2. Supernatural Agents

3. This page intentionally left blank

4. Supernatural
Agents
Why We Believe in Souls, Gods,
and Buddhas
ilkka P yysiäinen
1
2009

5. 1
Oxford University Press, Inc., publishes works that further
Oxford University’s objective of excellence
in research, scholarship, and education.
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Copyright © 2009 by Oxford University Press, Inc.
Published by Oxford University Press, Inc.
198 Madison Avenue, New York, New York 10016
www.oup.com
Oxford is a registered trademark of Oxford University Press
All rights reserved. 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 Oxford University Press.
Library of Congress Cataloging-in-Publication Data
Pyysiäinen, Ilkka.
Supernatural agents: why we believe in souls, gods, and buddhas
/ Ilkka Pyysiäinen.
p. cm.
Includes bibliographical references and index.
ISBN 978-0-19-538002-6
1. Psychology, Religious. 2. Cognitive science. 3. Soul.
4. Buddhas. 5. Gods. I. Title.
BL53.P99 2009
202'.1—dc22 2008039709
987654321
Printed in the United States of America
on acid-free paper

6. Acknowledgments
I had been working with the idea of this book for a couple of years
when Harvey Whitehouse kindly invited me to work as a visiting
researcher at the newly founded Center for Anthropology and Mind
at the Institute of Social and Cultural Anthropology at Oxford
University. I am very grateful for this unique opportunity. It was at
Oxford, in the summer of 2007, that the ideas of this book finally
began to take shape and different pieces of the puzzle started gradu-
ally to find their place. I am also hugely indebted to Justin Barrett,
Emma Cohen, Jon Lanman, and Harvey Whitehouse, as well as
Pascal Boyer, for help and advice during that summer. The seminars,
lectures, and informal discussions we had have been an important
source of inspiration.
I also want to thank all others who have in many ways helped
and inspired this project. This book would never have been written
had I not met in March 1995 with Tom Lawson, whose support,
advice, and friendship is much appreciated. Bob McCauley helped
me and my family move to Oxford. Thanks also to Barbara de Bruine
for kind help in many practical matters, as well as to the staff of the
Tylor Library.
Over the years, I have had numerous enjoyable discussions with
István Czachesz, Nicholas Gibson, Jeppe Jensen, Pierre Liénard,
Luther Martin, Joel Mort, Tom Ryba, Benson Saler, Jesper Sørensen,
and Don Wiebe; I am grateful to them for help, advice, and friend-
ship. This work was begun during my term as an Academy Research
Fellow ( project 1111683) at the Helsinki Collegium for Advanced
Studies. I am grateful both to the Academy of Finland that has

7. vi acknowledgments
funded my work and to the Collegium that has provided me with a first-class
academic environment for many years. Its past and present directors, Raimo
Väyrynen and Juha Sihvola, have supported my work with keen and much-
appreciated interest. I also thank the fellows of the Collegium and my colleagues
at its administration: Minna Franck, Hanna Pellinen, Marjut Salokannel, Taina
Seiro, Iiris Sinervuo, Maria Soukkio, Tuomas Tammilehto, and Aarno Villa.
Thanks also to my first teacher of comparative religion, Juha Pentikäinen,
who showed me the importance of not forgetting one’s own cultural heritage,
in my case the long traditions of Finnish scholarship on folk beliefs. I began
the study of religion together with Veikko Anttonen, now the godfather of my
daughter Milja. I would not be what I am without his help, support, and always
perceptive criticism of my most ambitious speculations. For many years I have
also had the pleasure of working closely with Kimmo Ketola and Tom Sjöblom
in Helsinki. I could never have done it all alone.
I have learned a lot from the members of the governing board of the Mind
Forum, a two-year series of multidisciplinary workshops on the study of mind
at Helsinki—Riitta Hari, Johannes Lehtonen, Kirsti Lonka, Anssi Peräkylä,
Stephan Salenius, Mikko Sams, and Petri Ylikoski—as well as all the partici-
pants in our seminars, too numerous to be listed here. Last but by no means
least, I thank my students and Ph.D. students who have at times worked very
hard in advancing the cognitive science of religion in Finland: Elisa Järnefelt,
Kaisa Maria Kouri, Jani Närhi, Outi Pohjanheimo, Mia Rikala, Aku Visala, espe-
cially, and many others. Thanks also to such important fellow scholars as Scott
Atran, Tamás Bíró, Joseph Bulbulia, Armin Geertz, Harry Halén, Sara Heinä-
maa, Timo Honkela, Markus Jokela, Jutta Jokiranta, Kai Kaila, Klaus Karttunen,
Simo Knuuttila, Hanna Kokko, Teija Kujala, Arto Laitinen, Jason Lavery, Mar-
jaana Lindeman, Dan Lloyd, Petri Luomanen, Brian Malley, Juri Mykkänen,
Matti Myllykoski, Markku Niemivirta, Asko Parpola, Anna Rotkirch, J. P. Roos,
Heikki Räisänen, Pertti Saariluoma, Risto Saarinen, Mikko Salmela, Stephen
Sanderson, Matti Sintonen, Jason Slone, Todd Tremlin, Reijo Työrinoja, Risto
Uro, and Ted Vial. The kind of interdisciplinary work that underlies this book
would not have been possible without many good contacts across disciplinary
boundaries. Finally, I am grateful to Cynthia Read and Oxford University Press
for accepting my book for publication. It has been a pleasure to work with
Meechal Hoffman, Martha Ramsey, and Mariana Templin. The two anonymous
reviewers made some good suggestions for improvements.
I am also happy that my family had the wonderful opportunity to enjoy a
summer at Oxford with me while I was finishing this book. Thanks for being
there, Sari, Miihkali, and Milja!

