic reasoning seems to be ecologically rational; it worked successfully in the ancient environment in which probability was represented in the frequency format.

A similar argument can be made about the Wason selection task. One might speculate that information about cheaters (those who receive the benefit of cooperation without contributing to it) was particularly salient in ancient societies (Cosmides & Tooby 1992). Indeed, failing to detect cheaters is a serious challenge to the maintenance of altruistic behaviours (e.g., Trivers 1971). It turns out that the Wason selection task becomes less challenging when the content of the statement to be falsified is an example of a social exchange rule. Alternative versions of the selection task were devised to test the hypothesis that a participant’s performance improves when the statement tested is a cheater-detection rule (Cosmides 1989). For instance, Richard Griggs and James Cox (1982) asked participants to imagine a police officer checking whether people drinking in a bar respect the following rule: ‘If you drink alcohol, then you must be over twenty-one years of age.’ The cards had on their visible sides one of the following: ‘Beer’, ‘Coke’, ‘22 years’, and ‘16 years’. In this situation, the majority of participants (correctly) choose the ‘Beer’ card and the ‘16 years’ card. It could be argued, then, that human deductive reasoning is ecologically rational.

In effect, Gigerenzer introduces an alternative conception of rationality, namely the ecological conception of rationality, and argues that human reasoning meets the requirements of ecological rationality. Thus, we seem to have two conceptions of rationality that yield two different interpretations of the experimental results. Our conclusion is pessimistic (‘Humans are irrational’) when human reasoning performance is evaluated in light of the standard picture. However, our conclusion is optimistic (‘Humans are rational’) when human reasoning performance is evaluated in light of the ecological picture.

A problem with this ‘ecological rationality objection’ is that it seems to conflate biological adaptiveness and rationality. For instance, Stanovich writes:

Evolutionarily adaptive behavior is not the same as rational behavior. Evolutionary psychologists obscure this by sometimes implying that if a behavior is adaptive it is rational. […] I think a conflation of these interests is at the heart of the disputes between researchers working in the heuristics and biases tradition and their critics in the evolutionary psychology camp. (Stanovich 2009, 55–56)

Certainly, it could be biologically adaptive to have a cognitive system that is especially good at dealing with frequencies but not very good at dealing with probabilities. But is such a system not only adaptive but also rational? This issue is closely related to the aim or purpose of cognition, which we will discuss in the next section.

Another problem is that the ecological picture seems to imply that humans used to be rational, but they no longer are. The concern, expressed again by Stanovich, is that human reasoning styles probably used to be successful in the ancient environment, but are no longer successful in the current environment, where probabilities tend not to be represented in a frequentist format, and information about cheaters no longer has special biological significance:

Unfortunately, the modern world tends to create situations where some of the default values of evolutionarily adapted cognitive systems are not optimal. […] [M]odern technological societies continually spawn situations where humans must decontextualize information – where they must deal abstractly and in a depersonalized manner with information. (Stanovich 2004, 122)

It is therefore a challenge to defend optimism even when adopting the ecological picture of human rationality. Our temporary conclusion is that none of the objections reviewed so far constitute a serious threat to the pessimistic interpretation of the experimental results. In the next section, we will explore another issue, which is closely related to the dispute between the standard picture and the ecological picture.

1.6 The Aim of Cognition

Aiming at Truth

Why do we accept the standard picture of rationality rather than, say, the ecological picture? The idea that good reasoning aligns with logical and mathematical rules is often associated with the idea that the aim or purpose of cognition is truth; because the aim of cognition is truth, and because logical and mathematical rules are the best guides to truth, human agents ought to reason in accordance with logical and mathematical rules.

But is the purpose of cognition really truth? Does cognition aim at truth? Perhaps Gigerenzer’s fundamental disagreement with the standard picture lies in his dissatisfaction with the idea of truth as the aim of cognition. He does not believe that truth is the aim of cognition, or at least he does not believe that it is the only aim: ‘[A]daptive behavior has other goals than logical truth or consistency, such as dealing intelligently with other people’ (Gigerenzer 2008, 12). From this perspective, human reasoning should not be evaluated in terms of logical and mathematical rules. Rather, it should be evaluated in terms of its contribution to biological and ecological success.

This takes us to the topic of the aim of cognition, which is the last topic in this chapter. We cannot fully address the issue of the aim of cognition here, but we will identify some considerations that need to be taken into account in a full investigation of this issue. As the aim of cognition is too broad a target for our investigation, let us focus on the mental state of believing and explore what its aim might be.

What does the ‘aim’ of belief mean? There could be more than one correct answer to this question, but our tentative answer is that the ‘aim’ of belief is the function of the belief-forming mechanisms (i.e., the mechanisms that produce beliefs in response to some inputs, including perceptual inputs and other beliefs). To say that the aim of belief is truth is to say that the belief-forming mechanisms have the function of tracking truth (i.e., producing true beliefs, avoiding false ones, etc.) (e.g., Velleman 2000; see also Miyazono 2018).

By ‘function’ we mean so-called ‘aetiological’ function; on this understanding, function aligns with evolutionary design (e.g., Millikan 1989; Neander 1991). In other words, X has the aetiological function of doing Y if Xs were selected for doing Y in their evolutionary history. For example, hearts have the aetiological function of pumping blood: hearts were selected for pumping blood in their evolutionary history. Again, kidneys have the aetiological function of filtering metabolic wastes from blood: kidneys were selected for filtering metabolic wastes from blood in their evolutionary history. Similarly, belief-forming mechanisms have the aetiological function of tracking truth if the mechanisms were selected for tracking truth in their evolutionary history.

Thus, our questions are: ‘Do belief-forming mechanisms really have the function of tracking truth?’ and ‘Were they selected for tracking truth in their evolutionary history?’

There are two tempting arguments against the idea that belief-forming mechanisms have the function of tracking truth. Both arguments, in our view, are problematic.

The first argument is that the function of belief-forming mechanisms cannot be tracking truth because ‘natural selection does not care about truth; it cares only about reproductive success’ (Stich 1990, 62).

This argument might seem compelling at first glance, but it does not stand careful scrutiny. The problem with this argument becomes obvious when we run a parallel argument about hearts. The function of hearts cannot be pumping blood because natural selection does not care about pumping blood; natural selection only cares about reproductive success. The weakness of the parallel argument is clear: the choice between pumping blood and reproductive success as the function of hearts is a false dichotomy. After all, hearts contributed to reproductive success by pumping blood, which is why the function of hearts is to pump blood. Similarly, the choice between tracking truth and reproductive success as the function of belief-forming mechanisms is a false dichotomy. Perhaps belief-forming mechanisms contributed to reproductive success by tracking truth, in which case the function of the mechanisms is to track truth (Millikan 2009; Wilkins & Griffiths 2012).

The second argument is as follows: the function of belief-forming mechanisms cannot be tracking truth because, as we have seen in the heuristics and biases studies, such mechanisms – and reasoning mechanisms especially – lead to systematic errors and biases. In other words, systematic errors and biases constitute evidence that tracking truth is not the function of belief-forming mechanisms.

This argument is not compelling either. Systematically failing to do X is perfectly consistent with having the function of doing X. For example, sperm has the function of fertilizing an egg, although most sperm fail to fulfil their function. Although most sperm fail to fertilize an egg, some sperm do fertilize an egg, and that is how sperm has contributed to reproductive success in its evolutionary history. Similarly, it is possible that belief-forming mechanisms have the function of tracking truth, despite them systematically producing false beliefs. Although these mechanisms often produce false beliefs, it is possible that they do produce true beliefs in many other cases, and that is how belief-forming mechanisms have contributed to survival and reproductive success, at least in their evolutionary history. Let us put this a different way. Most sperm do not fertilize an egg, not because sperm do not aim at fertilizing an egg but because some failure can be tolerated as long as some sperm successfully fertilize an egg in biologically crucial contexts. Similarly, human belief-forming mechanisms exhibit systematic errors and biases, not because belief-forming mechanisms do not aim at tracking truth but because some errors and biases (even systematic ones) can be tolerated as long as some beliefs successfully track truth in biologically crucial contexts.

To sum up, the arguments above fail to refute the idea that belief-forming mechanisms have the function of tracking truth. In other words, the idea that truth is the aim of belief has not been ruled out yet. Before closing this chapter, we would like to mention a particularly relevant discussion by Ryan McKay and Daniel Dennett (2009).

Positive Illusions

McKay and Dennett endorse the view that in most cases adaptive beliefs are true beliefs. But there is an exception: positive illusions (i.e., optimistically biased judgments about oneself [see Box 1C for a list of the most common positive illusions]).

BOX 1C: Positive Illusions

Definition:

Positive illusions are patterns of beliefs about oneself, the world, and the future characterized by ‘systematic small distortions of reality that make things appear better than they are’ (Taylor 1989, 228).

Types of positive illusions:

• Illusion of control – to believe that one can control independent, external events (e.g., Langer & Roth 1975).
• Illusion of superiority – to believe that one is better-than-average in a variety of domains, including attractiveness, intelligence, and moral character (e.g., Brown 2012; Wolpe, Wolpert, & Rowe 2014).
• Optimism bias or unrealistic optimism – to predict that one’s future will be largely positive and will yield progress, and that negative events will not be part of one’s life (Sharot 2011; Weinstein 1980).

Examples:

• Control: You believe that you have a better chance at winning in a betting situation if you are the one dealing the cards or throwing the dice.
• Superiority: You believe that you are smarter than your colleagues even if the evidence does not support that.
• Optimism bias: You believe that you are immune to a highly infectious disease that other people are catching.

According to McKay and Dennett, positive illusions are ‘adaptive misbeliefs’: in other words, they are false beliefs that are adaptive in themselves. For example, your positive illusion about your health might enhance your actual health and life expectancy. If McKay and Dennett are correct, truth might not be the function of positive illusions (or the mechanisms that produce them).

Please note that positive illusions are importantly different from the errors that are due to the adoption of heuristics. In the case of heuristics, false beliefs are merely tolerated; the errors themselves do not have evolutionary value. However, in the case of positive illusions, false beliefs positively contribute to fitness. The function of positive illusions (or, more precisely, the underlying mechanism responsible for positive illusions) is perhaps something other than tracking truth.

Dennett and McKay’s argument is certainly compelling. But is it really true that positive illusions do not have the function of tracking truth? As Ruth Millikan (2004) points out, it may be too hasty to draw that conclusion. For example, it is conceivable that the original and basic role of positive illusions is tracking truth, but they later acquired the other derivative roles of contributing to survival and reproduction in a different way, which could interfere with the original function in some cases: ‘[M]any biological systems ride piggyback on systems developed earlier for other purposes. Systems whose jobs were to distort certain beliefs would have to ride on more general systems whose basic jobs were to produce true beliefs’ (Millikan 2004, 86).

Another worry is that some forms of optimism involving a denial of existing threats are maladaptive, and the challenge is how to distinguish them from positive illusions. In 2020, people all around the world experienced the devastating effects of COVID-19. Some people seemed to firmly believe in their relative safety and immunity (‘I won’t be infected,’ ‘Even if I were infected, I would just have mild symptoms,’ etc.), and due to underestimating their personal risk they also exhibited reckless behaviour. For the purpose of survival, denying health threats does not seem to be a good strategy – especially during a global pandemic. We can imagine that our ancestors were more frequently exposed to life-threatening diseases and viruses. In that ancient environment, the negative impact of simplistic optimism may have easily outweighed its positive impact.

Positive illusions have been conceived as the capacity to respond effectively to threats, not as the denial of those threats, and thus they may be immune from this problem (Bortolotti 2018; Taylor 1989). However, questions remain as to how to distinguish non-adaptive optimism from adaptive positive illusions.

1.7 Summary

• The standard picture of rationality claims that rationality consists in reasoning in accordance with the rules of logic, probability, and decision-making. However, studies of reasoning errors and biases, in particular the ones in the heuristics and biases programme, reveal that we systematically fail to reason in accordance with the rules of logic, probability, and decision-making. Thus, if we adopt the standard picture and take the results of the empirical studies at face value, we draw the pessimistic conclusion that humans are irrational.
• There are three main objections from optimists, most notably Gigerenzer: (1) according to the feasibility objection, the standard picture fails to take relevant constraints into account; (2) according to the meaninglessness objection, the results of the reasoning studies should be dismissed, in particular the probabilistic ones, because they involve meaningless questions about the probability of single events; and (3) according to the ecological rationality objection, we should replace the standard picture with the ecological picture, which evaluates human cognition not in terms of logical and mathematical rules but rather in terms of biological success in the relevant environment. None of these objections, however, convincingly refutes the pessimistic interpretation of the experimental results.
• The standard picture is related to the idea that the aim of cognition is tracking truth. This idea has some plausibility, but there are some interesting cases, such as the case of positive illusions, in which beliefs seem to have a function that is distinct from tracking truth.

Further Resources

Articles and Books

Among the active participants in the rationality debate, Daniel Kahneman and Gert Gigerenzer are a pleasure to read. We recommend two books where they present their take on the relationship between intuition and rationality in an accessible way: Kahneman’s Thinking, Fast and Slow (2011) and Gigerenzer’s Gut Feelings: The Intelligence of the Unconscious (2007).

The Science article by Tversky and Kahneman, ‘Judgment under uncertainty: Heuristics and biases’ (1974), is essential for anyone interested in reasoning and rationality. This and other important papers in the heuristics and biases programme are included in a collection of papers: Judgment under Uncertainty: Heuristics and Biases (1982), edited by Kahneman, Slovic, and Tversky. To get a good understanding of the rationality debate, see the articles by Kahneman and Tversky and by Gigerenzer in the journal Psychological Review (1996). It would also be helpful to read the following: Without Good Reason: The Rationality Debate in Philosophy and Cognitive Science (1996) by Edward Stein, where the notion of the standard picture of rationality is introduced and discussed; the paper by Richard Samuels, Stephen Stich, and Michael Bishop (2002), where the debate is summarized and reinterpreted; and Chapter 1 of Lisa Bortolotti’s Irrationality (2014), where the significance of the debate for our appreciation of human rationality is examined.

Online Resources

Kahneman’s research is summarized in his Nobel Prize Lecture in 2002, ‘Maps of bounded rationality’, which is available online. You can also watch Gigerenzer’s TEDx talk ‘How do smart people make smart decisions?’ (2012).

It may also be useful to read Gregory Wheeler’s Stanford Encyclopedia of Philosophy entry on bounded rationality (2020), and Andrea Polonioli’s interviews with Thomas Sturm on the science of rationality (2017) and with Ralph Hertwig on biases, ignorance, and adaptive rationality (2016), both of which are available on the Imperfect Cognitions blog.

Questions

1. What do you mean when you say that a person is rational? How do you judge whether a person is rational or not?
2. What do pessimism and optimism about human rationality have in common? What sets them apart?
3. Is the standard picture of rationality a good measure of how good human reasoning is? Why?
4. To what extent should human agents replace the use of heuristics with the application of the rules of logic, probability, and decision-making? Is it even feasible for human agents to do so?

2.1 Introduction

It has been a popular idea in the philosophical tradition – as popular as the idea that humans are rational animals – that humans are self-knowing animals (where self-knowledge is the knowledge humans have about their own mental lives). In fact, both rationality and self-knowledge have been described as distinctive of humans: among all animals, human beings are the only ones that can reason rationally and think about their own beliefs, desires, emotions, and so on. The idea of humans as rational animals and the idea of humans as self-knowing animals are not independent of one another. Rationality and self-knowledge are intertwined, and they jointly support human agency. The sophisticated capacity to consider one’s own beliefs and reflect on one’s own choices (in deliberation and self-examination or in interpersonal settings such as in a discussion with others) seems to rely on good reasoning and self-knowledge.

It turns out that psychological studies in the 20th and 21st century threaten not only the idea that humans are rational animals but also the idea that humans are self-knowing animals. For example, on the basis of a wide range of empirical studies, Timothy Wilson provides a ‘not […] very encouraging’ portrait of self-knowledge according to which ‘[p]eople have limited access to their own personalities, the reasons for their responses, their own feelings, and how they will feel in the future’ (Wilson 2002, 158). The main aim of this chapter is to examine relevant psychological studies and see whether their results support such a pessimistic portrait of self-knowledge.

We start by clarifying the idea that humans are self-knowers (Section 2.2) and then examine the implications of relevant psychological studies (Section 2.3). These studies threaten some ideas about how and to what extent humans can attain self-knowledge. In particular, the studies seem to threaten the idea that humans have non-observational access to their own minds (i.e., they can know their minds without observing and interpreting their own outward behaviour). Empirical evidence provides numerous and striking examples of humans actually relying on self-observation or self-interpretation to learn what they are thinking or feeling – figuring out their own minds in the same way that they figure out other minds, by observing and interpreting outward behaviour. The crucial question is whether self-observation and self-interpretation happens in some cases but not all cases (in which case humans do have some non-observational access to their minds, as discussed in Section 2.4) or whether it happens in all cases (in which case they completely lack non-observational access, as discussed in Section 2.5). Other interesting questions intersect with issues about the process humans rely on to attain self-knowledge: ‘Is self-knowledge fallible?’, ‘Do we know our minds better than an external observer?’, and ‘To what extent can human agents know their own mind?’ The empirical evidence has something important to contribute to such questions too.

This chapter defends a modest view of human self-knowledge. Although human agents are capable of knowing their own minds non-observationally, they rely on the process of self-observation and self-interpretation more frequently than previously acknowledged, which suggests that their knowledge of themselves is not always peculiar or special.

2.2 Clarifying Self-Knowledge

Self-Knowledge and Its Targets

Claire hears one of her friends saying ‘Claire is so mean! I have known her since school and she has never bought me a drink, not even on my birthday!’ Claire always thought of herself as a generous person. Her friend must be wrong!

Last night, Jagbir met a handsome man on a roller-coaster ride at the local fair when they were asked to share the same two-seater. Now she can’t stop thinking about him. Jagbir thinks she has fallen in love with him. Will she be able to find out who he is?

Claire believes that she is a generous person. Jagbir is convinced that she has fallen in love. But are they right? Are they good self-knowers? More generally, how well do humans know themselves? To answer this question, we need to investigate two issues: the philosophical questions about self-knowledge (e.g., ‘What does it mean to have self-knowledge?’, ‘What is necessary for a person to have self-knowledge?’, and ‘What is sufficient for a person to have self-knowledge?’) and the psychological questions about self-knowledge (e.g., ‘Do humans satisfy a sufficient condition for having self-knowledge?’, ‘Do humans fail to satisfy a necessary condition for having self-knowledge?’, and ‘What do empirical studies say about these issues?’). We discuss the philosophical questions in this section (Section 2.2) and move on to the psychological questions in the next (Section 2.3).

Self-knowledge is knowledge of one’s own mind. Here we do not need to commit to a full theory about the nature of knowledge, which is the focus of epistemology; we are only committed to the view that knowledge involves belief (i.e., when X knows that P, X believes that P). This means that self-knowledge involves beliefs about one’s mind.

Self-knowledge is knowledge of one’s own mind. One can know about one’s own mental states (such as the desire to eat a cheeseburger or the wish to become an astronaut), one’s psychological processes (such as the process of calculating the sum of two numbers), one’s psychological capacities (such as the capacity to remember people’s names in a list quickly and accurately), one’s dispositions (such as the tendency to have paranoid thoughts), one’s character traits (such as whether one is generous or mean), one’s deep values (such as valuing family more than income), and so on.

We cannot assume that all forms of self-knowledge are similar and equally good. Perhaps humans have relatively good self-knowledge of some of the items above but not others. For example, it is likely that humans have a better grasp of their own beliefs and desires – what Quassim Cassam (2014) calls ‘trivial self-knowledge’ – than of their own dispositions, character traits, and deep values – what Cassam calls ‘substantive self-knowledge’. Knowing one’s own beliefs and desires seems to be relatively simple; people just know what they believe and what they want. In contrast, knowing one’s own dispositions, character traits, and deep values requires serious investigation, including observation and analysis of one’s past and present behaviour.