8. Preface
A book needs a good title. To the extent that Supernatural Agents
fulfills this purpose, it yet is nothing more than just a title. “Super-
natural” is here not a technical, explanatory concept and does not
necessarily identify any clearly demarcated phenomenon. I only use
it as catchy term to refer to the representations of nonhuman agents
described and analyzed in this book. These include ghosts and spirits
in Finno-Ugric and western European traditions, God within the
Christian traditions, and buddhas and bodhisattvas in the Buddhist
traditions. My aim is to show that the mental representation of God
and of buddhas are made possible by the same mental mechanisms
that are used in representing ourselves and our fellow human beings
as embodied agents (“souls”). God, buddhas, and human beings are
agents in the sense of animated organisms that have a mentality or
mind. Agency thus consists of the two properties of animacy ( liveli-
ness, self-propelledness) and mentality (beliefs and desires). Mentality
is subjectively experienced as an “I” or a “self” that inhabits the body;
usually we do not think of ourselves as simply identical with the
material body.
I draw from the cognitive science of religion, trying to show how
humans are able to represent agency in all its varieties and to draw
agentive inferences that feel natural and compelling, irrespective
of whether the agent in question can be directly perceived or not.
The cognitive science of religion emerged in the 1980s and 1990s in
religious studies and in anthropology and psychology of religion. It
has by and large focused on producing empirically testable hypoth-
eses about religious phenomena (Lawson and McCauley 1990; Boyer

9. viii preface
1994b, 2001, 2003a; Barrett 2000, 2004a; Pyysiäinen 2001b; Pyysiäinen and
Anttonen 2002; Whitehouse 2004; Tremlin 2006; Paiva 2007; Barrett 2007).
Not every scholar involved in this project is a natural scientist, and not every proj-
ect directly aims at empirical testing of theories, however (e.g., Martin 2004a,
2005; Pyysiäinen 2004d; Cohen 2007; Vial 2004; Sørensen 2007). It is equally
important to try to apply the findings of cognitive science in attempts at under-
standing and explaining religion as it is found in archaeological and historical
records and documents, doctrinal treatises, archival sources, and ethnographic
reports (e.g., Mithen 1998, 2004; Pearson 2002; Luomanen et al. 2007a). This
book is an example of this kind of approach to religious materials.
The first part of the book consists of an outline of the cognitive structures
that make it possible to mentally represent agents and agency; some method-
ological remarks; and important conceptual clarifications. In the second part,
I describe and analyze various kinds of conceptions of souls and spirits in folk
traditions, representations of the Christian God in their varieties, and repre-
sentations of supernatural agents in Buddhism. I try to show that conceptions
of souls and supernatural agents are cognitively natural and that even the most
abstract theological conceptualizations are elaborations of folk-psychological
notions. Analysis develops bottom-up, in the sense that I proceed from spirit
beliefs toward more abstract ideas about gods, buddhas, and bodhisattvas.
I am well aware that I am comparing traditions many scholars regard as
incommensurate: ancient Finno-Ugric conceptions of soul and Scholastic the-
ology, everyday Christian beliefs about God and Buddhist doctrine, and so forth.
The received view in anthropology has long been that beliefs and practices
receive their meaning in a cultural context and that cultural systems cannot be
understood from outside (e.g., Geertz 1973, 1993; cf. Boyer 1994b; D’Andrade
2000). The task of the anthropologist is to “read” particular cultures like texts
and to interpret them.
My own, essentially comparative project is based on the realization that
there are crossculturally recurrent patterns and that cultures do not have any
clear boundaries or an essence. As all comparison is always made from some
point of view and all similarities and dissimilarities are relative to that point of
view, there are many ways of comparing religious beliefs and practices (Boyer
1994b; Lawson 1996). As Tom Lawson (1996) points out, the problem of com-
parative religion has been precisely the lack of an independent measuring
stick that would make comparison possible; here I suggest that a comparative
exploration of religious traditions is possible by linking religious concepts and
ideas to the human cognitive architecture that channels the cultural spread of
ideas and beliefs. This is not to deny cultural or any other differences. To quote
the Oxford anthropologist Robert Ranulph Marett, “the special function of the
comparative method is to testify to a unity in difference, as in this case consti-
tuted by the human mind . . . amid an endless diversity of outer circumstance”
(Marett 1920, 81). Analyzing crossculturally recurrent patterns is not to deny

10. preface ix
differences in semantic contents but only to recognize important continuities
and invariances. It is mostly these continuities that I explore in the pages to
follow.
This book is intended for various kinds of audiences, and there are thus
various ways of reading it. An easy way to grasp the basic ideas will be to read
the summaries at the end of each chapter, and then to read the general conclu-
sion; the main text provides detailed justifications for the claims made, based
on extensive source materials.

11. This page intentionally left blank

12. Note on Transliteration
The transliteration of Sanskrit and Pali words has been simplified so
that only long vowels (such as ā), palatal ñ, and palatal ś are indicated
using diacritical marks, while cerebral (retroflex) consonants and
syllabic r, for example, are not indicated.

13. This page intentionally left blank

14. Contents
Note on Transliteration, xi
PART I Human Agency
1. Religion: Her Ideas of His Ideas of Their Ideas . . . , 3
2. Mind Your Heads, 43
PART II Supernatural Agency
3. Souls, Ghosts, and Shamans, 57
4. God as Supernatural Agent, 95
5. Buddhist Supernatural Agents, 137
6. Conclusion, 173
Appendix: Cognitive Processes and Explaining Religion, 189
Notes, 211
Sources and Literature, 229
Index, 281

15. This page intentionally left blank

16. I
Human Agency

17. This page intentionally left blank

18. 1
Religion
Her Ideas of His Ideas of Their Ideas . . .
1.1 Epidemiology of Intuitive and Reflective Ideas
My aim in this book is to show how knowledge about the cognitive
architecture of the human mind can help us understand the nature
and spread of beliefs about both human souls and spirits and super-
natural agents such as gods. Dan Sperber and others, in developing
what they call the epidemiology of representations, have focused on
the spread of concepts and beliefs (representations) (1985, 1996,
2006; Claidière and Sperber 2007; Sperber and Hirschfeld 2004).1
Epidemiology studies the ecological patterns of phenomena in
populations—in medical epidemiology, pathological diseases; in
psychological epidemiology, people’s beliefs and desires. As cul-
tures consist of a differential distribution of representations among
individuals, the psychological and the cultural are not so much
two different “levels” as measures of the spread of representations
(Sperber 2006; see also Pyysiäinen 2005c, in press). Culture thus
serves as a kind of “precipitation of cognition and communication
in a human population” (Sperber 1996, 97). I will discuss the ontol-
ogy of the mental and the cultural in section 1.2; here I only wish to
emphasize that my analysis by and large focuses on the psychological
mechanisms that constrain the spread of concepts of and beliefs
about supernatural agents (see Boyer 1994b).
Such constraining means that beliefs about supernatural agents
are not arbitrarily created in minds. The architecture of mind (see
Anderson and Lebiere 2003) instead shapes beliefs, thus creating
cross-culturally recurrent patterns. This implies that not all concepts