Claire is angry with her friend for having described her as a mean person, but did her friend have a point? Claire needs to think about her past behaviour. Did she ever buy him a drink? Knowledge of one’s own dispositions, character traits, and deep values is typically attained in the same way as knowledge of other people’s dispositions, character traits, and deep values – namely by thinking about the past and interpreting the available information. If Claire cannot remember ever buying her friend a drink, then maybe she is not as generous as she thought she was.

Again, knowledge of one’s psychological processes is not usually available to conscious awareness and thus cannot be introspected. Jagbir believes that she has fallen in love with the man from the roller-coaster ride, but she may not know what caused her emotional state of arousal during the ride nor what changed in her body when the emotion took hold. Moreover, it is possible that knowing how her emotional state emerged will reveal that it is not love that she is feeling, but rather a sense of excitement caused by the thrill of the ride that she mistakenly categorized as attraction for the handsome man.

As Wilson writes:

People possess a powerful, sophisticated, adaptive unconscious that is crucial for survival in the world. Because this unconscious operates so efficiently out of view, however, and is largely inaccessible, there is a price to pay in self-knowledge. (Wilson 2002, vii)

We have seen some examples of unconscious reasoning processes in the previous chapter. The heuristics guiding our reasoning processes might not be available to consciousness (although the output of the heuristic processes is usually available). For example, before the studies conducted by Kahneman and Tversky, people may not have been aware of the fact that they relied on the heuristic of representativeness when establishing the probability of events in the Linda experiment (although they were aware of the output of the heuristic process). It is overwhelmingly plausible that human agents do not have introspective access to unconscious cognitive processes of this kind: knowledge about the heuristic processes that guide human reasoning requires agents to interpret the results of sophisticated psychological experiments.

Self-Knowledge: Privileged and Peculiar

Self-knowledge is a form of knowledge. But philosophers tend to assume that self-knowledge is not just an example of knowledge among others; there is something special or distinctive about it. As we will see, the crucial issue in this chapter is whether self-knowledge is really special or distinctive.

In what sense would self-knowledge be special or distinctive? The following example from Alex Byrne is helpful:

Consider Jim, sitting in his office cubicle. Jim believes that his pen looks black to him; that he wants a cup of tea, that he feels a dull pain in his knee, that he intends to answer all his emails today; that he is thinking about paperclips; that he believes that it is raining. Jim also has various equally humdrum beliefs about his environment: that it is raining, that his pen is black, and so on. Furthermore, he has some opinions about the psychology of his officemate Pam. He believes that her pen looks green to her; that she wants a cup of coffee, that her elbow feels itchy; that she is thinking about him; that she believes that it is raining. (Byrne 2018, 5–6)

Among the three groups of beliefs in this example, one might think that the first group, Jim’s beliefs about his own mind, is more epistemically privileged than the other two groups, Jim’s beliefs about the environment and Jim’s beliefs about Pam’s mind.

The notion of ‘being epistemically privileged’ can be understood in multiple ways: ‘epistemically privileged’ can mean ‘more reliable’, ‘more likely to be true’, ‘more likely to be knowledge’, and so on. In any case, the basic idea is clear enough: it is relatively easy to imagine that Jim’s belief that it is raining is false. Perhaps it was raining one hour ago when Jim looked out of the window. He still believes that it is raining one hour later, but it stopped raining and he did not realize. Again, it is not difficult to imagine that Jim’s belief that Pam wants a cup of coffee is false. Perhaps Jim mistakenly assumes that all his co-workers, including Pam, are coffee lovers, but Pam prefers tea to coffee.

In contrast, it is relatively difficult to imagine how Jim’s belief that he believes that it is raining is false, or how his belief that he wants a cup of tea is false. When Jim fails to notice that it has stopped raining, his belief that it is raining (his belief about the environment) is false. However, his belief that he believes that it is raining (his belief about his own belief) is true. Again, when Jim fails to notice Pam’s preference for coffee, his belief that Pam wants tea (his belief about Pam) is false. However, Jim’s belief that he wants a cup of tea (his belief about his own desire) is true. Epistemic privilege is one aspect in which self-knowledge is said to be special and distinctive.

There is another aspect in which self-knowledge is said to be special and distinctive. Self-knowledge is thought to be acquired by a peculiar process that is different in kind from the process of knowing about the environment or about other minds. For example, Jim’s belief that he believes that it is raining is acquired by a peculiar process that is different in kind from the process of Jim acquiring the belief that it is raining or the belief that Pam believes that it is raining. The peculiar process is only available to Jim when he tries to know about his own mind, or to Pam when she tries to know about her own mind. The process is not available to Jim when he tries to know Pam’s mind, nor to Pam when she tries to know Jim’s mind. The peculiarity of the process by which self-knowledge can be acquired has been described in different ways, such as ‘introspective’, ‘direct’, ‘non-inferential’, ‘authoritative’, or ‘a priori’ (see Box 2A for some clarification of the meanings of these terms). This peculiarity by which one acquires knowledge of one’s own mental states is the second aspect in which self-knowledge is said to be special and distinctive.

BOX 2A: Key Terms in the Philosophical Literature on Self-Knowledge

Introspection (literally, ‘looking inward’ or ‘looking within’) is one way of learning about one’s own mind, by simply realizing what one’s own thoughts and feelings are. For example, Swati knows that she is feeling sad by introspection if all it takes for her to know about her sadness is to turn her attention to how she is feeling.

A priori knowledge is knowledge that can be attained independently of any experience that justifies the knowledge. For example, Swati knows a priori that 2+2=4 or that all bachelors are unmarried without relying on observation or induction.

Direct (non-inferential) self-knowledge is knowledge that can be attained by being aware of one’s own mental state without becoming aware of anything else (it is often called ‘knowledge by acquaintance’). For example, Swati knows that she feels sad directly because her feeling of sadness is something she introspects. Introspecting the feelings requires no inferential process.

First-person authority is the idea that one is more authoritative about the content of one’s own mind than about the contents of other people’s minds. For example, Swati’s friends do not challenge her assertion that she feels sad because they assume that she is the expert when it comes to the contents of her own mind, as they are all experts about the contents of their own minds when they attribute mental states to themselves.

Privileged access is the idea that one’s judgment about own mental state is more likely to be true, or more likely to be knowledge, than one’s judgment about other issues. For example, when Swati judges that she is sad, that judgment is more likely to be true (i.e., she is in fact sad) or more likely to be knowledge (i.e., she knows that she is sad) than her judgment that one of her friends is sad or that it is raining outside.

Peculiar access is the idea that one can appeal to a peculiar and special method to acquire self-knowledge. For example, Swati appeals to a peculiar and special method when she judges that she is sad. Others cannot appeal to this method when they judge that she is sad.

What does it mean to say that self-knowledge is peculiar? What is the difference in kind between Jim’s access to the contents of his own mind and his access to the contents of other minds or his access to the environment? Fully answering this question requires a discussion of how knowledge of other minds is acquired; this is the topic of Chapter 8, in which we discuss the nature of mindreading (i.e., the process of attributing mental states to others). For now, it can be understood in the following way:

Our access to others’ minds is similar to our access to the non-psychological aspects of our environment. Jim knows that his pen is black by seeing it; Pam could know the same thing by the same method. Likewise, Jim knows that Pam wants a coffee by observing her behavior. Anyone else – including Pam – could know the same thing by the same method. Our peculiar access to our own minds is not like this: one can come to know that one wants a coffee without observing oneself at all. (Byrne 2018, 8, our emphasis)

The idea is that peculiarity has something to do with observation, or the lack thereof. Jim’s knowledge of his environment and of Pam’s mental states is observational or interpretive: his knowledge of the pen depends on the observation of the pen and its properties, and his knowledge of Pam’s mental states depends on him observing and interpreting her behaviour. If Pam answers ‘Coffee, please’ to the question ‘Would you like something to drink?’, Jim can observe this exchange and conclude that she wants a cup of coffee. In contrast, Jim does not need any observation of his own behaviour to know that he believes that it is raining or that he wants a cup of tea. The fact that there is no need for observation or interpretation is what makes self-knowledge peculiar. (Note that this does not imply that Jim cannot know his mental states by observing his behaviour. It is obvious that Jim, if he wants to, can gain knowledge of his mental states by observing his behaviour. The point is that he does not have to observe his behaviour to know his own mental states. Observation and interpretation are not necessary.)

To sum up, our answer to the philosophical questions about self-knowledge is that self-knowledge is knowledge (a form of belief) of one’s own mind and of its constituents, such as one’s own mental states, processes, dispositions, character traits, and so on. Self-knowledge seems to be special and distinct in the sense that, first, it is epistemically privileged in comparison to other forms of knowledge and, second, it is acquired through a peculiar and non-observational process.

2.3 Challenges to Peculiarity

Peculiarity and Parity

Let us move on to the psychological questions about self-knowledge. How well do humans know their own minds according to psychological inputs? Do they really have privileged access? How about peculiar access? The psychological studies we will see below could threaten ideas of self-knowledge both as privileged and as peculiar. We will outline some issues concerning self-knowledge as privileged, but our main focus will be peculiarity.

Several researchers in psychology and philosophy are sceptical about the peculiarity thesis of self-knowledge for empirical reasons (Bem 1972; Carruthers 2009, 2011; Gopnik 1993; Wegner 2002; Wilson 2002). These researchers emphasize the similarity, rather than the difference, between self-knowledge and knowledge of other minds. In other words, their emphasis is not on the peculiarity of self-knowledge but rather on the parity between self-knowledge and knowledge of other minds.

Individuals come to ‘know’ their own attitudes, emotions, and other internal states partially by inferring them from observations of their own overt behavior and/or the circumstances in which this behavior occurs. Thus, to the extent that internal cues are weak, ambiguous, or uninterpretable, the individual is functionally in the same position as an outside observer, an observer who must necessarily rely upon those same external cues to infer the individual’s inner states. (Bem 1972, 2)

[T]he system that is employed when one identifies and attributes mental states to oneself is none other than the mindreading system that underlies one’s capacity to attribute mental states to other people. […] In order to attribute thoughts to oneself, then, the mindreading faculty is forced to interpret the available sensory evidence. This can concern one’s physical circumstances and overt behavior, or it can involve one’s own visual imagery, affective feelings, and inner speech. The result is that all access to one’s own propositional attitudes is sensory-based and interpretive in nature. (Carruthers 2013, 145–146)

But should we deny the idea of peculiar self-knowledge? Is it really true that self-knowledge is observational and interpretive just like knowledge of other minds? We divide this question into two sub-questions. The first sub-question is whether self-knowledge is observational and interpretive in at least some cases. The second sub-question is whether self-knowledge is observational and interpretive in all cases.

If the answer to the first sub-question is ‘No’ (which implies a ‘No’ answer to the second sub-question), then the idea of peculiar self-knowledge is unscathed. We call this the ‘extreme peculiarity view’. If, in contrast, the answer to the second sub-question is ‘Yes’ (which implies a ‘Yes’ answer to the first sub-question), then the idea of peculiar self-knowledge is denied completely. We call this the ‘extreme parity view’. Peter Carruthers, for example, defends this extreme position according to which we have no peculiar self-knowledge of our own propositional attitudes and decisions: ‘[T]here is just a single faculty involved in both forms of activity [knowing one’s own mental states and knowing somebody else’s], using essentially the same inputs, which are all perceptual or quasi-perceptual in character’ (Carruthers 2009, 123).

If the answers are ‘Yes’ to the first sub-question and ‘No’ to the second sub-question, the idea of peculiar self-knowledge is partially challenged, but not completely denied. Peculiar self-knowledge can occur; it is just that it is not as widespread as we might assume. We call this the ‘moderate view’. Daryl Bem seems to be the defender of the moderate view. In the quote above, he says that we know our mental states ‘partially by inferring them from observations of their own overt behavior and/or the circumstances in which this behavior occurs’, and this happens not always but rather when ‘internal cues are weak, ambiguous, or uninterpretable’ (Bem 1972, 2).

Thus, we have three positions: the extreme peculiarity view, the extreme parity view, and the moderate view. Our main task below is to see which view is the one that the empirical evidence recommends.

Dissonance Studies

Let us start by looking at what Bem takes to be the evidence for self-observation: instances of cognitive dissonance.

In a classic study by Leon Festinger and James Carlsmith (1959), participants in one condition (the $1 condition) performed repetitive and boring tasks for one hour. Then they received $1 compensation for playing a role which involved the job of telling the next person who was waiting for her turn that the tasks were fun and enjoyable. Participants in another condition (the $20 condition) performed the same tasks for the same time. They received $20 compensation for playing the role and telling the next person that the tasks were fun and enjoyable. (The remaining participants were in a control group and received no monetary compensation.)

Afterwards, all participants indicated how much they enjoyed the tasks. The result was that those in the $1 condition rated the tasks significantly more enjoyable than those in the $20 condition.

Bem argues that this experimental result is explained by the idea of self-observation in a fairly straightforward manner. To see how Bem’s idea works, consider the following case. Let us think about a participant in the $1 condition, Sally, and another in the $20 condition, Christopher. You observe Sally’s and Christopher’s behaviour and try to figure out how much they enjoyed the tasks. You observe two facts about Sally’s behaviour: she told the next person that the tasks were fun (A1) and that she received only $1 as compensation (B1). On the basis of this pair of observations, you conclude that Sally enjoyed the tasks a lot; you think ‘Sally told somebody else that the tasks were fun despite only receiving $1 as compensation, so the tasks must be really fun.’ Next, you observe two facts about Christopher’s behaviour: he told the next person that the tasks were fun (A20) and he received $20 as compensation (B20). On the basis of this pair of observations, you conclude that Christopher did not enjoy the tasks very much; you think ‘Christopher told somebody else that the tasks were fun, but he obviously did so for the money’ (cf., Bem 1967).

Now think about Sally and Christopher themselves. According to Bem’s self-perception theory, Sally and Christopher do not have peculiar access to how much they enjoyed the tasks. Thus, they need to figure out how they felt about the tasks by observing their own behaviour, just like you do when you assess how much they enjoyed the tasks. Sally observes two facts about her own behaviour: A1 and B1. Based on this pair of observations, Sally concludes that she enjoyed the tasks a lot. Next, Christopher observes two facts about his own behaviour: A20 and B20. Based on this pair of observations, Christopher concludes that he did not enjoy the tasks. Thus, Bem’s self-perception theory nicely predicts and explains Festinger and Carlsmith’s result. If Bem is correct, then this experiment supports the idea of self-observation.

But how plausible is Bem’s account of Festinger and Carlsmith’s experiment? Bem’s theory is presented as an alternative to Festinger’s own account. According to Festinger’s cognitive dissonance theory, the experimental result is explained by the motivation for reducing dissonant ideas rather than the process of self-observation.

In the $1 condition, Sally is aware of three facts: that she told the next person that the tasks were fun (A1); that she received only $1 as compensation (B1); and that the tasks were boring (C1). A1, B1, and C1 are dissonant in the sense that they do not fit together well. Sally might be thinking ‘If the tasks are boring, why did I tell the next person that the tasks were fun, and I was paid only $1 for it?’ So, motivated to dissolve the dissonance creating a feeling of discomfort, Sally revises her belief about C1, which is arguably the easiest observation to revise, thus coming to believe that the tasks were actually fun.

The same thing does not happen in the $20 condition. Christopher is aware of three facts: that he told the next person that the tasks were fun (A20); that he received $20 as compensation (B20); and that the tasks were boring (C20). Christopher’s observations are not dissonant. It is coherent to believe that one receives a high compensation for being asked to perform a boring task and to lie to other participants. Christopher does not need to revise his belief about C20 to eliminate a feeling of discomfort due to dissonance.

Note that Festinger’s account is not committed to the idea of self-observation. In fact, his account seems to presuppose that Sally and Christopher have non-observational access to how much they enjoyed the tasks (C1 and C20). Otherwise, Sally cannot be aware of C1, which is dissonant with A1 and B1.

Thus we have two accounts of Festinger and Carlsmith’s experiment: Bem’s self-perception account, which supports the idea of self-observation, and Festinger’s cognitive dissonance account, which does not support the idea of self-observation. Which is more plausible? Answering this question is not easy because the two theories predict the same results in many scenarios. In other words, many scenarios (just like the one in Festinger and Carlsmith’s study) can be adequately explained by either of the two theories. But only the cognitive dissonance theory predicts that affective or motivational factors (e.g., the feeling of discomfort) are involved in dissonant cases, and this hypothesis seems to be supported by further studies. In any case, we will not pursue the debate between the cognitive dissonance theory and the self-perception theory here (see Cooper 2007 for an overview); it is sufficient to say at this point that it is not obvious that Bem’s self-perception account is the best explanation of dissonance cases.

Confabulation Studies

Let us now move on to another body of research relevant to the investigation of self-knowledge: the literature on broad confabulation. In this context, confabulation means ‘making up a story’ without intending to deceive. It has been defined as an ill-grounded claim people make when they do not realize that their claim is ill grounded (Hirstein 2005) or as an unsubstantiated claim people make in response to a question that they cannot answer because the relevant information is not known or accessible to them (Turner & Coltheart 2010). For a different understanding of confabulation, where confabulation is a symptom of disorders characterized by severe memory impairments (narrow confabulation), see Chapter 7.

The cases of confabulation that we will consider here are those where agents offer an inaccurate explanation of their own behaviour (see the examples in Box 2B). For example, when asked why one made a decision, one might answer by offering an explanation that does not reflect the actual process behind the decision. Suppose you ask Karl why he chose to watch the film Casablanca and he says that he wanted to see a classic film starring Humphrey Bogart when, in fact, his choice was due to the fact that on the way home from work he had seen a poster advertising Morocco as a fun holiday destination and that ‘primed’ him to select a film that was set in that country. Karl has not lied to you; he is convinced that he is providing an accurate explanation for his choice of film.

BOX 2B: Some Contexts in Which People Confabulate

Consumer choice:

People offer explanations for their consumer preferences that do not match the most likely causal processes leading to their choices and do not include causally relevant factors. For example, one might choose a product due to an advertisement having made that product salient to them but explain their choice in terms of the product having some qualities that the alternatives do not have. See the stockings experiment (Nisbett & Wilson 1977) reviewed in this section.

Attitude explanation:

People offer explanations for their attitudes (aesthetic or moral judgments, political attitudes, etc.) that are unlikely to reflect the most salient factors contributing to having those attitudes. For example, people in a romantic relationship asked to evaluate their relationship after receiving some negative information about their partner do not realize that their subsequent negative evaluation is largely determined by receiving the negative information (Wilson & Kraft 1993).

Hypnosis:

People under hypnosis tend to explain their actions without acknowledging the role of the hypnotic suggestion on their behaviour. For example, after receiving the hypnotic suggestion that they will see a stranger in the mirror, people under hypnosis claim that the image they see in the mirror is not theirs, and explain this by finding physical differences between themselves and the image (such as different eye colour) (Bortolotti, Cox, & Barnier 2012).

Confabulation fits nicely with the idea that humans lack peculiar access to their own minds. If, for example, Karl lacks peculiar access to the real cause of his choice of film, namely the priming effect, then he needs to consider observational inputs to figure out why he made that choice. He eventually reaches the reasonable (but false) conclusion that he wanted to watch a classic film starring Humphrey Bogart. How often do such confabulations happen? Is self-knowledge always confabulatory? Or is it only sometimes confabulatory?

Confabulation studies can be divided into two groups. The first group of studies involves confabulation in unusual circumstances. The most notable example is the study by Michael Gazzaniga (2000), who observed confabulations among so-called ‘split-brain patients’. In split-brain cases, the corpus callosum (a bundle of fibres connecting the right hemisphere and the left hemisphere of the brain) is severed as a method of treating severe epilepsy. Gazzaniga’s split-brain patient confabulated about the movement of his left hand. The left hand is controlled in the right hemisphere of the brain. The verbal centre of the brain is in the left hemisphere. With the corpus callosum severed, the verbal centre in the left hemisphere loses access to the right hemisphere, which is responsible for the left hand. Thus, the patient (or the patient’s left hemisphere) makes up a false story when asked to explain the reason why the left hand does what it does. The split-brain case provides good evidence that agents attain knowledge of themselves (e.g., of the reasons for their current or past behaviour) based on self-observation and self-interpretation. It is likely that what they do when asked for information they do not have is simply interpret their own behaviour as they would interpret the behaviour of another agent. But there are obvious limitations to these studies. One problem is that the relevant cases can be relatively rare, which makes it difficult to conduct reliable, large-scale empirical studies. Another problem is that it is far from obvious that the finding about these unusual cases can be generalized to more common, everyday cases of self-knowledge. The confabulations in these special cases do not tell us much about how self-knowledge works in general.

The second group of studies investigates confabulation in everyday cases and is more relevant to our discussion about the peculiar nature of self-knowledge. We will now discuss two classic studies of this kind.

The Stockings Experiment. In their landmark paper ‘Telling more than we can know: Verbal reports on mental processes’, Richard Nisbett and Timothy Wilson (1977) describe an experiment (disguised as a consumer survey) conducted in a bargain store just outside of Ann Arbor, Michigan. Participants were asked to evaluate the quality of four pairs of nylon stockings (A, B, C, and D, from left to right) and indicate which pair was the best. The result revealed a strong position effect on the choices by the participants: D was frequently chosen (A = 12%, B = 17%, C = 31%, and D = 40%). Participants’ choice was based on the position, not the quality, of the pair of stockings. In fact, all pairs of stockings were exactly the same; there was no difference in quality. However, when asked the reason for their choice, participants did not acknowledge the position effect. Instead, they confabulated and spoke about some qualities of their chosen pair of stockings being the main reason for their choice, such as: ‘I chose this pair of stockings because its knit is superior.’ When the experimenter explicitly asked participants about the relevance of the position, participants adamantly denied the idea.

Choice Blindness. More recently, Petter Johansson and his colleagues in Lund performed a series of experiments that show a different type of confabulation at play (Johansson et al. 2005; Johansson et al. 2006). The original study by Johansson and colleagues (2005) is inspired by the stockings experiment by Nisbett and Wilson but features two important differences.

First, participants in this experiment chose from a range of non-identical items. The items were photos of the faces of female strangers (Photo A and Photo B) and participants had to choose which one they found most attractive.

Second, this study involves a deceptive manipulation. After participants made their choice, and before they could be asked about the reason for it, the photos were swapped. Unbeknown to the participants, the experimenter swapped the photo that had been chosen by the participants (say, Photo A) with one that had not been chosen (say, Photo B). Next, the experimenter asked the participants why they chose Photo B, pretending that no swapping had taken place. The surprising result of the experiment was that no more than 26% of the photo swaps were noticed by participants, who we could say were choice-blind. Indeed, many participants confabulated by giving articulate and plausible reasons for a choice they apparently never made: ‘She looks like an aunt of mine, and she seems nicer than the other one,’ ‘I thought she had more personality, in a way. She was the most appealing to me,’ and so on.

The behaviour observed by Nisbett and Wilson in the stockings experiment and by Johansson and colleagues in the choice blindness studies are striking cases of self-observation and self-interpretation. When explaining why they made their choices, participants do not exhibit any peculiar access to their mental states and processes. Instead, they rely on the same observational and interpretive processes on which they rely when they think about why other people make choices.

For example, think about a hypothetical case in which you observe a participant, Angelina, in the stockings experiment by Nisbett and Wilson. Angelina chooses pair D and you are asked why she chose the pair rather than the other pairs. Perhaps you will answer in a way that makes her choice sensible or reasonable; you might say ‘She chose it because she thinks that its knit is superior’ or ‘She chose it because she thinks it is more elastic than the others.’ You will not say ‘She chose it because of the position effect’ unless you are a psychology student. As Nisbett and Wilson point out, ‘[S]uch factors [e.g., the position effect] should seem particularly implausible as reasons for liking or disliking an object […] [and] it seems outrageous that such a judgment as one concerning the quality of a nightgown might be affected by its position in a series’ (Nisbett & Wilson 1977, 252). The experiment’s result can be nicely explained by the self-observation hypothesis: Angelina needs to observe her own behaviour when reporting why she chose pair D, just like you need to when you explain why she chose pair D.

Again, think about the hypothetical case where you observe a participant, Neville, in the choice blindness study by Johansson and colleagues. Neville chooses Photo A and you are asked why he chose Photo B (since Photo A has been switched with Photo B). Because you do not have any direct or introspective access to Neville’s preference for Photo A, you might fail to recognize the swap. Perhaps you will answer the question so that his choice seems sensible or reasonable; you might say ‘He chose it because he thinks that the woman in the photo seems nicer than the other one’ or ‘He chose it because he thinks that the woman had more personality.’ The experimental result can be nicely explained by the self-observation hypothesis: Neville needs to observe his own behaviour when reporting why he (allegedly) chose Photo B, just like you do when explaining why he (allegedly) chose it.

Success and Failure of Self-Knowledge

Before moving on, let us consider an issue concerning confabulation studies. It is tempting to think that cases of confabulation involve inaccurate or unsuccessful judgments about one’s own mental states. Participants in the studies discussed above are bad at self-knowledge. Because of this, confabulation studies might threaten not only the idea of peculiar self-knowledge but also the idea of privileged self-knowledge. But is this really the right interpretation of confabulation studies?

Based on the evidence of pervasive confabulation concerning reasons for attitudes and choices, Krista Lawlor (2003) argues that mental-state self-attributions lack authority as they are not as accurate as third-party attributions and fail to correlate with a person’s future behaviour. On the basis that ill-grounded explanations of attitudes and choices are virtually indistinguishable from well-grounded ones and are common, one might argue that we should be genuinely concerned about the reliability of self-knowledge (Scaife 2014, 471). Indeed, some interpret the standard philosophical account of confabulation as an instance of ‘failed mind-reading’ (Strijbos & de Bruin 2015); confabulation shows that people make mistakes when attributing mental states to themselves:

[I]f confabulation turns out to be a widespread phenomenon in everyday social practice, this would seriously undermine first-person authority of mental state attribution. (Strijbos & de Bruin 2015, 298)

Consider the stockings experiment once more. Imagine that participants choose the rightmost pair of stockings because of position effects. When asked to explain their choices, they answer that they chose that pair because of its qualities. As they do not mention the role of the position effect in their choice, their explanation is ill grounded. They confabulate. Not only do they offer an ill-grounded explanation, but, as a result of confabulating, they also form the belief that the pair of stockings they chose is the brightest, and that belief is false. Participants who are asked for an explanation of their choice produce an ill-grounded causal claim due to their ignorance of the mental processes underlying their choices.

Whether the form of non-clinical confabulation we are examining here involves a failure of mental-state self-attribution depends on what we take successful mental-state self-attributions to require (Bortolotti 2018a). In their original paper on priming effects, Nisbett and Wilson are clear that participants’ verbal reports are inaccurate because participants ignore the mental processes leading to their choices and, as a result, misidentify the reasons for their choices. Confabulation is evidence for the view that people are blind to the processes responsible for their choices, but this does not imply that they are also blind to the choices they made. Independently of whether research participants can identify the reasons for their choices, their choices are authentic in the sense that they are sincerely reported and genuinely endorsed. If successful mental-state self-attributions require awareness of one’s attitudes and choices, then they are not threatened by confabulation.

Does successful mental-state self-attribution require that people are aware of the mental processes responsible for their choices? This sounds like an implausibly demanding requirement. In the cases where confabulation has been observed and documented (such as consumer choice, moral judgments, and hiring decisions), causal factors leading to the choice are likely to be psychological processes that involve priming effects, socially conditioned emotional reactions, and implicit biases that cannot be directly experienced or easily observed, but rather need to be inferred on the basis of the systematic and scientific study of human behaviour.

Does successful mental-state self-attribution require that people’s subsequent behaviour is explained and reliably predicted on the basis of that self-attribution? This also sounds like an implausibly demanding requirement and one that imposes more stability and consistency on people’s mental life than is reasonable to expect. We do not know whether people who claim to have chosen a pair of stockings for its texture would choose the softest pair of stockings at their next consumer choice survey, but should they not do so, the fact that mental-state self-attributions fail to shape their future behaviour does not speak so much against self-knowledge as against the crystallization of preference criteria for stockings. This is even more evident in the case of choice blindness.

In the case of the experiment on faces, Johansson and colleagues embrace a preference misattribution account of choice blindness, suggesting that participants are (literally) blind to their choices and misattribute preferences to themselves. The explanation would be that participants who fail to detect the manipulation chose Photo A because they found that face most attractive; when asked why they chose Photo B (after it was switched for Photo A), they misattribute a preference to themselves and come to think that this face is most attractive. However, some have argued that preference change is the best interpretation of the experiment (see Bortolotti & Sullivan-Bissett 2019; Lopes 2014), meaning that participants do not misattribute preferences to themselves when they defend a choice that is different from the one they initially made, but rather change their preference in the course of the study due to the manipulation in the experiment. In this case, participants have an initial preference for Photo A (which explains their choice), but when asked why they chose Photo B, they form a new preference for Photo B (which explains them being able to offer reasons for choosing that photo).

Dominic McIver Lopes (2014) claims that, in the manipulation condition of the experiment, participants give reasons for what has become their new choice. For Lopes, it is no surprise that preferences are changeable and amenable to manipulation. The newly ascribed preferences are real and are newly formed as a result of the participant’s beliefs about the choices they made (Lopes 2014, 29–30). Given that most participants do not detect the manipulation, they come to believe that they chose the face that is presented to them as their choice. Participants do not realize that their preference shifted. In the manipulation condition, reasons are often given in the past tense: ‘I thought she had more personality in a way,’ ‘I chose her because she smiled,’ or ‘I chose her because she had dark hair’ (Johansson et al. 2005, 118). This strongly suggests that people believe that their new preference was there all along, failing to appreciate the preference shift and the role of the manipulation in determining this preference shift. The participant’s belief that the choice presented to them was the one that they had made determines the preferences they ascribe to themselves when they give reasons for the choice.

Perhaps subjects in the manipulation condition changed their preference as a result of their choice. Since they did not notice the manipulation, they believed that they chose the displayed face, and that fact determines their preference. On this hypothesis, […] subjects’ reasons do not accord with their initial preference as revealed by their initial choice, but they do accord with their eventual preference as determined by what they took to be their choice. (Lopes 2014, 29–30)

When asked again about their preferences at a later time (Johansson, Hall, & Sikström. 2008), participants in the manipulated condition chose the face that they provided reasons for choosing in the experiment (i.e., they provided reasons for why they chose the switched photo) and not the face they had originally chosen. According to Lopes, this suggests that reasons cement preferences, meaning that when one gives reasons for a choice, the preference expressed in that choice ‘sticks’ and becomes more stable.

2.4 The Moderate View

So far, we have seen some empirical evidence for self-observation. In some cases, including rather unusual cases (split-brain patients, hypnosis, etc.) and fairly mundane cases (priming effects in consumer choice, choice blindness, etc.), we rely on observation in the same way that we rely on observation to come to know somebody else’s mind. This means that the extreme peculiarity view is hopeless. We are then left with two options: the moderate view and the extreme parity view. In the former case, self-observation is at work in some contexts, such as the ones in the experiments above, but not in others. The peculiarity of self-knowledge is limited, but it is not denied. In the latter case, self-observation is always at work; what we have seen in the experiments above is just the tip of the iceberg. There is nothing peculiar about self-knowledge; self-observation is all there is to self-knowledge. Let us examine the two options in more detail.

The moderate view has been supported by many authors in philosophy and psychology. On the psychology side, for example, Bem defends the idea of self-observation, but adds that self-observation happens ‘to the extent that internal cues are weak, ambiguous, or uninterpretable’ (Bem 1972, 2). On the philosophy side, for example, Byrne defends the idea of peculiar self-knowledge, but adds that ‘there is certainly room for debate about the importance of third person (“behavioral”) access to one’s mental states’ (Byrne 2018, 2).

According to the moderate view, self-observation is at work in some contexts, but not in others. The crucial question for this view is: ‘What are the contexts in which self-observation is at work?’

The answer to this question might be found when we look carefully at the contexts in which people confabulate. Confabulation experiments seem to put participants in rather peculiar contexts. For example, the stockings experiment involves a rather peculiar situation in which the participants are first asked to choose among identical stockings and then they need to respond to the question ‘Why did you choose this pair?’ This situation is unlike our everyday choice situations where, for example, we choose from different stockings on display in a supermarket and nobody asks us ‘Why?’ immediately after the choice.

Let us consider two hypotheses. The first is that self-observation happens in the context in which participants are supposed to choose among identical items. The second hypothesis is that self-observation happens in the context in which participants are supposed to answer the ‘Why?’ question.

Let us begin with the first hypothesis. As Johansson and colleagues have noted, the stockings experiment ‘involved a rather strange and contrived task’ in which experimenters ‘have given [the participants] the artificial choice between identical stockings’ (Johansson et al. 2006, 689). Since all items are the same, perhaps participants do not have a strong preference, or no preference at all, which creates a context in which ‘internal cues are weak’.

However, it is not clear that the choice blindness study can be explained by this hypothesis. After all, participants choose among different items (e.g., Photo A and Photo B) in the experiment. Of course, it is conceivable that participants do not have a strong preference despite their choice among different items. After all, the choice does not have any significant implications for their lives. But not all cases in which confabulation and choice blindness are observed involve something inconsequential such as trivial consumer choices or assessing the attractiveness of strangers’ faces.

For instance, choice blindness has been observed in cases where people were asked about their moral principles or political commitments. In this context, the failure to detect the manipulation causes greater concern than in the context of consumer choice or aesthetic judgments. It is perhaps understandable that participants’ preferences about the attractiveness of strangers’ faces can be easily manipulated, because people may not come to make those choices with existing preferences and may not care much about the attractiveness of faces that they have never seen before. However, preferences about moral and political issues (e.g., ‘It is more important for a society to promote the welfare of its citizens than to protect their personal integrity’) are expected to be more stable, and they may even count as self-defining for those agents who see themselves as engaged in politics (Bortolotti & Sullivan-Bissett 2019).

Participants in the studies by Hall and colleagues (Hall, Johansson, & Strandberg 2012; Hall et al. 2013) were asked to complete what the authors describe as a ‘self-transforming’ questionnaire on either foundational moral principles (condition 1) or topical moral issues (condition 2). This happened just before a general election in Sweden at a time when people who were usually not interested in politics were likely to think about these kinds of issues in their daily lives. Participants had to rate their agreement with a statement using a nine-point scale, and then explain their ratings to the experimenter. The transforming part of the experiment was that two of the statements read out by the experimenter were actually the opposite of the statements originally rated. The ratings given were kept the same, but the statement was reversed, so participants were in effect presented with the opposite of the opinion they expressed earlier. The experiment was designed to see whether participants would be led to endorse a view that was in opposition to the one they had just stated.

Once the participant had read the reversed statement, an experimenter would summarize their view back to them with a question such as ‘So you don’t agree that [statement]?’ or ‘So you do agree that [statement]?’ This mechanism was in place to ensure participants were sure of what they were committing themselves to. The manipulated trials were understood as corrected when participants noticed something strange immediately (spontaneous detection) or claimed that something was amiss only later at the time of debriefing (retrospective correction). Trials were understood as accepted when the participant showed no sign of having noticed that the reversal of the opinion they originally expressed was being fed back to them.

In the first condition, in which the questionnaire was on foundational moral principles, around 33% of the trials were spontaneously detected, and a further 8% were retrospectively detected after the experiment. In condition two, nearly 50% of the manipulations were spontaneously detected, but very few participants claimed to detect the manipulations retrospectively. Framed for individuals, 69% of participants accepted at least one of the two reversed statements (Hall, Johansson, & Strandberg 2012, 4). The manipulation was performed very subtly, so even participants who noticed something strange did not detect the manipulation as such, although they did declare that they must have previously misread or misunderstood the statement.

There was no correlation between self-evaluation of strength of moral conviction and correction, so those ‘participants who believed themselves to hold strong moral opinions in general were no more likely to correct the manipulations’ (Hall, Johansson, & Strandberg 2012, 3). Although there was a positive relationship between level of agreement and spontaneous detection, ‘a full third (31.3%) of all manipulated trials rated at the endpoints of the scale (1 or 9) remained undetected, which shows that not even extreme levels of agreement or disagreement with statements guarantees detection’ (Hall, Johansson, & Strandberg 2012, 3). However, those participants who claimed to be politically active were more likely to spontaneously detect the manipulation in condition two when compared to politically active participants in condition one. From this we learn that those who identified as politically active were less likely to be manipulated into misidentifying their attitudes or were less likely to have their attitudes changed. It would be interesting to examine whether there are any other individual differences that can explain the correction/acceptance rates among participants.

Let us now consider the second hypothesis that self-observation happens in the context in which participants are supposed to answer the ‘Why?’ question. Nisbett and Wilson (1977) said the following in the opening of their classic article:

‘Why do you like him?’ ‘How did you solve this problem?’ ‘Why did you take that job?’ In our daily life we answer many such questions about the cognitive processes underlying our choices, evaluations, judgments and behavior. (Nisbett & Wilson 1977, 231)

Nisbett and Wilson are certainly right: we face many ‘Why?’ questions in daily life. But there are some reasons to think that ‘Why?’ questions (unlike ‘What?’ questions such as ‘What do you think?’ or ‘What do you want?’) create the context in which self-observation is likely. Perhaps ‘Why?’ questions create the context in which ‘internal cues are weak, ambiguous, or uninterpretable’ (Bem 1972, 2). For example, a ‘Why?’ question might be taken as asking the cause of the decision rather than the decision itself. Participants do not have strong internal cues about the cause of their decision and, thus, they have no choice but to appeal to self-interpretation.

Nisbett and Wilson propose a distinction between ‘content’ (i.e., what occupies one’s consciousness at that time) and ‘process’ (i.e., the causal origin of content) (but see also Wilson 2002). We confabulate about the latter, but not about former. In fact, Nisbett and Wilson accept that ‘we do indeed have direct access to a great storehouse of private knowledge’ of the content of consciousness and the content belongs to ‘private facts that can be known with near certainty’ (Nisbett & Wilson 1977, 255).

Alternatively, the ‘Why?’ question might be taken as asking participants to justify their decision. The ‘Why?’ question creates the context in which participants are supposed to justify their decision rather than simply report their psychological states and processes. Under the psychological pressure to justify their decision, participants appeal to self-observation which enables them to explain their own behaviour in terms of beliefs and desires that make their behaviour reasonable or justifiable (Sandis 2015). For example, a participant, Jeremiah, chooses pair D because of the position effect. He interprets the experimenter’s question about his choice as a request for a justification. This is a difficult request; there is no good reason for Jeremiah’s choice as the stockings to choose from were identical (unless we believe that we are in a situation in which the position of an item is a good reason to choose it). Under the psychological pressure to justify his choice, Jeremiah says that pair D has the best texture, which latches onto generally plausible reasons for choosing stockings and other similar items.

Note that this is not to say that Jeremiah’s answer is plausible or unproblematic. Even as a justification for his particular choice of the rightmost pair of stockings within a set of four identical pairs of stockings, his answer is still problematic. A justification is not supposed to disclose the causal processes that led to his choice, but rather to highlight what good reasons there are for that choice, regardless of whether those reasons did motivate him to make the choice (Sandis 2015). The problem is that the reason he provided to justify his choice does not match the features of the situation because all pairs of stockings are the same.

2.5 The Extreme Parity View

Extreme Parity and Parsimony

Now let us turn to the extreme parity view. According to this view, self-observation is always at work. Self-observation is all there is to self-knowledge. Shaun Nichols and Stephen Stich summarize the view:

[The extreme parity view] proposes to maintain the parallel between detecting one’s own mental states and detecting another person’s mental states quite strictly. The only information used as evidence for the inference involved in detecting one’s own mental state is the information provided by perception (in this case, perception of oneself) and by one’s background beliefs (in this case, background beliefs about one’s own environment and previously acquired beliefs about one’s own mental states). (Nichols & Stich 2003, 156)

Note that we do not have direct evidence for the extreme parity view. We only have the evidence for self-interpretation in some limited contexts such as abnormal cases (e.g., split-brain cases) and experimental contexts (e.g., the stockings experiment).