19. 4 human agency
and beliefs have an equal potential for becoming widespread. The most suc-
cessful representations in cultural selection are those that “match” people’s
mental architecture, in that an existing “slot ” corresponds to the form of the
representation in question. This makes it possible to connect the representa-
tion with information stored in one’s mind and to make inferences from it.
Representations such as “God is angry” are cognitively cheap, in that it does not
take much effort to make inferences from them, because human minds have
evolved to make inferences about other agents’ beliefs and desires in particular
(see Levinson and Jaisson 2006; Nichols and Stich 2003). “God is the ground
of being ” is a cognitively costly representation because it does not immediately
trigger any automatic inferential system in a person’s mind; it thus is more
difficult to spread. Sperber and Wilson’s (1988) relevance theory yields the fol-
lowing prediction:
(a) When the processing costs of two interpretations are equivalent, an
inferentially richer interpretation is favored
( b) When the inferential potential is the same, the less costly interpre-
tation is favored (Boyer 2003c, 352)
This means that cognitively costly representations are at a disadvantage in
cultural selection: they are either ignored or transformed into simpler forms
(Sperber 1996). To use Barrett ’s (1998, 611) example, “A cat that can never die,
has wings, is made of steel, experiences time backwards, lives underwater, and
speaks Russian” is a representation that cannot survive in oral transmission
without being distorted (see also Ward 1994). After a few retellings, a story
about this cat might be transformed into a story about an immortal cat, for
example. Memorizing the whole list of the cat ’s properties puts a heavy load
on working memory and thus leads to a simplification of the representation.
It also is not possible to memorize the representation of the bizarre cat as an
entity that belongs to some familiar category and then to deduce the various
properties of the cat one by one on the basis of this category membership (see
Boyer 2001; Sperber 2000b). There simply is no natural cognitive category
corresponding to this cat.
This argument only holds for orally transmitted representations. For exam-
ple, even though I had forgotten Barrett ’s exact characterization of the imag-
inary cat, I was able to check the list of the attributes in the journal article
where this bizarre feline lives just as Barrett once imagined it. Or not quite so,
because it is only the marks in ink that are preserved unchanged; the mental
representations these marks trigger in readers’ minds vary. We all imagine the
cat slightly differently. Verbal communication is not a process of copying ideas
from one mind to another, as it were. Instead, it is based on an active construc-
tion on the part of the reader or listener (Sperber 2000b; Sperber and Wilson
1988). Extremely complicated ideas (or instead their material tokens) can be
stored in books and on computers, but they are part of culture only insofar as

20. religion: her ideas of his ideas of their ideas . . . 5
they are part of cognitive causal chains connecting material causal processes
such as visual perception with semantic contents (Sperber 2006, 153 –61). In
other words, ideas in books are part of culture only on the condition that there
are persons who read the books and thus represent the ideas in their minds.
This challenges the traditional view of “world religions” and “religions of
the book ” as abstract entities that resemble Plato’s “ideas” (see chapter 2, sec-
tion 3). “Hinduism,” “Buddhism,” “Christianity,” or “Islam” do not exist some-
where outside of human minds which supposedly can only imperfectly reflect
the “true Hinduism,” “true Buddhism,” and so forth. Individuals can of course
have ideas about “true Hinduism,” but they are ideas in minds, not abstract
extramental entities. Several persons can also have “shared” ideas of a “true
religion,” in that their ideas more or less resemble each other, but not in the
sense of many minds somehow partaking in one and the same extramental
idea (chapter 2, section 1).
Naturally, persons can also have radically idiosyncratic (“heretic”) religious
ideas. Psychologist Justin Barrett (1999) coined the expression “theological cor-
rectness” ( TC) to refer to a continuum the endpoints of which consist of intui-
tive ideas easy to process in fast “on-line” reasoning and highly reflective ideas
that are cognitively costly and need institutional support (Barrett 1998; Barrett
and Keil 1996; Tremlin 2006, 75). While our everyday intuitions about reli-
gious concepts occupy the first end, the other end is typified by theological
and scientific ideas. As theology is normative, it is easy to understand TC as
referring to something that persons should believe although they do not (Slone
2004). However, this may be misleading because “heresies” are not necessarily
cognitively cheap and all officially accepted ideas are not necessarily cognitively
costly. Instead, I conceptualize the TC continuum on the basis of the nature
and amount of cognitive effort involved, without implying anything about the
normativeness of the ideas in question.
It is here that the discussion gets complicated; I present the more techni-
cal details in the appendix. I have used the so-called dual-process theories to
develop conceptual tools for describing the continuum of TC more broadly,
on the basis of cognitive effort involved, not on the basis of correctness versus
incorrectness. By so doing, I aim to accomplish two things: first, to provide a
richer and more adequate description of the representation of supernatural
agency in the traditions analyzed, and second, to be able to explain why and
how everyday concepts of supernatural agents develop to more reflective ones.
My analysis is cognitive, not historical. I try to show what kinds of cognitive
resources are needed for mentally representing supernatural agents and how
humans’ mental architecture channels the conceptual evolution of represen-
tations of supernatural agency. I focus on certain conceptual problems that
emerge in religious thinking about supernatural agents and try to show how
the ways these problems have been treated can be explained with reference to
humans’ mental architecture.