The extreme parity view is counterintuitive. For example, it does not explain those occasions in which we deliberate. The extreme parity view gets something right: people ‘read their own minds’, self-observe, and self-interpret much more frequently than they realize. But by being extreme, the view misses some important nuances. People sometimes deliberate (rarely and not very rationally) and it is overwhelmingly implausible to think otherwise. When making a decision that will have wide-ranging and long-term implications (e.g., which subject to study at university or whether to move to a foreign country), people consider the advantages and disadvantages of each outcome, ask for advice, and generally ‘make up’ their own mind without the need to self-observe.

Despite the lack of direct evidence for it, and despite its being counterintuitive, the extreme parity view can be defended by an indirect argument; we call this the ‘parsimony argument’ for the extreme parity. The extreme parity view gives a parsimonious explanation of the available data. The extreme parity view explains all cases of self-knowledge in a simple and unified manner in terms of self-observation. We do not need to posit additional peculiar methods for gaining self-knowledge. As Carruthers has noted, ‘In order to warrant the extra complexity, it needs to be shown that [the extreme parity view] on its own is inadequate, or else some positive evidence of an additional method should be provided’ (Carruthers 2011, 366).

Parsimony arguments are often used in philosophy (and in science). If we have two theories, theory A and theory B, which are equally good at explaining the available data, and if A is more complicated than B, then we ought to choose B since it is more parsimonious. So, if we have two hypotheses about self-knowledge, the moderate view and the extreme parity view, which are equally good at explaining the available data, and if the moderate view is more complicated than the extreme parity view, then we ought to choose the extreme parity view since it is more parsimonious.

The Parsimony Argument

• Premise 1: The moderate view and the extreme parity view are equally good at explaining the available data.
• Premise 2: The moderate view is more complicated than the extreme parity view.
• Premise 3: If the moderate view and the extreme parity view are equally good at explaining the available data, and if the former is more complicated than the latter, then we ought to choose the extreme parity view.
• Conclusion: We ought to choose the extreme parity view.

Should we accept the parsimony argument in this case? Is the extreme parity view really defendable?

Our focus will be on premise 1, namely that the moderate view and the extreme parity view are equally good at explaining the available data. Can this premise be challenged? Challenging premise 1 involves providing some cases that the extreme parity view cannot explain but that the moderate view can. Are there such cases?

The Interpretive Sensory-Access Theory

Nichols and Stich point out that ‘[i]t seems obvious that people can sit quietly without exhibiting any relevant behavior and report on their current thoughts’ (Nichols & Stich 2003, 157). For example, it is possible that a person, Joan, knows that she believes that there is an apple on the table, or that she wants to eat an apple, while sitting still in the living room. This case seems to pose a serious problem for the extreme parity view (but not for the moderate view, which allows for the possibility of peculiar and non-observational access). Being unable to explain such an obvious fact, Nichols and Stich say that the extreme parity view is ‘crazy’ and ‘is hard to take seriously’ (Nichols & Stich 2003, 156).

Carruthers notes that the case can be explained by a sophisticated version of the extreme parity view where materials for self-interpretation go beyond the observation of bodily behaviour. According to Carruthers’ (2009) interpretive sensory-access (ISA) theory, the materials for self-interpretation include both external cues (including own behaviour and environment) and inner sensory cues (including inner speech and mental imagery). The ISA theory can deal with the case of immobile Joan: although there is no informative external cue in the case, Joan can access her internal sensory cues (including inner speech and mental imagery) that help her to learn about her beliefs, desires, and so on. Joan can attribute to herself the belief that there is an apple on the table on the basis of her inner speech sentence ‘An apple is on the table’ or mental imagery of an apple on the table.

There is a sense in which the ISA theory is not that extreme; after all, the theory permits peculiar and non-interpretive access to internal cues (such as inner speech and mental imagery). That said, the ISA theory is extreme in that it denies any peculiar access to propositional attitudes (such as beliefs or desires). According to the ISA theory, metacognitive access to propositional attitudes is always self-interpretive, where self-interpretation is based not only on external cues but also on internal sensory cues.

Note that although the ISA theory recognizes the peculiar access to internal sensory cues, it is still more parsimonious than the moderate view. Any sensible account of metacognition, including the moderate view, would recognize the peculiar access to inner speech, mental imagery, and so on. Thus, the ISA theory, which recognizes the peculiar access to internal sensory cues but does not recognize the peculiar access to propositional attitudes, is more parsimonious than the moderate view, which recognizes peculiar access to propositional attitudes in addition to peculiar access to internal sensory cues. This means that the parsimony argument is still applicable.

There are, however, some worries about the ISA theory. It requires at least two theoretical assumptions: a psychological assumption and a metaphysical assumption. The psychological assumption is that internal sensory cues regularly accompany propositional attitudes. For example, the inner speech sentence ‘An apple is on the table’ regularly accompanies the belief that an apple is on the table. Otherwise (i.e., if there is no such a regularity), the theory fails to explain the case of immobile Joan. If Joan’s belief that an apple is on the table is not regularly accompanied by the inner speech sentence ‘An apple is on the table’ (or other internal sensory cues), then she is not able to identify her belief. The metaphysical assumption is that internal sensory cues do not constitute propositional attitudes. For example, the inner speech sentence ‘An apple is on the table’ does not constitute the belief that an apple is on the table, although the former reliably accompanies the latter. Otherwise (i.e., the inner speech sentence does constitute the belief), admitting peculiar access to the inner speech sentence amounts to admitting peculiar access to the belief, which contradicts the ISA theory’s denial of peculiar access to propositional attitudes.

One might worry about the compatibility of the psychological and metaphysical assumptions. If internal sensory cues regularly accompany propositional attitudes, then one might think that there must be an explanation of the regular co-occurrence of sensory cues and propositional attitudes. A simple explanation of the regular co-occurrence is that internal sensory cues are constitutive of propositional attitudes.

Setting this worry aside, another problem is that the ISA theory might fail to explain the case of immobile Joan after all because her internal cues are not informative enough for her to identify her propositional attitudes. Even though Joan has access to her internal sensory cues (such as the inner speech sentence ‘An apple is on the table’ or the mental image of an apple on the table), these sensory cues are not sufficient for her to identify her belief that there is an apple on the table.

For example, having the mental image of an X in your mind does not specify whether you are thinking about X or Y when you cannot discriminate X from Y in your visual imagination. Suppose that Joan recognizes a mental image of a geometrical shape with lots of angles. Suppose that she also has two hypotheses: that she is believing something about a chiliagon (i.e., a polygon with 1,000 sides) and that she is believing something about a myriagon (i.e., a polygon with 10,000 sides). Can she tell which one is correct? This is really difficult because distinguishing chiliagons from myriagons in one’s visual imagination is extremely difficult. As Descartes notes, the thing that I visually imagine when thinking about a chiliagon ‘differs in no way from the representation I should form if I were thinking of a myriagon, or any figure with very many sides’ (Descartes 1984, 50).

One might think, however, that this problem can be solved by inner speech. For example, Joan can conclude that she is believing something about a chiliagon when she recognizes the inner speech word ‘chiliagon’ or that she is believing something about a myriagon when she recognizes the inner speech word ‘myriagon’. But inner speech has its own problems too.

Having an inner speech sentence P in your mind does not specify whether you are believing that P, desiring that P, imagining that P, supposing that P, and so on. For example, what is Joan supposed to know about her propositional attitudes when she recognizes the inner speech sentence ‘Caesar died in his bed’? Does she believe that Caesar died in his bed, or does she imagine that Caesar died in his bed? As Hume points out, sensory cues (or ‘ideas’) are the same between somebody imagining that Caesar died in his bed and somebody believing that Caesar died in his bed. The believer ‘form[s] all the same ideas as [the imaginer] does’ and the believer ‘can’t conceive any idea that [the imaginer] can’t conceive, or conjoin any ideas that [the imaginer] can’t conjoin’ (Hume 1739/2007, 66).

To sum up, it is not clear that the ISA theory can successfully explain the case of immobile Joan; internal sensory cues might not be informative enough for knowing about one’s own propositional attitudes.

2.6 Summary

• Empirical data, especially those concerning confabulation, suggest that agents rely on self-observation to identify their own mental states – at least in some cases. This is consistent with what we called the moderate view, according to which humans know their own minds by self-observation in some cases but not always.
• The view that our self-knowledge always relies on self-observation and self-interpretation (what we called the extreme parity view) is certainly attractive because of its simplicity, but it faces the difficult challenge of explaining the case of immobile Joan. It is also not clear that the ISA theory can meet this challenge.
• According to the moderate view, our self-knowledge relies on self-observation and self-interpretation in some cases. This suggests that there are parts or domains of the mind to which people do have peculiar and direct access and parts or domains to which people do not have peculiar and direct access. The best thing to do if one wants to know about the parts or domains of the latter kind is to observe and interpret one’s own behaviour or learn from third-person testimony. This picture is compatible with the idea that the mind is fragmented, which will be discussed in Chapter 3.

Further Resources

Articles and Books

The stockings experiment discussed in this chapter is included in Richard Nisbett and Tim Wilson’s article ‘Telling more than we can know: Verbal reports on mental processes’ (1977), which is considered to be one of the most influential psychology papers of the 20th century. Wilson’s book Strangers to Ourselves: Discovering the Adaptive Unconscious (2002) is an excellent overview of salient psychological studies on the limitations of introspection, and it offers some interesting philosophical reflections on the implications of those studies.

For a thorough investigation of the phenomenon of confabulation, we recommend a book by William Hirstein, Brain Fiction: Self-Deception and the Riddle of Confabulation (2005), where clinical and non-clinical instances of confabulation are examined. For more information about cognitive dissonance and the self, see Elliott Aronson’s useful article (2019) that discusses the notion of hypocrisy.

For an empirically informed and philosophical discussion of self-knowledge, we recommend Peter Carruthers’ work, especially the target article (2009) in Behavioral and Brain Sciences and the book The Opacity of Mind: An Integrative Theory of Self-Knowledge (2011). Some papers that discuss the role of the empirical evidence on confabulation in the self-knowledge literature include work by Robin Scaife (2014) and Derek Strijbos and Leon de Bruin (2015). Other important contributions by philosophers include: Shaun Nichols and Stephen Stich’s Mindreading: An Integrated Account of Pretence, Self-Awareness and Understanding Other Minds (2003), Quassim Cassam’s Self-Knowledge for Humans (2014), and Annalisa Coliva’s The Varieties of Self-Knowledge (2016).

Online Resources

Brie Gertler’s Stanford Encyclopedia of Philosophy entry on self-knowledge (2020) has some useful discussion of the implications of the empirical evidence on theories of self-knowledge. Eric Schwitzgebel’s Stanford Encyclopedia of Philosophy entry on introspection (2019) is good background reading on the idea of self-knowledge as privileged and peculiar.

Keith Frankish’s article on the limitations of self-knowledge (2016) in Aeon and Joelle Proust’s article on metacognition (2016) in the Brains blog are both accessible and engaging. Lisa Bortolotti also wrote an accessible article for Aeon on everyday instances of confabulation (2018b), where the stockings experiment is mentioned.

Mitchell Green’s introductory lectures on the value and limits of self-knowledge (2019) are available at University of Edinburgh online courses.

Petter Johansson’s TED talk (2018) is an excellent introduction to the phenomenon of choice blindness.

Questions

1. Do you know yourself better than your family members or friends do?
2. If you have any special access to information about yourself, does this apply only to the content of your feelings and thoughts or does it extend to your character traits too?
3. In what way, if any, does the literature on confabulation undermine self-knowledge claims?
4. In what context does it make more sense to hypothesize a type of self-knowledge that is not limited to self-observation and interpretation?

3.1 Introduction

Our discussion so far seems to suggest that the mind is fragmented or divided rather than united. For instance, the fact that many people are vulnerable to reasoning biases, which we discussed in Chapter 1, suggests that their minds are fragmented. Recall the Linda experiment, where most research participants considered the conjunction ‘Linda is a bank teller and is active in the feminist movement’ more probable than ‘Linda is a bank teller’ because the description of Linda matched that of a feminist. Participants committed the conjunction fallacy because they chose the option that they felt was more ‘representative’ of Linda. Stephen Jay Gould describes his own experience with the Linda experiment as follows:

I know that [the feminist bank teller hypothesis] is least probable, yet a little homunculus in my head continues to jump up and down, shouting at me – ‘but she can’t just be a bank teller; read the description’. (Gould 1991, 469)

Gould says that he knows that the feminist bank teller hypothesis is mathematically less probable than the bank teller hypothesis because a conjunction of two events can never be more probable than one of the conjuncts, but while part of him is convinced that the bank teller hypothesis is more probable, another part of him still leans towards the feminist bank teller hypothesis. It is as if Gould’s mind is, by his own admission, fragmented.

The phenomenon of self-observation, discussed in Chapter 2, lends support to the idea of the fragmented mind as well. To self-observe is to treat one’s own mind as if it is somebody else’s. You learn about your own mind by observing and interpreting your own behaviour just like you do when you learn about somebody else’s mind by observing and interpreting her behaviour. This means that, in a sense, we are strangers to ourselves. This is also the title of Timothy Wilson’s (2002) book, which provides an overview of the significance of the studies on the limitations of introspection.

The very idea of self-observation suggests that the part of a person’s mind that is observing and interpreting is distinct from, and maybe even alienated from, the part of the mind that is being observed and interpreted. As we mentioned in Chapter 2, the most extreme example of this fragmentation or alienation is in the context of confabulation by split-brain patients (Gazzaniga 2000), where the brain is literally divided into two (i.e., the right hemisphere and the left hemisphere become physiologically alienated from one another). But the idea of a fragmented mind can be generalized to other, less extreme forms of self-observation too. According to Wilson, human agents are alienated from an important part of themselves – their ‘adaptive unconscious’.

Many human judgments, emotions, thoughts, and behaviors are produced by the adaptive unconscious. Because people do not have conscious access to the adaptive unconscious, their conscious selves confabulate reasons for why they responded the way they did. (Wilson 2002, 105)

We believe that the idea of a fragmented mind is plausible as it makes sense of some phenomena we have already described and other phenomena we will describe in the following chapters. This chapter will clarify the idea of a fragmented mind in light of the empirical evidence and various theoretical considerations. In particular, this chapter will address the literature on the dual-process theory (Section 3.2), which is an informative account of mental fragmentation. The dual process theory states that human cognition is the result of the interaction between two kinds of processes: Type-1 processes and Type-2 processes. This theory provides us with a plausible solution for the so-called ‘rationality paradox’, which concerns how to reconcile human apparent rationality with human apparent irrationality. There are, however, some issues with the dual process theory. The following questions arise: ‘What are Type-1/Type-2 processes?’, ‘What distinguishes Type-1 processes from Type-2 processes?’, and ‘How do Type-1 and Type-2 processes interact? Do they compete with each other all the time? Or are Type-1/Type-2 processes dominant most of the time?’ (Section 3.3).

This chapter also addresses whether the dual-process theory implies the dual-system theory (i.e., Type-1/Type-2 processes correspond to two distinct cognitive systems) (Section 3.4) and the dual-state theory (i.e., Type-1/Type-2 processes correspond to two distinct mental states or attitudes) (Section 3.5).

3.2 The Dual-Process Theory

Basic Ideas

What does it mean to say that the mind is fragmented or divided rather than united? By ‘fragmented’ we do not just mean that the mind has many parts, constituents, or systems. The crucial part of what we call ‘fragmentation’ is the idea that different parts of the mind can pull us in different directions, or more specifically that there can be some kind of conflict between different parts of the mind. Plato’s doctrine of the tripartite structure of the soul is a classic expression of a fragmented mind. According to Plato, the soul is composed of three parts: appetite, spirit, and reason. Crucially, Plato does not only say that the soul is divided into parts, he also says that different parts can be in conflict with one another, pulling the soul in different directions. Plato explains this idea in the famous Chariot Allegory:

Remember how we divided each soul in three at the beginning of our story – two parts in the form of horses and the third in that of a charioteer? […] The horse that is on the right, or nobler, side is upright in frame and well jointed, with a high neck and a regal nose; his coat is white, his eyes are black, and he is a lover of honor with modesty and self-control; companion to true glory, he needs no whip, and is guided by verbal commands alone. The other horse is a crooked great jumble of limbs with a short bull-neck, a pug nose, black skin, and bloodshot white eyes; companion to wild boasts and indecency, he is shaggy around the ears – deaf as a post – and just barely yields to horsewhip and goad combined. (Plato 1995, 43–44)

The idea of a fragmented mind has been expressed in many ways by different authors in philosophy and psychology (see Frankish & Evans 2009), and several accounts describe a sort of duality in the human mind (see Box 3A for three versions of duality).

BOX 3A: Theories of Duality in Human Cognition

Dual process:

‘Dual-process theories hold that there are two distinct processing modes available for many cognitive tasks: one (type 1) that is fast, automatic and non-conscious, and another (type 2) that is slow, controlled and conscious.’ (Frankish 2010, 914)

Dual system:

‘Human cognition is composed of two multipurpose reasoning systems, widely known as System 1 and System 2, the former supporting type 1 processes, the latter supporting type 2 ones.’ (Frankish 2010, 914)

Dual attitude (or dual state):

‘Dual attitudes are defined as different evaluations of the same attitude object: an automatic, implicit attitude and an explicit attitude.’ (Wilson, Lindsey, & Schooler 2000, 101)

Currently, the best expression of the idea of mental fragmentation can be found in the dual-process theory (Evans 2003, 2010; Evans & Over 1996; Kahneman 2011; Sloman 1996; Stanovich 1999, 2011; Stanovich & West 2000). The dual-process theory has been applied to different domains, including reasoning, learning, and social cognition (for an overview, see Evans 2008). There are also dual-process theories of other psychological phenomena, such as moral judgments (Chapter 4) or mindreading (Chapter 6).

The focus of this chapter is the dual-process theory of reasoning according to which reasoning performance is determined by the interaction between two reasoning processes of different kinds: Type-1 processes and Type-2 processes. Typically, Type-1 processes are evolutionarily older, unconscious, automatic, low effort, and associative, whereas Type-2 processes are evolutionary more recent, conscious, controlled, high effort, and rule-based.

Just like the two horses in Plato’s Chariot Allegory, Type-1 processes and Type-2 processes can pull us in different directions. In the Linda experiment, for example, Type-2 processes should follow the mathematically canonical procedure and reach the conclusion that the bank teller hypothesis is more probable than the feminist bank teller hypothesis: ‘I know that [the feminist bank teller hypothesis] is least probable.’ In contrast, Type-1 processes follow a heuristic procedure driven by judgments of representativeness and reach the conclusion that the feminist bank teller hypothesis is more probable than the bank teller hypothesis: ‘[A] little homunculus in my head continues to jump up and down, shouting at me – “but she can’t just be a bank teller; read the description”’ (Gould 1991, 469).

We will now discuss two issues concerning the dual-process theory. In the rest of this section, we focus on the reasons for accepting the dual-process theory. Next (in Section 3.3), we focus on the details of the theory.

Empirical Reasons: Reasoning Biases

There are both empirical and philosophical reasons for favouring the dual-process theory (or at least supporting the basic idea of it). We will start with the empirical reasons and then turn to the philosophical reasons.

A clarification is needed with regard to the empirical reasons for favouring the dual-process theory. It would not be accurate to assume that the dual-process theory is an empirical theory in psychology that is directly testable via experiments. Instead, it is one of general hypotheses in psychology that can make sense of a series of relevant studies as interconnected, just like the massive modularity hypothesis, the language of thought hypothesis, and the adaptive unconsciousness hypothesis. In fact, Jonathan Evans and Keith Stanovich argue that the dual-process theory is a ‘broad framework’ or ‘metatheory’; it ‘cannot be falsified by the failure of any specific instantiation or experimental finding’ and its central role is to ‘stimulate new research and accumulate enough supportive evidence’ (Evans & Stanovich 2013b, 263). The dual-process theory has been successful in making sense of relevant empirical studies as interconnected. In Thinking, Fast and Slow, for example, Daniel Kahneman (2011) interprets the entire heuristics and biases programme in light of the dual-process theory.