21. 6 human agency
All thinking naturally takes place in time and in social and cultural situa-
tions. I have chosen to focus on the cognitive features of agent representation
and have kept references to sociocultural environments to a minimum. I do so
not because cognition is somehow more important or essential but because a
single volume cannot address both. I focus my analysis on an interesting and
important part of what is going on in social and cultural reality. Sociality refers
to systems of more than one individual. Cognition is the link that connects
individuals in a social system and makes culture possible by stabilizing infor-
mation in interpersonal communication (see Jaisson 2006; Sperber 2006).
The human ability to attribute beliefs and desires to others and to think what
others might be thinking about is a cognitive capacity. Thus, although I focus
on the cognitive aspects of religious traditions, I also suggest some ways for
analyzing the noncognitive components of cognitive causal chains could com-
plete the analysis (see Sperber 2006, 161 –62).
To briefly summarize the argument in the appendix: humans use two dif-
ferent reasoning strategies that have been variously labeled intuitive and reflec-
tive, spontaneous and rational, and so forth. If we want to avoid misleading
connotations, we can just speak of the A- and B-systems (Pyysiäinen 2004c) or
systems 1 and 2 (Stanovich and West 2000). These systems can be differenti-
ated on the basis of the neural processes involved, the cognitive mechanisms
involved, and the kinds of contents processed. When cognitive scientists of
religion speak of “intuitions,” they mean certain kinds of basic concepts and
beliefs, usually without specifying the mechanisms that support them and help
differentiate them from nonintuitive ones (Barrett 2000, 2004b; Boyer 1994b,
2001, 2003b; Pyysiäinen 2001b; cf. Sperber 1997). One exception to this is
Tremlin, who discusses dual-process theories at some length (2006, 172–82).
Social psychologists have long known that humans seem to have two dif-
ferent reasoning strategies they apply in differing kinds of reasoning tasks, and
neuroscientists and cognitive scientists have made progress in differentiating
between two types of neurocognitive mechanisms supporting these strategies.
However, the kinds of phenomena that are typically studied within the social
sciences often involve both automatic and controlled processes (Bargh 1994, 3).
Thus, real-life religious concepts (as distinct from imaginary examples) can
hardly be completely intuitive or reflective. Instead, each type of process makes
a differential contribution to the cognitive processing of particular religious con-
cepts in particular contexts.
The two strategies can be roughly distinguished by such criteria as speed,
amount of emotion involved, type of motivation, type of information consulted,
form of reasoning employed, and amount of “extracranial” scaffolding needed
(see chapter 2, section 2.1). The intuitive A–system is responsible for fast, asso-
ciative, and emotionally colored thinking with purely practical goals, using
innate information and information derived from the environment through
analogical reasoning. It operates reflexively, not reflectively, drawing inferences

22. religion: her ideas of his ideas of their ideas . . . 7
and making predictions on the basis of temporal relations and similarity. It
employs knowledge derived from personal experience, concrete and generic
concepts, images, stereotypes, feature sets, and associative relations, relying
on similarity-based generalization and automatic processing. It serves such
cognitive functions as intuition, fantasy, creativity, imagination, visual recogni-
tion, and associative memory. Some authors also argue that it is a subsymbolic
pattern-recognition system that relies on connectionist, parallel distributed
processing.
The A-system thus supports what is often called “everyday thinking,” which,
in contrast to reflective or scientific thinking, proceeds from the immediate
experience of individuals; aims at short-term, practical efficacy, not at creating
general theories; seeks evidence and not counterevidence; makes use of indi-
vidual cases as evidence; personalizes values and ideals; makes use of abductive
inference; and presents arguments in the form of narratives (Denes-Raj and
Epstein 1994; Epstein 1990; Epstein and Pacini 1999; Epstein et al. 1992).
The reflective B-system serves such cognitive functions as deliberation,
explanation, formal analysis, and verification. It seeks logical, hierarchical, and
causal-mechanical structure in its environment, using information from lan-
guage, culture, and formal systems. It is a rule-based system capable of encod-
ing any information with a well-specified formal structure and relies heavily
on external memory stores such as books and pictorial representations. It thus
seems to work by computing digital information syntactically in some kind
of “language of thought ” (Fodor 1975; Paivio 1986; cf. Fodor 2000a). Highly
reflective B-system reasoning leans on rules, conventions, and databases that
often take time to learn and master. As it is not possible to hold all relevant
knowledge in mind and to consult it while performing mental operations,
external memory stores and “scaffolding ” are needed (Clark 1997). Thus, the
nature and amount of empirical information and institutional support avail-
able to a person also has an effect on her or his reasoning strategy (see Giere
2004; Giere and Moffatt 2003; see also chapter 2, section 2.1 here). Although
individuals differ from one another in their ability to understand complex ideas
and to engage in various forms of reasoning (Neisser et al. 1996; see also Miller
and Penke 2007), the distinction between intuitive and reflective reasoning
does not reduce to a difference in intelligence or between types of minds; it is
instead a difference in contexts and motivations (see Pyysiäinen 2004c). Reflec-
tive reasoning is used in different kinds of context and for different purposes
from intuitive reasoning.
There thus seem to be two different cognitive strategies based on different
kinds of cognitive mechanisms with a more or less distinct neural realization.
Reasoning about supernatural agents can be more or less intuitive or reflective,
not simply intuitive or reflective. The two systems can contribute in differing
degrees to inferences made. The relative cheapness or costliness of mental
representations thus is based on the relative amount of reflective (B-system)

23. 8 human agency
versus intuitive (A-system) processing. Consequently, representations are more
or less cheap or costly in terms of the cognitive resources needed, although
no simple way of quantifying cognitive costliness seems to be available ( but
see Barrett 2008; Sperber 2005). I use the term “costliness” in opposition to
the term “relevant ” (Sperber and Wilson 1988): processing a costly concept or
belief spends so much of one’s cognitive resources that persons are more or less
likely to interrupt the processing and try to find alternative means of achieving
the same benefits with a lower cost. As such, cognitive costliness is a relative
and not an absolute measure.
I substitute the more widely studied distinction between intuitive and
reflective thinking for the more narrow idea of TC. When I want to emphasize
that certain religious ideas are officially accepted doctrines, I speak of “pre-
scribed” ideas, not of “theologically correct ” ideas, in order to avoid confusion.
I do not mean that it is a defining characteristic of either prescribed or costly
ideas that people do not really “believe” in them (cf. Slone 2004). I mean that
the functions of costly ideas are different from those of more intuitive repre-
sentations (see Pyysiäinen 2004c).
When we study real-life religious traditions, we meet a whole spectrum of
representations from intuitive and idiosyncratic to prescribed and costly. We
encounter the former ones especially in interview materials and experimen-
tal data and the latter ones in theological treatises especially (see also Wiebe
1991). The challenging task of a scholar of religion is to somehow relate the
two to each other. Historically, the costly prescribed ideas seem to develop on
the basis of more intuitive ones (Pyysiäinen 1997, 2004d). When “theologians”
(as they might be called—of whatever tradition) try to solve problems of coher-
ence and consistency, they develop ever more abstract systems with no empiri-
cal constraints. As the price of the apparently increasing rationality of beliefs,
it becomes increasingly difficult to make folk-psychologically relevant infer-
ences from theological concepts (Boyer 2001, 320 –22). In my effort to describe
the human cognitive structure that affects some of the ways this development
unfolds, I will focus on everyday conceptions of souls and spirits, “Christian”
conceptions of God, and “Buddhist ” conceptions of supernatural agents. My
aim is to show that the concept of agency is the key to understanding the role of
representations of supernatural beings in human social systems. This means,
for one thing, a rethinking of the old dichotomy between “Tylorianism” and
“Durkheimianism” in anthropology of religion.
1.2 Strange Bedfellows: Tylor and Durkheim
Viewing religion as a system of beliefs and practices related to gods and
other supernatural beings is currently not, perhaps, the most popular way of
understanding religion. It is often argued that such an approach to religion is