Wason Selection Task

Having said that, the dual-process theory has been associated with some particular empirical findings. For example, the dual-process theory of reasoning was originally introduced in an interpretation of the Wason selection task experiments, which we saw in Chapter 1. In one experiment (Wason & Evans 1974), participants were asked to select cards that need to be turned over to verify the given hypothesis and to explain the reason for selecting the cards. Participants’ actual choices were strongly influenced by a primitive matching bias (selecting cards depicting the items that are explicitly mentioned in the hypothesis), but the reasons offered by participants have little to do with their matching bias. Rather, their reasons were apparently logical, post hoc rationalizations. (This reminds us of the stockings experiment by Nisbett and Wilson [1977] discussed in Chapter 2, where participants’ choices were strongly influenced by the position effect, of which they were not aware. When asked to justify their choice, they offered a post hoc rationalization that had nothing to do with the position effect.)

To explain the result of the Wason selection task experiments, Wason and Evans propose a dual-process account, which posits two distinct processes:

1. The processes underlying reasoning performance (e.g., matching bias) are not generally available for introspection.
2. Introspective accounts of reasoning performance reflect one’s tendency to construct a justification for one’s own behaviour consistent with one’s knowledge of the situation.

Belief Bias

The dual-process theory of reasoning has been associated with the belief bias, which is the tendency of one’s beliefs to influence one’s judgment of logical validity such that inferences with believable (i.e., plausible given one’s background beliefs) conclusions are more likely to be regarded as valid than ones with unbelievable conclusions.

An inference is logically valid if its conclusion is guaranteed to be true when all of its premises are true. In an experiment (Evans, Barston, & Pollard 1983) on logical validity, participants were asked to evaluate four types of inferences.

• Valid-Believable: A valid inference with a believable conclusion. For example, ‘No cigarettes are inexpensive. Some addictive things are inexpensive. Therefore, some addictive things are not cigarettes.’
• Valid-Unbelievable: A valid inference with an unbelievable conclusion. For example, ‘No addictive things are inexpensive. Some cigarettes are inexpensive. Therefore, some cigarettes are not addictive.’
• Invalid-Believable: An invalid inference with a believable conclusion. For example, ‘No addictive things are inexpensive. Some cigarettes are inexpensive. Therefore, some addictive things are not cigarettes.’
• Invalid-Unbelievable: An invalid inference with an unbelievable conclusion. For example, ‘No cigarettes are inexpensive. Some addictive things are inexpensive. Therefore, some cigarettes are not addictive.’

The result showed a significant belief bias: inferences with believable conclusions are more likely to be regarded as valid than those with unbelievable conclusions (92% of participants took Valid-Believable inferences to be valid, while only 46% took Valid-Unbelievable inferences to be valid). But there is also another tendency: valid inferences are more likely to be regarded as valid than invalid ones (46% of participants took Valid-Unbelievable inferences to be valid, while only 8% took Invalid-Unbelievable inferences to be valid). Evans and colleagues interpret these results in terms of the dual-process theoretical idea that these two tendencies reflect an interplay between two kinds of processes: a non-logical process responsible for the belief bias and a logical process responsible for sensitivity to logical structure.

Philosophical Reasons: The Rationality Paradox

We will now move on to philosophical reasons for favouring the dual-process theory. The main theoretical consideration concerns what is known as ‘the rationality paradox’, which can be summarized as a simple question: ‘We can put a man on the moon, so why can’t we solve those logical reasoning problems?’ (O’Brien 1998, 23).

As we have seen, human agents deviate from the rational norms of reasoning: for example, people systematically judged that it was more probable that Linda was a feminist bank teller than a bank teller. But this is paradoxical. If people are so irrational that they make frequent mistakes in basic reasoning problems, like in the Linda experiment, then how did humans achieve remarkable things in science, technology, culture, and other areas? How can creatures who systematically fail to solve the Linda experiment and other basic problems of reasoning solve the much more difficult and complicated logical and mathematical problems that are required, for example, in the project of sending humans to the Moon and bringing them back to Earth safely? Note that the answer to this problem cannot (just) be about individual differences. For example, the answer cannot just be that NASA’s employees are super smart or especially immune to biases and errors. The real paradox is that one and the same person can be rational in some contexts and irrational in other contexts. For example, a science professor who can easily solve complicated maths problems might fail in the Linda experiment.

A special case of the rationality paradox concerns the fact that the remarkable achievements by humans include scientific achievements, which, in turn, include psychological achievements such as the findings of human reasoning errors in the heuristics and biases programme. But how can creatures who systematically fail in the Linda experiment and other basic problems of reasoning successfully solve logical and mathematical tasks that are required for designing and conducting successful psychological studies on reasoning and reveal reasoning errors such as the conjunction fallacy? This version of the rationality paradox can be put in the form of a self-defeating paradox, which we call the ‘self-defeating rationality paradox’.

Suppose that the empirical evidence supports the sceptical view that humans systematically make errors in some basic reasoning problems, such as the Linda problem. However, an implication of this is that humans are not really capable of doing psychology; after all, doing psychology requires the cognitive capacity for solving logical and mathematical tasks that are much more difficult than those required in the Linda experiment. But if humans are not capable of doing psychology, then we should not trust the findings of psychology, including those about reasoning biases. This means that we lose the empirical evidence for the sceptical view of reasoning. The sceptical view defeats itself !

The Self-Defeating Rationality Paradox

• Premise 1: The experimental psychology of reasoning reveals that human reasoning capacity is not sufficiently reliable (see Chapter 1).
• Premise 2: If human reasoning capacity is not sufficiently reliable, then we should not trust the product of human reasoning capacity, including scientific findings and theories, and in turn the experimental psychology of reasoning.
• Conclusion: Thus, the experimental psychology of reasoning is self-defeating.

The rationality paradox (and its special case: the self-defeating rationality paradox) seems to suggest that, on the one hand, humans systematically fail to solve even basic reasoning tasks in some contexts (such as in the heuristics and biases experiments), and, on the other hand, they systematically succeed in solving difficult ones in other contexts (such as when calculating the orbit of a space rocket or analysing experimental data in a psychological study).

The dual-process theory helps us to make sense of this rather puzzling situation. According to the dual-process theory, human cognitive performance is determined by the complex interplay between Type-1 and Type-2 processes. Type-1 processes are dominant in some contexts, while Type-2 processes are dominant in other contexts. Systematic reasoning failures are typically seen in Type-1-dominant contexts, while systematic reasoning successes are typically seen in Type-2-dominant contexts. (However, we should be sceptical about the simplistic idea that people always make mistakes in Type-1-dominant contexts, or people do not make any mistakes in Type-2-dominant contexts. For instance, people might appeal to heuristic processes in Type-1-dominant contexts, but heuristics work perfectly fine in many cases: ‘In general, these heuristics are quite useful’ [Tversky & Kahneman 1974, 1124].)

With the dual-process theory, we can also respond to the self-defeating rationality paradox. Premise 1 can be true only when it is understood as a claim about Type-1-dominant contexts. Thus, the argument should look like this:

• Premise 1: The experimental psychology of reasoning reveals that human reasoning capacity in Type-1-dominant contexts is not sufficiently reliable.
• Premise 2: If human reasoning capacity in Type-1-dominant contexts is not sufficiently reliable, then we should not trust the product of human reasoning capacity, including scientific findings and theories, and in turn the experimental psychology of reasoning.
• Conclusion: Thus, the experimental psychology of reasoning is self-defeating.

But now we see what’s wrong with this argument. In particular, Premise 2 is false because even if human reasoning capacity in Type-1-dominant contexts is not sufficiently reliable, human reasoning capacity in Type-2-dominant contexts might be sufficiently reliable, in which case we might trust the product of human reasoning capacity, including scientific findings and theories, in a Type-2-dominant context.

3.3 Processes and Interactions

Type-1 and Type-2 Processes

Let us now move on to consider the details of the dual-process theory. So far, we have not offered a detailed characterization of it; we have only said that reasoning performance is determined by the interaction between Type-1 and Type-2 processes. There are at least two crucial questions to be asked:

• Process: What are ‘Type-1 processes’ and ‘Type-2 processes’?
• Interaction: How do Type-1 processes and Type-2 processes interact?

Let us begin with the process question. Type-1 processes and Type-2 processes are distinguished by their properties. It is assumed that there are two clusters of properties: the cluster of Type-1 properties and the cluster of Type-2 properties. Type-1 processes instantiate Type-1 properties and Type-2 processes instantiate Type-2 properties.

We should not expect that Type-1/Type-2 properties in Table 1 are always co-instantiated and that Type-1/Type-2 processes are the ones with all of the Type-1/Type-2 properties. In fact, dual-process theorists are not usually committed to such a strong claim: ‘The main misuse of such tables is to treat them as strong statements about necessarily co-occurring features – in short to aid in the creation of a straw man’ (Stanovich & Toplak 2012, 5).

Table 1. Type-1 and Type-2 properties (Evans 2008)

Type-1 cluster properties	Type-2 cluster properties
Unconscious or preconscious	Conscious
Implicit	Explicit
Low effort	High effort
Automatic	Controlled
Rapid	Slow
High capacity	Low capacity
Default process	Inhibitory
Holistic and perceptual	Analytic and reflective
Evolutionarily old	Evolutionarily recent
About evolutionary rationality	About individual rationality
Shared with animals	Uniquely human
Non-verbal	Verbal
About modular cognition	About fluid intelligence
Associative	Rule-based
Domain-specific	Domain-general
Contextualized	Abstract
Pragmatic	Logical
Parallel	Sequential
Stereotypical	Egalitarian
Universal	Heritable
Independent of general intelligence	Linked to general intelligence
Independent of working memory	Limited by working memory capacity

Some propose to distinguish essential and accidental items in the list. Essential properties are what define Type-1/Type-2 processes and accidental properties are statistically but not necessarily correlated with essential properties. Evans and Stanovich argue that essential Type-1 properties are the property of not requiring working memory and the property of being autonomous; essential Type-2 properties are the property of requiring working memory and the property of involving ‘cognitive decoupling’, which is the ‘ability to distinguish supposition from belief and to aid rational choices by running thought experiments’ (Evans & Stanovich 2013a, 236).

Type-1 properties and Type-2 properties form statistically (but not necessarily) correlated property clusters: Type-1 clusters and Type-2 clusters. An intriguing suggestion by Richard Samuels (2009) is that Type-1/Type-2 clusters constitute natural kinds, or ‘homeostatic property clusters’ (Boyd 1991). In other words, Type-1/Type-2 properties are statistically (but not necessarily) co-instantiated, and this tendency to co-instantiate is not an accident; it is due to underlying mechanisms.

Samuels proposes the following analogy. Influenza is associated with its characteristic symptoms, such as coughing and fever, that tend to be (but are not always) co-instantiated, and any co-instantiation can be explained by an underlying causal mechanism, namely the presence of the flu virus. Similarly, Type-1/Type-2 properties tend to be (but are not always) co-instantiated, and any co-instantiation can be explained by an underlying causal mechanism. The causal mechanism is ‘System 1/System 2’, which will be discussed in Section 3.4.

Interaction between Type-1 and Type-2 Processes

Let us move to the interaction between Type-1 and Type-2 processes. One might think that Type-1 and Type-2 processes operate in a parallel manner and that they compete with one another for the final output. In the Linda experiment, for example, Type-1 processes are operating and eventually generate the answer that the feminist bank teller hypothesis is more probable than the bank teller hypothesis. Type-2 processes are also operating and eventually generate the answer that the bank teller hypothesis is more probable than the feminist bank teller hypothesis. And the final output (the person’s answer to the question) is determined as the result of the competition between these two kinds of processes. This model of interaction between Type-1 and Type-2 processes is known as ‘the parallel-competitive model’ (e.g., Sloman 1996). This model implies that the reasoning mistake in the Linda experiment is due to the ‘weakness’ of Type-2 processes; they were defeated in the competition by Type-1 processes.

The parallel-competitive model states that Type-1 and Type-2 processes operate in parallel. Alternatively, we might think that Type-1 and Type-2 processes operate in a sequential manner (i.e., Type-2 processes come after Type-1 processes). More precisely, we might think that Type-1 processes are the default processes and Type-2 processes intervene in Type-1 processes only when necessary. This model of interaction is known as ‘the default-intervention model’ (e.g., Evans & Stanovich 2013a). This model implies that participants’ reasoning mistakes in the Linda experiment were due to the ‘laziness’ (Kahneman 2011, 32), rather than the ‘weakness’, of Type-2 processes; Type-2 processes simply fail to intervene in Type-1 processes.

The default-intervention model states that Type-2 processes are ‘lazy’; Type-1 processes are the default position and Type-2 processes come into play only when necessary. That said, Type-2 processes at least have the capacity to intervene in Type-1 processes. But we can think of another model that denies that Type-2 processes have this capacity. Type-2 processes are not ‘lazy’, they are ‘incompetent’. Type-2 processes are not able to intervene in Type-1 processes; at best they can rationalize the output from Type-1 processes in a post hoc manner. Type-2 processes are similar to lawyers rather than police officers. Unlike police officers, lawyers are not able to intervene in a defendant’s behaviour. Their job is rather to rationalize and justify the defendant’s behaviour in a post hoc manner. Let us call this ‘the lawyer–client model’ (Haidt 2001). This model implies that participants’ reasoning mistakes in the Linda experiment are due to the ‘incompetence’, rather than the ‘laziness’ (in the case of the default-intervention model) or ‘weakness’ (in the case of the parallel-competitive model), of Type-2 processes; they are not able to intervene in Type-1 processes in the first place.

Despite sounding radical, the lawyer–client model is plausible to some extent. For example, it is consistent with the selection task experiment discussed in Chapters 1 and 3 (Wason & Evans 1974). Participants’ performance is largely determined by unconscious Type-1 processes, which exhibit the primitive matching bias. In contrast, Type-2 processes are not vulnerable to such a bias, but they are not capable of intervening in Type-1 processes in the first place. Type-2 processes just rationalize the performance of Type-1 processes in a post hoc manner. A similar analysis is applicable to the stockings experiment discussed in Chapter 2 (Nisbett & Wilson 1977). In this experiment, choice is largely determined by unconscious Type-1 processes, which are strongly influenced by the position effect. In contrast, Type-2 processes are not vulnerable to such an effect, but they are not capable of intervening in Type-1 processes in the first place. Type-2 processes just rationalize the Type-1 choice in a post hoc manner.

Which model is correct? This is not an easy question to answer, but a conciliatory answer is available: each model captures some of the interactions between Type-1 and Type-2 processes. In fact, it is far from obvious that we need to choose only one model and reject the alternatives. The idea is that in many cases, Type-2 processes just ‘wait and see’ (which is consistent with the default-intervention model). Even when they are operating, their main job is to rationalize what Type-1 processes do (which is consistent with the lawyer–client model). But sometimes Type-2 processes do intervene, which creates a real competition between Type-1 and Type-2 processes (which is consistent with the parallel-competitive model).

We will come back to this issue in the discussion of moral judgment in Chapter 4.

3.4 The Dual-System Theory

Exactly Two Systems?

We now turn to more controversial ideas that are often associated with the dual-process theory (see Box 3A). The first idea, known as ‘the dual-system theory’, extends the dual-process theory ‘by making specific proposals about the underlying cognitive mechanisms responsible for type-1 and 2 processing’ (Evans 2012, 125). Type-1 and Type-2 processes are attributed to ‘two distinct cognitive systems, with different structures, functions and evolutionary histories’ and thus ‘we have, in effect, two minds’ (Frankish 2010, 919).

Gould talks about the internal disagreement within him, between a part of him, a ‘little homunculus’, who says that Linda ‘can’t just be a bank teller; read the description’, and another part of him, ‘I’, who ‘knows that [the feminist bank teller hypothesis] is least probable’. The dual-system theory takes this description rather seriously and interprets it as a disagreement between two distinct systems: System 1 and System 2.

Is the dual-system theory plausible? Should we really attribute Type-1 processes and Type-2 processes to two distinct systems? It depends on what we mean by ‘systems’. Different authors use the term ‘system’ in different ways. In the early stage of the development of dual-process theory, authors did not draw a sharp theoretical distinction between the ‘process’ talk and the ‘system’ talk, often going back and forth between them relatively freely (e.g., Stanovich 1999). This suggests that, for these authors, the distinction between the ‘process’ talk and the ‘system’ talk is more or less a kind of terminological choice, or perhaps the two are different ways of expressing the same thing. The dual-system theory, in this deflationary terminology, is not very interesting; it simply collapses into the dual-process theory.

Recently, however, authors have tended to be more careful about the distinction between the ‘process’ talk and the ‘system’ talk, and are often in favour of the former. For example, Keith Stanovich and Maggie Toplak avoid the ‘system’ talk because it tends to be associated with some unwarranted assumptions: ‘the two processes in dual-process theory map explicitly to two distinct brain systems’ and ‘what is being referred to [by the term “System 1” or “System 2”] is a singular system’ (Stanovich & Toplak 2012, 4). For Stanovich and Toplak, if the dual-system theory comes with these assumptions, then it is not defensible. Evans joins Stanovich and Toplak and avoids the ‘system’ talk because ‘[a] cognitive or neural system, in a well-defined sense, should be described with some fairly specific assumptions about its inputs, outputs and functions’, and ‘[i]t has been evident for some time that what people have been calling System 1 is in fact a multitude of automatic systems’ (Evans 2011, 90).

These authors avoid the ‘system’ talk because it implies the implausible idea that there are exactly two cognitive systems in the human brain. This is what Samuels calls the ‘token thesis’ according to which ‘each mind contains two particular cognitive mechanisms or systems […] each human mind exhibits a fundamental, bipartite division into particular systems’ (Samuels 2009, 133).

The token thesis is implausible and the dual-process/system theorists explicitly deny such an idea. For example, according to Stanovich (2004, 2011), what is called ‘System 1’ is a collection of autonomous systems or mechanisms (‘the autonomous set of systems’ or ‘TASS’). Many of these systems are modular: that is, they are informationally encapsulated and are dedicated to specific tasks (see Box 3B).

BOX 3B: Modularity

Definition of modularity:

‘[M]odular systems can be defined as systems made up of structurally and/or functionally distinct parts. While non-modular systems are internally homogeneous, modular systems are segmented into modules, i.e., portions of a system having a structure and/or function different from the structure or function of other portions of the system.’ (Calabretta & Parisi 2005, 309)

What is a mental module?

It is a system within the human mind that is responsible for specific cognitive tasks. Some of the module’s characteristics include fast processing and informational encapsulation, which means that the module is not affected by information from other systems (Fodor 1983).

Massive modularity hypothesis:

‘[It] maintains that our cognitive architecture – including the part that subserves “central processing” – is largely or perhaps even entirely composed of innate, domain-specific computational mechanisms or modules.’ (Samuels 1998, 575)

Again, ‘System 2’ is divided into at least two parts: the algorithmic part and the reflective part. The distinction between the algorithmic part and the reflective part corresponds to the distinction between cognitive ability and thinking disposition. Cognitive abilities are the ‘measures of the ability of the algorithmic mind to sustain decoupled representations (for purposes of inhibition or simulation)’, while thinking dispositions are the ‘measures of the higher-level regulatory states of the reflective mind’ (Evans & Stanovich 2013a, 230).

The dual-system theory is not that plausible if it is understood as the token thesis. Can there be another interpretation of it?

Samuels suggests another, more promising, interpretation of the dual-system theory which he calls the ‘type thesis’. The dual-system theory is not about the number of cognitive mechanisms found in the human mind but rather about the number of kinds of mechanisms found in the human mind. That is, the claim is not that there are two cognitive mechanisms; rather there are two kinds of mechanisms. In Samuels’ words: ‘[E]ach mind is comprised of two types or kinds of cognitive system […] each exhibits a fundamental, bipartite division into kinds or types of cognitive system’ (Samuels 2009, 133).

A similar proposal is to interpret the dual-system theory as what Evans calls the ‘two-minds hypothesis’, according to which there are two minds, both of which include multiple systems: ‘there are forms of cognition which are ancient and shared with other animals, and those that are recently evolved and distinctively human’ (Evans 2012, 125) and ‘each mind can have access to multiple systems in a meaningful sense of that term’ (Evans 2012, 125–126).

Two Agents?