24. religion: her ideas of his ideas of their ideas . . . 9
mistaken, ethnocentric, and obsolete. After all the linguistic, rhetorical, and
whatever “turns” the social sciences have gone through, it may seem fool-
hardy even to try to revive such an old-fashioned view of religion. The sup-
posed antagonism between the (neo-)Tylorian emphasis on spirit beliefs and
the Durkheimian emphasis on social functions continues to be an important
dividing line in the study of religion (see Pyysiäinen 2001b, 25– 76; Stocking
1995; Strathern and Stewart 2007). Philosopher Michael Levine (1998, 37)
has said: “given the dominant view in religious studies which sees religion as
a cultural system (Geertz 1973), it is a mistake to see gods as essential—even
when present in a system.” Nonetheless, beliefs about souls, spirits, ghosts,
gods and demons are found everywhere and seem to form a recurrent pat-
tern within and across cultures, quite irrespective of what philosophers and
theologians happen to consider important. Whether my colleagues regard
belief in supernatural agency as part of (true) “religion” or not, I argue that
beliefs about supernatural agents form a widespread phenomenon that can
and should be studied as an integral part of folk traditions (see Lawson and
McCauley 1990; McCauley and Lawson 2002, 13–16; see also Boyer 1994b,
19–34; Pyysiäinen 2001b, 1–5; Spiro 1968).
While Tylor emphasizes the content of religious beliefs ( belief in spirits),
Durkheim argues that it is the functions of beliefs that should define religion.
Underlying the Durkheimian view is the often implicit view that, although reli-
gious beliefs may testify to the “rich tapestry of human folly” (Boyer 2001, 2),
they are promulgated primarily to serve some important function in society.
But even if that is granted, it hardly means there cannot be additional rea-
sons for the persistence of religion (Hinde 1999). Even within the functionalist
frame of reference, we can still ask why precisely certain kinds of belief serve
a given function. Content and function may not have an arbitrary relationship,
and “Tylorianism” and “Durkheimianism” need not be regarded as mutually
exclusive (see Whitehouse and Laidlaw 2007). I argue that religious beliefs can
function in societies the way they do because “the social” is based on the notion
of agency (section 1.5.1 here).
Tylor (1903, 1:424–25), the first to lecture in anthropology at Oxford (begin-
ning in 1893), found religion in adoration of idols, sacrifice, or belief in a
supreme deity. For him, the minimum definition of religion was belief in
spiritual beings. Such belief seemed to appear among all “low races” of which
there was any evidence available. Although it was possible that humans had
lacked religion at some point in evolution, no evidence of this had survived.
Thus, it was better to admit that there was no way of deciding either way. Tylor
(1:425–27) called belief in spirits “animism”—a term he adopted from Georg
Stahl (1659–1734), who regarded animism as a scientific version of the idea
of the soul as the vital principle.2 Animism included two ideas that together
formed “one consistent doctrine”: first, individuals had souls capable of con-
tinued existence after death, and second, spirits formed a hierarchy of spiritual

25. 10 human agency
beings that were in control of events in the material world and human life both
here and hereafter. This doctrine led to the observed attitudes of reverence and
propitiation. Simple animism characterized tribes “very low in the scale of
humanity” but ascended to and persisted in “the midst of high modern cul-
ture,” in a deeply modified form but with unbroken continuity (see also Radin
1957, 192–253).
Spirit beliefs originated in the attempts of “primitive man” to explain two
things: what made the difference between a living and a dead body,3 and what
were the “human shapes” that appeared in dreams and visions? The “ancient
savage philosophers” inferred that every man had a life and a phantom. Both
were closely connected with the body: the life enabled the body to feel, think,
and act; the phantom was a representation of the live body in the mind. Both
could also be separated from the body. From this, the primitives concluded that
life and body belonged to each other, being manifestations of a single soul. As
they were thus united, the result was the idea of a ghost-soul that could animate
any physical form ( Tylor 1903, 1:428–29). The conception of a human soul
thus served as a model on which humans framed ideas of all kinds of spiritual
beings, “from tiniest elf that sports in the long grass up to the heavenly Creator
and Ruler of the world, the Great Spirit ” (2:110).
Tylor emphasized that the natives of Australia, for example, were “a race
with minds saturated with the most vivid belief in souls, demons, and dei-
ties.” The same held for African and American natives: despite the lack of
organized religion, these peoples had beliefs about supernatural beings. Tylor
(1903, 1:423) therefore reproached Sir Samuel Baker for having ignored pub-
lished evidence when he argued at the Ethnological Society of London in
1866 that the most northern tribes of the White Nile were without a belief
in a Supreme Being, had no worship, and lacked “even a ray of supersti-
tion.”4 Likewise, J. D. Lang had denied the existence of any religion among
the aboriginals of Australia but reported them to have beliefs about evil spirits
as well as of divinities. In the same vein, W. Ridley had reported that the
aboriginals of Australia had traditions about supernatural beings but denied
true religion to them ( Tylor 1903, 1:418–19). Tylor ’s basic idea was that all
spirit beliefs and their related practices hid some real meaning, no matter
how “crude and childish”; the explanation was that a reasonable thought had
once given life to practices that now looked like “superstitious folly.” Even
Christianity was somehow related to ideas that ran back to the very origins
of human civilization (1:421). “Civilized men,” however, found it extremely
difficult to animate nature. Such beliefs were only “survivals” from the dawn
of humanity. This idea of survivals was related to Tylor ’s erroneous concep-
tion that some peoples had a longer evolutionary history than others and thus
were “more evolved” or more advanced. We realize today that it makes no
sense in evolutionary terms to imply that some peoples have traveled faster
in evolution; evolution has no predetermined general goal, and many cultural