In his book Thinking, Fast and Slow, Kahneman talks about the dual-system theory in the following way: ‘systems’ in the dual-system theory are ‘agents within the mind, with their individual personalities, abilities, and limitations’ (Kahneman 2011, 28). It turns out, however, that his descriptions are only useful metaphors. For him, ‘System 2 calculates products’ is shorthand for ‘[m]ental arithmetic is a voluntary activity that requires effort, should not be performed while making a left turn, and is associated with dilated pupils and accelerated heart rate’ (Kahneman 2011, 29). This suggests that Kahneman’s version of the dual-system theory collapses to the dual-process theory.

But why can’t we take the agential talk literally rather than metaphorically? Kahneman argues that the literal interpretation ‘is considered a sin in the professional circles in which I travel, because it seems to explain the thoughts and actions of a person by the thoughts and actions of little people inside the person’s head’ (Kahneman 2011, 28–29). Against Kahneman, however, one might argue that the dual-agent theory is helpful since, for example, it helps us to make sense of why two conflicting beliefs can coexist. In the Linda experiment, for example, Gould believes that ‘[the feminist bank teller hypothesis] is least probable’ but, at the same time, he also believes that Linda ‘can’t just be a bank teller’ (Gould 1991, 469). In other words, he has two conflicting beliefs: the belief that (1) the feminist bank teller hypothesis is less probable than the bank teller hypothesis; and the belief that (2) the former is more probable than the latter.

But this creates a puzzle. Can a person simultaneously adopt two beliefs that obviously conflict with one another? Alfred Mele describes the puzzle as follows: believing two contradictory beliefs simultaneously ‘is not a possible state of mind: the very nature of beliefs precludes one’s simultaneously believing that p is true and believing that p is false’ (Mele 2001, 7). The idea of two sub-agents explains how conflicting beliefs can coexist. It is certainly puzzling that one agent believes two obviously conflicting beliefs simultaneously. But the puzzle is resolved if it turns out that each of the two beliefs is attributed to one of two sub-agents. Attributing belief 1 to sub-agent 1 and belief 2 to sub-agent 2 is no more puzzling than attributing two distinct beliefs to two distinct agents. Let us call this ‘the conflicting belief argument’ for the dual-agent theory. The conflicting belief argument, or something similar to it, has been expressed in psychology and philosophy. Here are two examples.

Sloman on Criterion S

Steven Sloman (1996) proposes something similar to the conflicting belief argument (although his commitment to the dual-agent theory is not very clear). Attributing two reasoning systems to a single human is warranted if what he calls ‘Criterion S’ is satisfied, and a reasoning task satisfies Criterion S ‘if it causes people to simultaneously believe two contradictory responses’ (Sloman 1996, 11). According to Sloman, Criterion S is satisfied in the Linda experiment. Gould believes that (1) the feminist bank teller hypothesis is less probable than the bank teller hypothesis, and that (2) the former is more probable than the latter.

Davidson on Mental Partitioning

In defending what he calls ‘the partitioning of the mind’, which is somewhat similar to the idea of a dual cognitive agent, Donald Davidson (2004) argues that two conflicting beliefs cannot be attributed to a single agent because doing so makes the mind of the agent unintelligible – it compromises the very practice of explaining and predicting the behaviour of the agent by attributing beliefs and desires to the agent. Thus, two conflicting beliefs need to be attributed to two independent parts of the agent: ‘[I]f parts of the mind are to some degree independent, we can understand how they are able to harbor inconsistencies’; ‘[T]he point of partitioning [in the mind] was to allow inconsistent or conflicting beliefs and desires and feelings to exist in the same mind, while the basic methodology of all interpretation tells us that inconsistency breeds unintelligibility’ (Davidson 2004, 184).

However, the conflicting belief argument is not very compelling. The first problem is that it is not clear why two conflicting beliefs cannot be attributed to a single agent. An agent believing two conflicting beliefs simultaneously would make the agent irrational, but we have already seen several times in this book that humans can be irrational (to at least some extent in some cases). The attribution of two conflicting beliefs to the same agent is problematic only for someone who embraces a rather idealized conception of agents where agents need to be rational (in the sense of having mutually consistent beliefs).

The second problem of the conflicting belief argument concerns the crucial assumption in this argument, namely the assumption that there are two conflicting beliefs. Note that this assumption is crucial for the conflicting belief argument, which relies on one agent having two conflicting beliefs (e.g., Gould believing that the feminist bank teller hypothesis is less probable and that it is more probable than the bank teller hypothesis). There is no puzzle if the assumption is false (e.g., Gould believes that the feminist bank teller hypothesis is less probable than the bank teller hypothesis and imagines that the former is more probable than the latter).

Does Gould really simultaneously believe that the feminist bank teller hypothesis is less probable and that it is more probable than the bank teller hypothesis? Isn’t it more plausible that, after learning the correct answer, Gould no longer believes that the feminist bank teller hypothesis is more probable? He believes that the feminist bank teller hypothesis is less probable than the bank teller hypothesis. In contrast, it is not clear that Gould also believes that the feminist bank teller hypothesis is more probable than the bank teller hypothesis. It is certainly true, from his description, that he cannot rid himself of the thought that the feminist bank teller hypothesis is very likely to be true. But this thought does not have to be understood as a belief. This possibility is discussed in the next section.

3.5 The Dual-State Theory

Two States

Another idea that tends to be associated with the dual-process theory is that Type-1 processes and Type-2 processes involve two kinds of mental states (or attitudes). Let us call this ‘the dual-state theory’.

Psychologists working on people’s attitudes have observed that attitudes are typically unstable and that they are likely to change when people are asked to reflect on the object of their evaluation, suggesting that people have some Type-1 attitudes that are largely unconscious and implicit and some Type-2 attitudes that emerge only after reflection is prompted and are conscious and explicit. What is particularly interesting is that the content of Type-1 attitudes about an object of evaluation contrasts with the content of Type-2 attitudes. For instance, experiments indicate that research participants are likely to shift their commitments when they are asked to give reasons for previously held attitudes (Hodges & Wilson 1993) and are more vulnerable to the effects of evidence manipulation when they are asked to give reasons for their attitudes (Wilson, Hodges, & LaFleur 1995).

Here is one example from the dual-attitude literature. Couples who have been dating for a few months are asked to provide reasons why they are attracted to their partner and are then asked to rate their commitment to the relationship and the likelihood that they will live together or get married in the future (Seligman, Fazio, & Zanna 1980). Participants’ responses are elicited via questioning. Guided by the formulation of the questions, some couples are invited to offer intrinsic reasons for being together (‘I date X because …’), whereas others are invited to offer extrinsic and more instrumental reasons (‘I date X in order to …’). Research participants invited to give extrinsic reasons ended up rating their attitudes towards their partners more negatively and tended not to predict living together or getting married in the future. They did not seem to realize that their reports were biased by the way in which the questioning was conducted: the fact that they provided reasons for their attitudes does not guarantee that they reported an attitude that accurately represented how they felt and what they thought about the relationship.

Timothy Wilson and Dolores Kraft (1993) designed a similar study on attitudes towards one’s partner (but without evidence manipulation). Participants were asked to report their attitude towards their relationship. They were then divided in two groups and were asked, first, to report their attitude towards their relationship again and, second, to predict the future of their relationship. In one group, they were also asked to list reasons for the success or failure of their relationship before making the prediction. Results show that participants who were asked for reasons for the state of their relationship experienced a shift between their initial attitude towards the relationship and their prediction.

Psychological models attempt to explain the instability of attitudes when people are asked to provide reasons for them. One model appeals to retrieval mechanisms and the idea that attitudes towards an object or person are constructed mainly on the basis of those aspects of the object or person to be evaluated that are easily accessible when the relevant information is retrieved (Wilson et al. 1984). By looking for reasons for a previously reported attitude, people justify their attitude based on the most accessible reasons for having that attitude. These reasons might be different from what caused the initial attitude and may give rise to a different attitude. Jaideep Sengupta and Gavan Fitzsimons give up the idea that people have persistent attitudes independent of context and say that ‘attitudes are freshly computed based on available contextual cues’ (Sengupta & Fitzsimons 2004, 711). That means that attitudes are fleeting, being formed from scratch every time they are elicited.

A new search for reasons may cause participants to form an entirely new attitude towards their relationship. Initially, participants may report an attitude that reflects how they feel about their relationship at that time (maybe an attitude formed via a Type-1 process). The attitude reported after a search for reasons may be affected by the weighing of benefits and costs (maybe an attitude formed via a Type-2 process). It is not surprising that different processes give rise to different attitudes. This explains the effects of evidence manipulation that participants experience in the dating couples study.

Non-Doxastic Dual-State Theory

It is tempting to think that the dual-process theory implies the dual-state theory. Type-1 processes and Type-2 processes give rise to two kinds of mental states. Uriah Kriegel nicely expresses this idea: ‘[w]herever there is a duality of processes, there is likely also a duality of products’ and ‘[t]he associationist system likely produces one mental state, the rationalist system another’ (Kriegel 2012, 474). Again, it is tempting to think that the dual-state theory is implied by the dual-system theory. Keith Frankish writes that ‘[d]ual-system theorists typically assume that the two reasoning systems have separate databases, and it is attractive to link the two types of belief we have distinguished with the two systems’ (Frankish 2009, 274).

Let us consider this proposal in relation to Gould’s response to the Linda experiment. Gould has a mental state, M1, about the feminist bank teller hypothesis being less probable than the bank teller hypothesis; he also has a mental state of a distinct kind, M2, about the feminist bank teller hypothesis being more probable than the bank teller hypothesis.

There are various versions of the dual-state theory, with different accounts of what M1 and M2 actually are. We can roughly distinguish doxastic dual-state theory from non-doxastic dual-state theory. According to the former, both M1 and M2 are broadly doxastic states (‘doxastic’ means ‘related to beliefs’) (e.g., M1 and M2 are different kinds of beliefs). According to the latter, at least one of them is not doxastic (e.g., M1 is a belief and M2 is a non-doxastic state).

Gendler on Aliefs and Beliefs

Let us start with Tamar Gendler’s non-doxastic dual-state proposal. Gendler (2008a) explicitly justifies her proposal, the alief/belief distinction, based on the heuristics and biases programme and the dual-process theory. On her view, M1 is belief and M2 is what she calls an ‘alief ’. Gould believes that the feminist bank teller hypothesis is less probable than the bank teller hypothesis, while he alieves that the feminist bank teller hypothesis is more probable than the bank teller hypothesis.

Gendler introduces the concept of aliefs with some examples. One example involves an experience of the Grand Canyon Skywalk. A man, let us call him Wilhelm, is convinced that the solid clear glass floor of the Grand Canyon Skywalk is safe, but when he is actually on it he suddenly feels extreme fear and becomes unable to move. In this case, Wilhelm believes that the Skywalk is perfectly safe (otherwise he would not have stepped on it in the first place). Nonetheless, according to Gendler, Wilhelm alieves that the Skywalk is dangerous, which explains his affective and behavioural responses.

Gendler’s general characterization of an alief is as follows:

A paradigmatic alief is a mental state with associatively linked content that is representational, affective and behavioral, and that is activated – consciously or nonconsciously – by features of the subject’s internal or ambient environment. Aliefs may be either occurrent or dispositional. (Gendler 2008a, 642)

This characterization suggests that an alief is a composite mental state that is constituted by the following components:

1. a representational component (e.g., ‘Really high up and a long way down.’);
2. an affective component (e.g., ‘Not a safe place to be!’); and
3. a behavioural component (e.g., ‘Get off !’).

However, understanding aliefs in this way makes Gendler’s proposal problematic (e.g., Currie & Ichino 2012; Dogett 2012). Aliefs are presented as composite entities that are constituted by representational, affective, and behavioural components. But why is it that the representational, affective, and behavioural components constitute a composite state? Why can’t we simply explain cases like the Skywalk case in terms of representational, affective, and behavioural processes without positing a composite state that is constituted by them? Introducing a composite state seems to be explanatorily redundant.

In some places, Gendler hints at a different, less problematic characterization of aliefs: ‘[T]o have an alief is to have an innate or habitual propensity to respond to an apparent stimulus in a particular way’ (Gendler 2011, 41). These remarks suggest another understanding of aliefs: an alief is understood not as a composite mental state that is constituted by representational, affective, and behavioural components, but rather as a propensity or disposition to go through a series of cognitive, affective, and behavioural processes that are statistically (but not always) correlated. To say that Wilhelm in the Skywalk case alieves that the Skywalk is dangerous is to say that he is disposed to go through some particular cognitive (e.g., thinking about the idea of falling down), affective (e.g., fear), and behavioural (e.g., unable to move) processes (see Schwitzgebel 2013 for a dispositional account of beliefs).

Doxastic Dual-State Theory

Frankish on Type-1 and Type-2 Beliefs

Keith Frankish (2009) proposes a doxastic dual-state theory according to which what we commonly call ‘belief ’ may actually be two or more different types of mental states with distinct features: Type-1 beliefs and Type-2 beliefs. On his view, M1 is a Type-2 belief and M2 is a Type-1 belief. Gould Type-2 believes that the feminist bank teller hypothesis is less probable than the bank teller hypothesis, while he Type-1 believes that the feminist bank teller hypothesis is more probable than the bank teller hypothesis.

A key feature of beliefs is consciousness: we might think of beliefs as conscious states, but some beliefs are actually never actively thought about and they remain in the background. Frankish’s example is that we have many beliefs about how a tap works, but we rarely think about these beliefs or express them out loud.

Another key feature is control: we tend to think of beliefs as mental states that we decide to adopt, but many beliefs are formed automatically as a result of sub-personal mechanisms. Consider the following example: people may glance out of the window and pick up an umbrella on their way out of the house; the best explanation for their action is that they believe that it is raining or that it will be raining soon. However, they have never decided to believe that it is raining or that it will be raining soon; they have not weighed up the reasons for and against the claim that it is raining.

There are some other relevant features of belief that Frankish considers, but simply by reflecting on consciousness and control it is easy to see that beliefs come in two types: those that are conscious and are typically the outcome of deliberation, and those that are unconscious and automatic. Frankish calls the former Type-1 beliefs and the latter Type-2 beliefs, aligning them with the dual-system theory, where Type-1 beliefs are understood as outputs of System 1 and Type-2 beliefs as the outputs of System 2. Type-1 beliefs work like behavioural dispositions, which are reflected in non-verbal behaviour but are not explicitly stated or challenged. Type-2 beliefs work like opinions, which are reflected in verbal behaviour but do not necessarily impact on people’s actions – when they do, it happens via the causal relationship between Type-1 and Type-2 beliefs. Frankish argues that the notion of Type-2 belief captures what in the philosophical literature has been known as an ‘acceptance’, almost like a policy to which agents explicitly commit. A useful example here is that of implicit racial bias. Many people have a Type-2-belief that all humans are equal, independent of their skin colour (their explicit policy). However, those people might also behave in ways that suggest that they have racist Type-1-beliefs (e.g., non-conscious and non-controlled associations between black people and violence).

Schwitzgebel on In-Between Beliefs

We would also like to mention Eric Schwitzgebel’s proposal. Although it is not strictly speaking a dual-state theory, it is relevant to our discussion – it could be described as a multiple-state theory since it recognizes beliefs, non-beliefs, and attitudes in-between belief and non-belief. Schwitzgebel suggests that it would be helpful to consider some mental states as neither beliefs nor non-beliefs: he calls them in-between cases (Schwitzgebel 2001).

Schwitzgebel discusses the example of Antonio, who is sometimes disposed to believe that there is a benevolent deity and is sometimes disposed to believe that God is just a beautiful metaphor. How Antonio feels about God and what he thinks about God changes depending on the context: on a beautiful spring day Antonio may be disposed to believe in God but at a dull church social event he may not. According to Schwitzgebel, there are numerous cases of in-between believing and a dispositional account of belief can handle those cases.

In-between cases of believing are cases where a mental state has some, but not all, of the dispositional properties of a belief. One dispositional property of a belief is that if you have a belief, then you are disposed to behave in a way that manifests your commitment to the truth of the content of your belief. Here is an example from Schwitzgebel where a person asserts something that sounds like a belief but does not have the right dispositional properties:

Daiyu’s belief that most pearls are white has entirely atypical effects. It does not cause her to say anything like ‘most pearls are white’ (which she would deny; she’d say instead that most pearls are black) or to think to herself in inner speech that most pearls are white. She would not feel surprise were she to see a translucent purple pearl. If a friend were to say to Daiyu that she was looking for white jewelry to accompany a dress, Daiyu would not at all be inclined to recommend a pearl necklace. Nor is she disposed to infer from her belief that most pearls are white that there is a type of precious object used in jewelry that is white. (Schwitzgebel 2012, 14)

For Daiyu’s mental state to be a belief it needs to have the dispositional properties of a belief. When a mental state fails to have the dispositional properties of a belief, such as when Daiyu does not infer from her mental state with the content ‘Most pearls are white’ that a precious object used in jewellery is white, then we cannot call that mental state a belief. Depending on how many of the dispositional properties of a belief Daiyu’s mental state has, it could be an in-between case or something else altogether. It is therefore conceivable that mental states resulting from System 1 fail to have the dispositional profile of well-behaved beliefs because they are not used in inference or are not explicitly endorsed. It is also conceivable that mental states resulting from System 2 fail to have the dispositional profile of well-behaved beliefs because they do not drive behaviour.

The three proposals above (Gendler’s belief/alief distinction, Frankish’s Type-1/Type-2 distinction, and Schwitzgebel’s in-between cases of believing) contribute to a better understanding of the complexity of human mental lives and help us acknowledge how heterogeneous human mental states can be. It is worth noticing, though, that each of these views seems to rely quite heavily on the over-idealized concept of agency that we challenged earlier.

Gendler’s proposal is supposed to answer puzzle cases such as the Skywalk case. ‘Why is it that Wilhelm in the Skywalk case feels extreme fear and becomes unable to move despite his belief that the Skywalk is perfectly safe?’ This is puzzling given the assumption that Wilhelm’s feelings and behaviour are always coherent with what he believes. But why should we accept this assumption? Isn’t this an over-idealized assumption about human agency?

Again, the idea that Antonio cannot really believe that God exists given his often-conflicting thoughts and emotions concerning God is grounded in the idea that, for Antonio to have that belief, his behaviour would have to be more coherent. Similarly, consider Sylvia, who systematically prefers to sit next to a white person rather than a black person on the bus (see Chapter 6 for a more thorough discussion of such cases). The idea that she cannot Type-1 believe that black people are equal to white people, but maybe only Type-2 believe it, is grounded in the idea that for Sylvia to Type-1 believe that black people are equal to white people her behaviour would always have to match her Type-1 belief. Antonio and Sylvia are unlikely to be as consistent in their thoughts and actions as we would like them to be – which does not necessarily tell us anything deep about the metaphysics of belief, but may simply be a by-product of their (and our) inescapable irrationality.

3.6 Summary

• The basic idea of the dual-process theory of reasoning is plausible. The theory’s empirical plausibility comes from its consistency with a variety of empirical studies, including the study of reasoning biases. The theory’s theoretical plausibility relies on the fact that it enables us to solve the rationality paradox as well as the self-defeating rationality paradox.
• However, there are a number of issues with the theory. In particular, there is no agreement on how Type-1 processes and Type-2 processes should be characterized, nor is there agreement on how Type-1 processes and Type-2 processes interact.
• The dual-process theory is often associated with the dual-system theory. But it is not a trivial task to find the interpretation of ‘system’ according to which the dual-system theory is plausible. It should not be interpreted as the token thesis (the idea that there are exactly two systems in the brain). Perhaps it should rather be interpreted as the type thesis (the idea that there are two kinds of systems in the brain).
• The dual-process theory is also associated with the dual-state theory. Frankish proposes a doxastic dual-state theory that distinguishes Type-1 beliefs from Type-2 beliefs. Gendler, in contrast, proposes a non-doxastic dual-state theory that distinguishes aliefs from beliefs. These attempts to distinguish between different types of states are promising and insightful but tend to assume that agents’ behaviour is rational.

Further Resources

Articles and Books

We highly recommend Jonathan Evans’ article (2003) in Trends in Cognitive Science and his review article (2008) in Annual Review of Psychology. Keith Frankish’s review article (2010) in Philosophy Compass and Frankish and Evans’ historical overview of the dual-process theory (2009) are accessible and informative.

Daniel Kahneman’s best-selling book Thinking, Fast and Slow (2011) integrates the heuristics and biases programme into the dual-process (or the dual-system) framework. Keith Stanovich’s Who is Rational?: Studies of Individual Differences in Reasoning (1999) provides an interesting take on the strengths and limitations of the dual-system theory and poses a question concerning how to think about individual differences in reasoning performance. Rationality and Reasoning (1996) by Jonathan Evans and David Over offers a dual-process account of deductive reasoning and its biases.

The dual-process theory and the dual-system theory have also been the target of challenges and have generated controversy. For an overview of the debate, see the journal Perspectives on Psychological Science, Volume 8, Issue 3 (2013), which includes a target article by Evans and Stanovich, followed by critical discussions. For psychological and philosophical discussions on the dual-process theory and the dual-system theory, we recommend the collection In Two Minds: Dual Processes and Beyond (2009), edited by Evans and Frankish.

Online Resources

For an accessible introduction to some of the themes discussed in this chapter, see Eric Mandelbaum’s Stanford Encyclopedia of Philosophy entry on associationist theories of thought (2017). On YouTube you will find Keith Frankish’s talk on the dual-system distinction (2013). There is an interesting post on dual-system theory and the relative benefits of the two systems in Scientific American, authored by Scott Barry Kaufman and Jerome L. Singer in 2012.

Questions

1. Which cognitive tasks in your everyday life are most likely to be due to Type-1 processing? Which are most likely to be handled by Type-2 processing?
2. What are the implications of the fragmentation of the mind, and in particular of the distinction between Type-1 and Type-2 processes or systems, for the prospects of improving the quality of human reasoning via education and training?
3. If you have a Type-1 attitude and a Type-2 attitude (this can be about racial differences, religious affiliation, or your romantic partner), which one is your real attitude?

4.1 Introduction

This chapter continues the discussion of the fragmentation of the mind but with a change of focus. In the previous chapter we discussed a fragmentation that concerns reasoning, in particular the fragmentation between Type-1 reasoning processes that are (typically) unconscious, automatic, low effort, associative, and so on, and Type-2 reasoning processes that are (typically) conscious, controlled, high effort, rule-based, and so on. In this chapter we discuss a different but related form of fragmentation, namely the fragmentation of the processes by which we arrive at moral judgments.

As we saw in the previous chapter, the dual-process theory is a useful theoretical framework for making sense of reasoning processes. The dual-process framework has also been applied in other fields of psychological study, such as research on learning and on social cognition (see Evans 2008 for an overview). We will argue in this chapter that the dual-process framework is also useful for making sense of the processes by which people make moral judgments, such as the judgment that it is wrong to harm an innocent child or that it is right to help an elderly stranger who is suffering.

Emerging views in the psychological and neuroscientific studies of moral judgments suggest a dual-process structure according to which moral judgments are determined by the interaction between two kinds of processes: emotion-based processes and reasoning-based processes. This chapter will investigate details of the dual-process structure of moral judgments. Of particular importance are the empirical studies of moral judgments by Jonathan Haidt and Joshua Greene. Both Haidt and Greene propose influential dual-process-theoretical accounts of moral judgments, but their accounts are importantly different in a number of ways. One of the objectives of this chapter is to clarify and discuss the difference between their two accounts.

We will begin by discussing a series of influential studies of moral judgments by Eliot Turiel and his colleagues and critiques of these studies by Richard Shweder and Jonathan Haidt (Section 4.2). This debate will lead us to discuss the dual-process theory of moral judgment according to which moral judgments are the product of the interaction between emotion-based processes and reasoning-based processes. There are, however, several important details that need to be specified. One issue is what we call ‘the interaction question’ (discussed in Section 4.3), which asks: ‘How do emotion-based processes and reasoning-based processes interact?’ In Haidt’s model, reasoning processes have a relatively weak role (of rationalizing affective judgments in a post hoc manner), but they have a stronger role (of competing with affective processes or intervening in them) in Greene’s model. Another issue is what we call ‘the process question’ (discussed in Section 4.4), which asks: ‘What are emotion-based processes and reasoning-based processes?’ Greene suggests the intriguing idea that the distinction between emotion-based processes and reasoning-based processes corresponds to the philosophical distinction between utilitarianism and deontology. This chapter will examine whether this view is defensible.

4.2 Harm and Emotion

Moral and Conventional

Let us start with the influential developmental psychological studies of moral judgments by Turiel and followers.

Moral–Conventional Tasks

One of Turiel’s main hypotheses is that the distinction between moral rules and conventional rules is psychologically real, and this distinction is made by both adults and children (even at an early stage of cognitive development). See, for example, the two stories below. Children are aware of the difference between a violation of a moral rule (‘a moral violation’) and a violation of a conventional rule (‘a conventional violation’). Below is an example of each:

A number of nursery school children are playing outdoors. There are some swings in the yard, all of which are being used. One of the children decides that he now wants to use a swing. Seeing that they are all occupied, he goes to one of the swings, where he pushes the other child off, at the same time hitting him. The child who has been pushed is hurt and begins to cry. (Turiel 1983, 41)

Children are greeting a teacher who has just come into the nursery school. A number of children go up to her and say ‘Good Morning, Mrs. Jones.’ One of the children says ‘Good Morning, Mary.’ (Turiel 1983, 41)

Turiel’s hypothesis is consistent with the empirical findings, in particular those involving what are known as ‘moral–conventional tasks’ in which adults and children reliably distinguish moral violations from conventional violations in multiple ways (Nucci 2001; Nucci & Turiel 1978; Smetana 1981; Turiel, Killen, & Helwig 1987). First, moral violations are typically regarded as more serious than conventional transgressions: for example, pushing the other child off the swing is worse than calling the teacher by her first name. Second, the wrongness of moral violations is more authority-independent than the wrongness of conventional violations: for example, pushing the other child off the swing is still wrong even if an authority figure (say, a teacher) says that it is OK, while calling the teacher by her first name is OK if an authority figure (say, the teacher herself) says that it is OK. Third, the wrongness of moral violations is more universal than the wrongness of conventional violations: for example, pushing the other child off the swing is wrong even in a country with a very different culture and values, while calling the teacher by her first name might be OK in a country with a very different culture and values.

Another important finding is that moral judgments and conventional judgments are justified differently. The wrongness of moral violations, but not the wrongness of conventional violations, tends to be explained or justified in terms of ‘harm’ in a broad sense (i.e., negative impacts on welfare, justice, or the rights of individuals involved). For example, the wrongness of pushing the other child off the swing tends to be explained or justified in terms of the harm that the act causes to the child, while the wrongness of calling the teacher by her first name is not usually justified by reference to the harm to the teacher.

These findings suggest that moral judgments are clearly distinguished from conventional judgments. Moral judgments, unlike conventional judgments, are the responses to perceived harms, and moral violations are judged to be seriously wrong, authority-independently wrong, universally wrong, and so on. Let us call this view ‘the harm account’ of moral judgments.

The studies by Turiel and colleagues are remarkable and they strongly support the harm account. However, several researchers, most notably Richard Shweder and Jonathan Haidt, are sceptical. Their main worry is that the findings by Turiel and colleagues might only be true about some particular group of people; they might not be generally applicable to other cultural or socio-economic groups. For example, an interview study (Shweder, Mahapatra, & Miller 1987) found that people in Bhubaneswar, India, tended to moralize harmless violations. For instance, people in Bhubaneswar treated a wide range of harmless violations related to food, clothing, sex roles, and other practices as moral violations, exhibiting the moral-type response to them (i.e., regarding the relevant violation as seriously wrong, authority-independently wrong, universally wrong, etc.). For example, ‘A widow in your community eats fish two or three times a day’, which does not seem to harm anyone, was treated as a moral violation rather than a conventional one.

However, this is not a conclusive objection to the harm account. As Turiel, Killen, and Helwig (1987) point out, on closer scrutiny the scenarios in the study by Shweder, Mahpatra, and Miller (1987) might involve some kind of harm given Indian people’s background assumptions about how the world works. For example, ‘A widow in your community eats fish two or three times a day’ is harmful according to an Indian worldview because the act will offend their deceased husband’s spirit and will cause the widow to suffer greatly. Again, ‘The day after his father’s death, the eldest son had a haircut and ate chicken’ (another item in the study by Shweder and colleagues) is harmful because the deceased father’s soul will not receive salvation if the proscription of eating chicken is violated. Thus, despite the challenge by Shweder and colleagues, one can still maintain that the harm account is universally true: moral judgment is a response to perceived harm both in the United States and in India. The difference between Indian people and American people lies in the fact that people in the two cultures have different worldviews; this has wide-ranging implications for what does and does not constitute a harmful action in each culture or worldview (e.g., a widow’s eating fish is harmful for Indian people, but it is not harmful for American people).

Morality and Harm

Jonathan Haidt sets out to resolve the dispute between Shweder and Turiel. Haidt’s aim is to overcome the limitation of the study by Shweder and colleagues, and find more convincing evidence against the harm account.

The Dog Eating Study

This study (Haidt, Koller, & Dias 1993) examines how participants from the United States and Brazil respond to harmful and harmless stories.

• Dog: A family’s dog was killed by a car in front of their house. They had heard that dog meat was delicious, so they cut up the dog’s body, cooked it, and ate it for dinner.
• Chicken: A man goes to the supermarket once a week and buys a dead chicken. Before cooking the chicken, he has sexual intercourse with it. He then cooks the chicken and eats it.
• Flag: A woman is cleaning out her closet and finds her old national flag. She doesn’t want the flag anymore, so she cuts it up into pieces and uses the rags to clean her bathroom.
• Promise: A woman was dying and on her deathbed she asked her son to promise that he would visit her grave every week. The son loved his mother very much, so he promised to visit her grave every week. However, after his mother died, the son was too busy to keep his promise.

These stories are prepared to be affect-laden. Some of the stories involve disrespect or disobedience (e.g., Flag and Promise), while others involve the emotion of disgust (e.g., Dog and Chicken). At the same time, these stories are prepared to be harmless. Dog, for example, involves a dead dog, which cannot be harmed. The dog eating is a private and secret act and, thus, nobody will be psychologically disturbed by watching the act or knowing about it.

Just like in the studies by Turiel and colleagues, participants were asked to answer some questions about each story, such as:

• Interference: ‘Should [the actor] be stopped or punished in any way?’
• Universality: ‘Suppose you learn about two different foreign countries. In country A, people [do that act] very often, and in country B, they never [do that act]. Are both of these customs OK, or is one of them bad or wrong?’ (Haidt, Koller, & Dias 1993, 617)

Participants in this study were from three cities in two countries: Philadelphia (United States), Porto Alegre (which is a relatively Westernized city in Brazil), and Recife (a less Westernized city in Brazil). They were also assigned to either a high socio-economic status (SES) group or a low SES group in the city they were in.

The study found that both the city and the SES group influenced the moralizing tendency of the participants (where moralizing tendency is the tendency to treat a violation as a moral violation, measured by their response to the Interference question and the Universality question). The more Westernized participants were, the less moralizing they were. The higher SES group was less moralizing than the lower SES group. The majority of high SES people in Philadelphia did not moralize offensive harmless acts, while the majority of low SES people in Recife did moralize them. For example, most of the high SES people in Philadelphia said that the act in Chicken should not be stopped or punished, and that the custom concerning the act in Chicken is not universal. In contrast, most of the low SES people in Recife said that the act in Chicken should be stopped or punished, and that the custom concerning the act in Chicken is universal.

In addition to the questions about Interference and Universality, participants answered a question about the harm involved in the story (e.g., ‘Do you think anyone was harmed by what the man did to the dead dog?’). It turned out that, against the harm account, an answer of ‘Yes’ to this question about harm was not a very good predictor of participants’ moralizing tendency: many participants moralized Dog and other stories even though participants regarded them as harmless. Thus, stories like Dog are harmless for those participants whatever background assumptions they might have about how the world operates. This means that the study by Haidt and colleagues poses a more serious problem for the harm account than the study by Shweder and colleagues.

During their interview, some participants actually tried to come up with potential harmful aspects of the act in question (e.g., eating dog meat is harmful because it would make a person sick). But it is likely that this type of reaction is a post hoc rationalization of their intuitive response. Indeed, participants did not revise their judgment when the interviewer explicitly denied any alleged harm in the act (e.g., the dog meat was thoroughly cooked and was thus perfectly safe to consume). In light of this, participants tried to find some other harmful aspect of the act. This suggests that it is not the case that participants first recognize a harmful aspect of dog eating and then make a moral judgment about it. Rather they first intuitively make a moral judgment that dog eating is wrong and then give a post hoc reason for that judgment by appealing to whatever harmful consequence they can think of. Their reaction is thus analogous to the participants of the stockings experiment by Nisbett and Wilson (1977) who first chose a pair of stockings unconsciously influenced by its location and then gave a post hoc justification for their decision by appealing to some superior quality of that particular pair of stockings.

The upshot of this is that the harm account is no longer true outside some limited cultural or socio-economic groups. Shweder’s and Haidt’s criticism of the harm account is intimately related to the criticism of the psychological studies that heavily rely on WEIRD participants (Henrich, Heine, & Norenzayan 2010a, 2010b), which we mentioned in the Introduction.

Shweder distinguishes the culture in which the ethics of autonomy (in which a moral agent is regarded as an autonomous and individual being) is dominant, the culture in which the ethics of community (in which a moral agent is essentially tied to the community to which they belong) is dominant, and the culture in which the ethics of divinity (in which morality is essentially tied to spirituality and the essence of moralizing for attaining spiritual purity and sanctity) is dominant. The harm account is closely related to the first kind of culture where the ethics of autonomy is dominant; it is not applicable to other cultures.

Haidt stresses the importance of what he calls ‘moral foundations’ (Graham, Haidt, & Nosek 2009; Graham et al. 2013; Haidt & Joseph 2004). Moral foundations are ‘cognitive modules upon which cultures construct moral matrices’ (Haidt 2012, 124). The five moral foundations are: the care/harm foundation, the fairness/cheating foundation, the loyalty/betrayal foundation, the authority/subversion foundation, and the sanctity/degradation foundation. Different foundations are dominant in different groups. The harm account is closely related to the care/harm foundation; it is not applicable to the cultural socio-economic groups in which the care/harm foundation is not dominant.

Morality and Affect

We have seen that the studies by Shweder and Haidt cast doubt on the harm account of moral judgment. They suggest that moral judgments are not necessarily a response to perceived harm. These studies, in particular Haidt’s study, suggest an alternative view that moral judgments are a response to affectively charged transgressions rather than to harmful transgressions. In fact, the main strategy in the dog eating study was to elicit moral judgments by presenting the ‘stories that are affectively loaded – disrespectful or disgusting actions that “feel” wrong – yet that are harmless’ (Haidt, Koller, & Dias 1993, 615).

A classic question about moral judgment is: ‘Are moral judgments based on reasoning or on emotion?’ A view would be that, when I judge, for example, that killing an innocent person is wrong, I am making the judgment on the basis of my reasoning about the consequences of killing or about relevant rules and duties. Another view would be that, when making a moral judgment about killing, I am making a judgment on the basis of my affective disapproval of the killing.

The dog eating study seems to support the latter view. It turns out, however, that the latter view is ambiguous. For example, the phrase ‘I am making this judgment on the basis of my affective disapproval of killing’ is open to at least two interpretations. It can mean that my judgment is causally influenced by affective disapproval. It can also mean that my judgment is (partly) constituted by affective disapproval. The former view takes affective inputs to be a cause of moral judgments (in which case affective inputs and moral judgments are separate or distinct events). For example, my judgment about killing an innocent person is caused by my affective disapproval of killing an innocent person. The latter view, in contrast, takes affective inputs to be a part of moral judgments (in which case affective inputs and moral judgments are not separate). For example, my judgment about killing an innocent person is (partly) constituted by my affective disapproval of killing an innocent person. Or, simply put, the judgment just is affective disapproval.

We therefore have two slightly but importantly different views about moral judgments and emotions: the view that moral judgments are causally influenced by affective inputs (we call this ‘causal moral sentimentalism’) and the view that moral judgments are partly constituted by affective inputs (we call this ‘constitutive moral sentimentalism’).

Causal moral sentimentalism and constitutive moral sentimentalism are distinct claims, but they are not completely independent. Suppose that it turns out that changes in affective inputs co-vary with changes in moral judgments. For example, one makes different moral judgments depending on whether or not one feels affective disapproval of an act. This is typically taken to be evidence for causal moral sentimentalism. But it could be taken to be evidence for constitutive moral sentimentalism as well: the best explanation of the co-variation would be that moral judgments are (partly) constituted by affective inputs. Thus, causal moral sentimentalism and constitutive moral sentimentalism can be supported by the same (or at least a similar) body of empirical evidence.

The studies by Haidt and colleagues mentioned above are coherent with causal/constitutive moral sentimentalism. Moral judgment (or the moral-type response) can be activated by offensive stories that do not involve harm, such as Flag or Chicken, which can be explained by the fact that: (1) even without the presence of harm, those stories evoke affective reactions of the relevant kind; and (2) moral judgment is caused/constituted by the affective reactions.

Causal/constitutive moral sentimentalism has been supported by other empirical studies too (Greene & Haidt 2002; Haidt 2007). In one remarkable study (Wheatley & Haidt 2005), for example, highly hypnotizable participants participated in group hypnosis sessions in which they were given the post-hypnotic suggestion to feel a flash of disgust either when they read the word ‘often’ or when they read the word ‘take’. Participants were then brought out of their hypnotic state and were asked to evaluate some moral transgressions presented as vignettes, such as the one below. There were two versions of each vignette, one with the word ‘often’ and one with the word ‘take’:

Congressman Arnold Paxton frequently gives speeches condemning corruption and arguing for campaign finance reform. But he is just trying to cover up the fact that he himself [will take bribes from/is often bribed by] the tobacco lobby, and other special interests, to promote their legislation. (Wheatley & Haidt 2005, 781)

The result was that participants rated the transgression in the vignette as being morally worse when a hypnotic trigger word was present in the vignette than when the word was absent. For example, in response to the ‘take’ version of the Paxton vignette, participants who received the post-hypnotic suggestion that they should be disgusted with the word ‘take’ rated Congressman Paxton’s act as morally worse than the ‘often’ version of it. This result can be nicely explained by the causal/constitutive moral sentimentalist idea that disgust with a word causes/constitutes their harsh moral judgment of a vignette containing the word.

These results strongly support the idea that affective inputs causally/constitutively influence moral judgments. However, it is important to note that these results do not necessarily show that affective processes dominate moral judgment, nor do they imply that reasoning processes do not play any role in moral judgment. Think back to the discussions of reasoning in earlier chapters. In Chapter 1 we saw that our reasoning is often based on heuristic processes, but in Chapter 3 we also argued that heuristic Type-1 processes interact with Type-2 processes. A similar thing can be said about emotion and reasoning in moral judgment: moral judgment is often affective, but affective processes might interact with reasoning processes. In other words, there can be an analogy between the dual structure concerning reasoning where Type-1 processes and Type-2 processes interact with each other and the dual structure concerning moral judgment where affective processes and reasoning processes interact with each other.

Indeed, the analogy between heuristic Type-1 processes in reasoning and affective processes in moral judgment seems apt (e.g., Sunstein 2005). An affective moral judgment can be understood as a form of heuristic Type-1 judgment. As Kahneman noted, a heuristic judgment replaces a difficult question with an easy one, for example replacing a difficult probabilistic judgment (e.g., ‘What is the probability of Linda being a bank teller?’) with an easy stereotypical judgment (‘Is Linda like a stereotypical bank teller?’). We can make sense of affective influence on moral judgment in the same way. Perhaps what we do is replace a difficult question of morality (‘Is this the right thing to do?’) with an easy question of emotion (‘How do you feel about this action?’).

We will examine the dual-process theory of moral judgment in the next section.

4.3 Interaction between Affective Processes and Reasoning Processes

Models of Interaction

We have discussed some basic ideas of the dual-process theory of moral judgment according to which affective processes and reasoning processes interact with each other. We will now clarify this theory by investigating two crucial questions (which correspond to the process question and the interaction question discussed in Chapter 3):

• Process: What are ‘affective processes’ and ‘reasoning processes’?
• Interaction: How do affective processes and reasoning processes ‘interact’?

Let us start with the interaction question. In Chapter 3 we discussed three models of interaction or relation between Type-1 and Type-2 processes of reasoning: the parallel-competitive model, the default-intervention model, and the lawyer–client mode. Analogously, we can think of three models of interaction between affective processes and reasoning processes (see Box 4A).

BOX 4A: Models of Interaction between Type-1 and Type-2 Processes

The parallel-competitive model of moral judgments

Affective processes and reasoning processes compete with each other as equally strong competitors. Affective processes are dominant in some cases, and reasoning processes are dominant in others.

The default-intervention model of moral judgments

Affective processes are the default processes and reasoning processes play a relatively weaker role. Affective processes are active by default, but in some cases reasoning processes can intervene and dominate.

The lawyer–client model of moral judgments

Affective processes are dominant (almost) all of the time and the role of reasoning processes is even weaker. The primary role of reasoning processes is not to intervene but rather to rationalize the output of emotion-driven moral judgments in a post hoc manner, just like a lawyer rationalizes the claims of her client in a post hoc manner.

Haidt (2001) proposes the social intuitionist model of moral judgment, which is a good example of the lawyer–client model. (Or, more precisely, the lawyer–client analogy is something we borrow from Haidt himself.) Reasoning processes are more like ‘a lawyer trying to build a case than a judge searching for truth’ (Haidt 2001, 814). Haidt also proposes the analogy of the press secretary for a secret administration: ‘Moral reasoning is often like the press secretary for a secretive administration – constantly generating the most persuasive arguments it can muster for policies whose true origins and goals are unknown’ (Haidt 2007, 1000).

The social intuitionist model has two major claims: first, that ‘moral intuitions (including moral emotions) come first and directly cause moral judgments’; and, second, that ‘[m]oral reasoning is usually an ex post facto process used to influence the intuitions (and hence judgments) of other people’ (Haidt 2001, 814). The social intuitionist model is nicely illustrated by what Haidt calls ‘moral dumbfounding’ (Haidt, Bjorklund, & Murphy 2000).

Julie and Mark are brother and sister. They are traveling together in France on summer vacation from college. One night they are staying alone in a cabin near the beach. They decide that it would be interesting and fun if they tried making love. At the very least it would be a new experience for each of them. Julie was already taking birth control pills, but Mark uses a condom too, just to be safe. They both enjoy making love, but they decide not to do it again. They keep that night as a special secret between them, which makes them feel even closer to each other. What do you think about that? Was it wrong for them to make love? (Haidt 2001, 814)

In response to this story, people first intuitively think that it is wrong for Julie and Mark to have sexual intercourse. When asked ‘Why is it wrong?’, they search for a post hoc justification of the intuitive answer. It turns out that the Julie and Mark story is cleverly constructed in such a way that it is very difficult to come up with a truly satisfactory justification of the intuitive answer. Some might appeal to the dangers of inbreeding, which is not a good justification: it is said that Julie and Mark carefully use two forms of birth control. Others might appeal to the bad effect on the relationship between Julie and Mark, which, again, is not a good justification: it is said that Julie and Mark feel even closer to each other. Eventually, many people end up saying something like ‘I know it’s wrong, but I just can’t come up with a reason why.’

Note that the social intuitionist model is not committed to the radical claim that reasoning processes only provide post hoc rationalizations of intuitive responses. Haidt carefully distances himself from Hume’s radical claim that ‘reason is, and ought only to be the slave of the passions, and can never pretend to any other office than to serve and obey them’ (Hume 1739/2007, 266). Hume’s model seems to be that the relationship between affective processes and reasoning processes is just like the one between slaves and their master, where the slaves can only obey the order of their master. In contrast, Haidt’s model is that the relationship between them is more like the one between a lawyer and their client: ‘Good lawyers do what they can to help their clients, but they sometimes refuse to go along with requests’ (Haidt 2012, 68). Reasoning processes might overweigh affective processes when a person has enough time to consider the question, when the person has a strong motivation to be accurate, when affective input is relatively weak, and so on. (Thus, the social intuitionist model contains an element of the default-interventionist model.)

Still, the social intuitionist model insists that it is atypical that reasoning processes refuse to go along with intuitive responses; post hoc rationalization is the typical activity and is the main job of reasoning processes. When it comes to determining one’s moral judgment and its content, the role of affective processes is overwhelmingly more important than that of reasoning processes.

In response, several philosophers have challenged Haidt’s social intuitionist model and the more general argument for the dominating role of raw emotional reactions in the formation of moral attitudes. For instance, Jeanette Kennett and Cordelia Fine (2009) have pointed out that Haidt attacks the causal role of reason in moral judgment by emphasizing: (1) the importance of automatic evaluations and intuitive processes in explaining human behaviour (which we reviewed in Chapter 3); (2) the fact that biases affect reasoning (which we reviewed in Chapter 1); and (3) the pervasiveness of post hoc justifications offered by human agents for their behaviour (which we reviewed in Chapter 2).

According to Kennett and Fine, based on (1) to (3), Haidt effectively argues that human agents are not reason responders. Agents can be characterized as reason trackers and reason responders. They are reason trackers when they recognize something as a reason for a judgment or an action, and can judge or act according to such a reason. They are reason responders when, in addition to tracking reasons, they respond to those reasons as reasons for a judgment or an action. Whereas non-human animals can track reasons, they most likely can’t respond to reasons. Reason tracking seems to be something automatic and often unconscious that maps onto System 1. Reason responding seems to be something deliberate and conscious that maps onto System 2.

Traditionally, moral philosophers have attributed to human agents the capacity to respond to reasons and the capacity to be guided by reasoning in their moral judgment. However, Haidt seems to suggest that the capacity for responding to reasons in human agents is an illusion. When it appears that human agents are responding to reasons, in reality they are typically playing the lawyer or the press secretary role, offering reasons as a post hoc justification for a moral judgment they have already arrived at by socially constructed intuitions. According to Kennett and Fine, this picture of moral judgment does not do justice to the fact that human agents have the capacity for moral agency and minimizes the role of reasoning in the formation of moral judgments. How are competent adult human agents different from non-human animals, young children, and incompetent adults when making moral judgments if we cannot describe them as reason responders? Haidt underestimates the fact that human agents can ‘effortfully override judgments based on moral intuitions’ (Kennett & Fine 2009, 93) because they have values and are responsive to reasons.

Despite their challenge to Haidt (namely that he exaggerates the illusory nature of reason responsiveness in human agents), Kennett and Fine recognize the importance of his work and appreciate the fact that the empirical evidence on moral judgment has forced philosophers to think more carefully about the role of intuitions.

It is certainly very plausible that much of the processing underpinning our everyday moral judgment and behaviour is done automatically and with little effortful cognition. Nonetheless, a close examination of the social cognitive psychology literature suggests that Haidt’s (2001) claim that ‘moral reasoning is rarely the direct cause of moral judgment’ (p. 815) is to overstate the primacy of automatic processes in social judgment, and to underplay the contribution of controlled processes. (Kennett & Fine 2009, 88)

More Than a Post Hoc Rationalization

The lawyer-client model of interaction (such as the social intuitionist theory) can be challenged for empirical reasons as well. In particular, Greene and colleagues conducted a series of fMRI experiments whose results better fit the parallel-competitive model or the default-intervention model than the lawyer–client model.

The Trolley/Footbridge Study

In this influential study, Greene and colleagues (2001) compare two different kinds of sacrificial dilemmas: personal and impersonal dilemmas involving the option of sacrificing someone’s life to save others. Personal sacrificial dilemmas involve the option of sacrificing someone’s life in an ‘up close and personal’ manner (e.g., the footbridge dilemma), while impersonal sacrificial dilemmas involve the option of sacrificing someone’s life in an impersonal manner (e.g., the trolley dilemma).

The Trolley Dilemma: A runaway trolley is headed for five people, who will be killed if it proceeds on its present course. The only way to save them is to pull a lever that will divert the trolley onto an alternate set of tracks where it will kill only one person instead of five. Should you divert the trolley?

The Footbridge Dilemma: A trolley threatens to kill five people. You are standing next to a large stranger on a raised footbridge that spans the tracks and stands in between the oncoming trolley and the five people. In this scenario, the only way to save the five people is to push this stranger off the bridge and onto the tracks below. He will die if you do this, but his body will stop the trolley from reaching the others. Should you push the stranger onto the tracks?

Although these two dilemmas share the same structure of saving five people by sacrificing one, the philosophical community are aware that they tend to elicit different responses: people tend to favour the sacrificing option in the case of the trolley dilemma (i.e., the response in favour of pulling the lever), while they tend to favour the anti-sacrificing option in the case of the footbridge dilemma (i.e., the response against pushing the stranger off the bridge). In this experiment, Greene and colleagues used fMRI to monitor the brain activities of participants while they responded to personal dilemmas (e.g., the footbridge dilemma), impersonal dilemmas (e.g., the trolley dilemma), and non-moral dilemmas. It was found that personal moral dilemmas, when compared to impersonal and non-moral dilemmas, evoke increased activities in brain areas associated with social/affective processing (medial frontal gyrus, posterior cingulate gyrus, and bilateral superior temporal sulcus). In contrast, impersonal and non-moral dilemmas, when compared to personal dilemmas, evoke increased activity in brain areas associated with working memory and abstract reasoning (dorsolateral prefrontal and parietal areas).

This result does not sit well with the lawyer–client model, in which reasoning processes typically play a minor role of post hoc rationalization. The experiment seems to suggest that, at least in the case of impersonal dilemmas, reasoning processes (which are manifested as a pro-sacrificing response to impersonal dilemmas) can either compete with affective processes (the parallel-competitive model) or intervene in them (the default-intervention model).

Greene and colleagues also analysed participants’ response times. It turns out that participants were slow to approve of the personal sacrificial option (e.g., pushing the stranger off the bridge) but relatively quick to disapprove of it. In contrast, approving of the impersonal sacrificial option (e.g., pulling the lever) and disapproving of it took an equally long time. This result is especially coherent with the default-intervention model. Since personal dilemmas elicit a strong affective reaction against the sacrificial option, approving of it can be a simple and quick matter (i.e., just letting the affective reaction manifest itself in the judgment), while disapproving of it requires reasoning processes to intervene in the affective processes and override the affective reaction, which can be slow.

The trolley/footbridge dilemmas (Greene et al. 2001) examine relatively ‘easy’ dilemmas in the sense that one form of response tends to be dominant in them: the sacrificial response (e.g., the response in favour of the pulling the lever) is dominant in impersonal dilemmas such as the trolley case, while the anti-sacrificial response (e.g., the response against pushing the stranger off the bridge) is dominant in personal dilemmas such as the footbridge case. Another contrasting experiment (Greene et al. 2004) examines a rather ‘difficult’ dilemma, known as the crying baby dilemma, in which people’s reactions tend to split.

The Crying Baby Dilemma: Enemy soldiers have taken over your village. They have orders to kill all remaining civilians. You and some of your townspeople have sought refuge in the cellar of a large house. Outside, you hear the voices of soldiers who have come to search the house for valuables. Your baby begins to cry loudly. You cover his mouth to block the sound. If you remove your hand from his mouth, his crying will summon the attention of the soldiers, who will kill you, your child, and the others hiding in the cellar. To save yourself and the others, you must smother your child to death. Should you smother your child?

It turns out that engaging in difficult dilemmas such as the crying baby dilemma causes increased activity in the anterior cingulate cortex, which is associated with dealing with two or more incompatible behavioural responses. Researchers also found that the sacrificial response to a difficult dilemma (e.g., the response in favour of smothering the child to save oneself and others) is correlated with increased activity in the dorsolateral prefrontal cortex, which is the seat of cognitive control and is necessary for overriding impulsive responses. These results provide further evidence of the view that reasoning processes can do something more than post hoc rationalization: reasoning processes can compete with affective processes or intervene in them.

Putting the evidence together, a reasonable view is that, as the phenomenon of moral dumbfounding suggests, reasoning processes rationalize the output of affective processes in a post hoc manner in some cases (i.e., the lawyer–client model). However, as the studies by Greene and colleagues suggest, reasoning processes can do something more than that in other cases: they either compete with affective processes (i.e., the parallel-competitive model) or intervene in them (i.e., the default-intervention model). This conclusion is consistent with what we suggested in the previous chapter concerning the interaction between Type-1 processes and Type-2 processes of reasoning: the conciliatory view according to which the parallel-competitive model, the default-intervention model, and the lawyer–client model are all correct to some extent, but wrong to some extent.

4.4 Affective Processes and Reasoning Processes

The Mapping Thesis

We will now move on to the process question about the dual-process theory of moral judgment. The dual-process theory says that moral judgments are produced by the interaction between affective processes and reasoning processes. But what are ‘affective’ processes and ‘reasoning’ processes?

Different theorists have slightly different terminology. Haidt distinguishes ‘moral intuition’ from ‘moral reasoning’. Moral reasoning is defined as ‘conscious mental activity that consists of transforming given information about people in order to reach a moral judgment’. It is also said to be ‘intentional, effortful, and controllable’ (Haidt 2001, 818). Moral intuition, in contrast, is defined as ‘the sudden appearance in consciousness of a moral judgment, including an affective valence (good–bad, like–dislike), without any conscious awareness of having gone through steps of searching, weighing evidence, or inferring a conclusion’ (Haidt 2001, 818). It is called moral intuition, rather than moral emotion, because the intuitive input can be unconscious and too fast to be regarded as an emotion.

Greene draws a distinction between ‘cognitive’ and ‘emotional’ processes. He defines ‘cognitive’ processes as the processes that deal with ‘cognitive’ representations: that is, representations that are ‘inherently neutral representations, ones that do not automatically trigger particular behavioral responses or dispositions’ (Greene 2008, 40). Emotional processes, in contrast, are the processes that deal with ‘emotional’ representations: that is, representations that ‘do have such automatic [behavioural] effects, and are therefore behaviorally valenced’ (Greene 2008, 40).

Let us recall our discussion of the process question about the dual process of reasoning. Type-1 and Type-2 processes of reasoning are distinguished by clusters of properties: the cluster of Type-1 properties and the cluster of Type-2 properties. Type-1 processes instantiate Type-1 properties and Type-2 processes instantiate Type-2 properties. We can think of a similar approach to the dual process of moral judgment. Affective processes and reasoning processes of moral judgment are distinguished by clusters of properties: the cluster of ‘affective’ properties (fast, effortless, automatic, unconscious, behaviourally valenced, etc.) and the cluster of ‘reasoning’ properties (slow, effortful, controllable, conscious, behaviourally neutral, etc.).

In the context of the process question, the most philosophically interesting issue is about Greene’s claim that the distinction between affective processes and reasoning (or cognitive) processes (which is a distinction concerning the type of processes in the brain) nicely maps onto the distinction between utilitarianism and deontology (which is a distinction concerning fundamental moral principles). Utilitarianism, which was developed by Jeremy Bentham and John Stuart Mill, states that the moral worth of an action (e.g., whether it is good or bad) is determined by the amount of utility or happiness the act brings about. Deontology, which was developed by Immanuel Kant, states that the moral worth of an action is determined by rules concerning duties and rights.

The result of the trolley/footbridge study (Greene et al. 2001) suggests that the distinction between affective processes and reasoning processes maps onto the distinction between personal moral dilemmas (e.g., in the footbridge dilemma) and impersonal moral dilemmas (e.g., in the trolley dilemma). The latter distinction, in turn, maps onto the distinction between the anti-sacrificial response (e.g., not pushing the stranger off the bridge in the footbridge dilemma) and the sacrificial response (e.g., pulling the lever in the trolley dilemma). The distinction between the anti-sacrificial response and the sacrificial response also maps onto the distinction between utilitarian moral judgment and deontological moral judgment. For example, the option of not pushing the stranger off the bridge, an anti-sacrificial response, is a typical deontological option that respects the right of the stranger or the duty of not using a stranger merely as a means to an end. The option of pulling the lever, a sacrificial response, is a typical utilitarian option that maximizes utility or happiness: the total utility is larger when one person is killed to save five people than when five people are killed.

Putting all of these distinctions together, we can conclude that the distinction between affective processes and reasoning processes maps onto the distinction between deontological and utilitarian judgments. In Greene’s words, ‘there is a natural mapping’ (Greene 2008, 63) between the two types of brain processes and the two types of moral principles (see Table 2).

Table 2. The mapping thesis

Ethical theories	Type of processes	Properties
Deontology	emotional, intuitive, affective	fast, effortless, automatic, unconscious, behaviourally valenced
Utilitarianism	reasoning, cognitive	slow, effortful, controllable, conscious, behaviourally neutral

Let us call Greene’s idea ‘the mapping thesis’ (Greene calls it ‘the central tension principle’) and formulate it as follows:

The Mapping Thesis: Deontological judgments are typically produced in affective processes, and utilitarian judgments are typically produced in reasoning processes.

With the mapping thesis at hand, Greene goes on to make some provocative claims: that ‘the psychological essence’ of deontology lies in emotional processes and ‘the psychological essence’ of utilitarianism lies in reasoning processes; that utilitarianism and deontology ‘are not so much philosophical inventions as they are philosophical manifestations of two dissociable psychological patterns, two different ways of moral thinking, that have been part of the human repertoire for thousands of years’; and that ‘the moral philosophies of Kant, Mill, and others are just the explicit tips of large, mostly implicit, psychological icebergs’ (Greene 2008, 38).

The mapping thesis bridges the gap between brain processes and principles of moral philosophy. By bridging the two, the thesis might enable us to draw some interesting consequences on the latter from the research of the former. In fact, Greene argues that his findings have a devastating consequence for philosophical deontology, which is perhaps the most controversial part of his project. Deontology, according to Greene, is discredited because it is the product of the affective, automatic, and heuristic processes ‘that are rather inflexible, and therefore likely to be unreliable, at least in some contexts’ (Greene 2008, 60). The psychological essence of deontology lies with the affective processes that are tracking something irrelevant to morality, namely how ‘up close and personal’ a transgression is. For example, the affective/deontological process is activated in the footbridge dilemma but not in the trolley dilemma because pushing the stranger off the bridge is something very ‘up close and personal’ while pulling the lever to save five people is not. Upon reflection, however, being ‘up close and personal’ does not seem to be a morally relevant property. Sacrificing someone’s life in an up-close and personal way and sacrificing someone’s life in a distant and impersonal manner do not seem to be morally different. Thus, we should be sceptical about the deontological judgment that pushing the stranger off the bridge is wrong.

Greene’s argument runs parallel to the following argument against the trustworthiness of heuristic Type-1 processes. Heuristics such as representational heuristics are inflexible and are thus likely to be unreliable in some contexts. Representational heuristics track something irrelevant to the probability of a given hypotheses, namely similarity to stereotypes (e.g., how Linda’s description is similar to that of a stereotypical bank teller). Thus, we should be sceptical of heuristic judgments (e.g., the judgment that Linda is more likely to be a feminist bank teller than a bank teller).

But, of course, these philosophical consequences follow only if the mapping thesis is true. (Note that we are not saying that these consequences follow if the mapping thesis is true: see Berker 2009; Kahane 2011; and relevant chapters in Sinnott-Armstrong 2008b.) The crucial question, then, is whether the mapping thesis really is true or not. We will examine this question in the rest of this chapter.

Content Interpretation

Evaluating the mapping thesis depends on what the thesis actually means. How should we interpret it? In particular, what is meant by ‘deontological’ judgments and ‘utilitarian’ judgments in the thesis?

An appropriate interpretation of the mapping thesis needs to satisfy two conditions. First, an appropriate interpretation makes it the case that the mapping thesis is empirically plausible in light of relevant empirical studies, such as the one by Greene and colleagues. Let us call this ‘the psychological adequacy condition’. Second, an appropriate interpretation makes it the case that the mapping thesis is philosophically adequate in the sense that ‘utilitarian judgments’ and ‘deontological judgments’ in the thesis are genuinely relevant to utilitarianism and deontology in moral philosophy. Let us call this ‘the philosophical adequacy

⚠️ Output truncated — the model hit its token limit. The conversion is incomplete. Try Gemini 2.5 Pro for longer books.

⚠️ Output truncated — the model hit its token limit. The conversion is incomplete. Try Gemini 2.5 Pro for longer books.