26. religion: her ideas of his ideas of their ideas . . . 11
developments are merely constrained by biology, not directly driven by it (see
Korthof 2004; Richerson and Boyd 2005).
By survivals, Tylor meant old customs and beliefs that survived in later
stages of cultural development because of mere ancestral authority (1903,
1:70 – 71). Superstitions, for example, were survivals, though not all survivals
were superstitions (children’s games, for example). This idea only makes sense
against the backdrop of Tylor ’s and others’ ideas of the time about a hierarchi-
cal scala naturae and a predetermined progress toward what was considered to
be higher forms of culture. To be a survival, it was not enough for an idea or a
practice to be old; it had to be somehow misplaced in its current context. For
Tylor, a sure sign of such misplacement was that the survival seemed somehow
foolish or erroneous by his lights; it had been rational in its original context
but had become obsolete in the modern context because of cultural progress.
Such an idea of a unilinear religious development has nothing to do with the
biological theory of evolution, however. When Tylor (1:vii) gave formal credit
to Darwin and Spencer, granting that his own work was arranged on its own
lines and came scarcely into any detailed contact with the work of “these emi-
nent philosophers,” he had in mind Spencer ’s social Darwinism more than
the evolutionary theory of natural selection operating on random variation (see
Sharpe 1975, 53–58).
Durkheim’s view of religion has often been regarded as a welcome alter-
native to Tylor ’s (see Anttonen 1992, 1996a,b; Douglas 1970, 1984; Stocking
1995, 82–83, 97–98, 162 –63, 361). Although Durkheim also was interested in
the most primitive form of religion, he focused more on structure than on
development. Notwithstanding some recent claims to the contrary, he was neg-
ative about the evolutionary theory of his time because it seemed to threaten
the autonomy of his new discipline of sociology (see Gans 2000; Schmaus
2003).5 Durkheim’s Les règles de la méthode sociologique was a polemical text;
even he himself did not follow too strictly its principle of explaining “the
social” only by the social (as opposed to the psychological). In Le suicide, for
example, the social consists of mental representations, and psychological argu-
ments are even more prominent in Les formes élémentaires de la vie religieuse
(Lukes 1977, 227–28; Pyysiäinen 2001b, 63– 70, 2005d).6
Durkheim’s basic idea was that religion should be defined only as a sym-
bolic expression of the reverence persons had toward the group ( la société ).
Although science could help reveal the psychological mechanisms behind
apparently meaningless ritual acts, this was not an explanation of the ritual.
Only the participants’ belief in the power of the ritual to raise spiritual forces
could be psychologically explained; the objective value of the ritual was based
on the fact that the group was regenerated by it. This fact called for a sociologi-
cal explanation (1925, 496–97, 1965, 389–90). Whether supernatural agents
were involved or not was quite inconsequential; actually, the primitives did not
even have the idea of the supernatural—because they had no idea of the natural

27. 12 human agency
(1925, 34, 118–22, 1965, 39–40; cf. Arbman 1939, 27). I have elsewhere tried
to show in detail that here Durkheim errs: all people all over the world have
intuitive expectations about the ordinary course of events, together with ideas
that violate these intuitive expectations (Geary 2005; Pyysiäinen 2001b, 55– 74,
2004d, 39–52, 81–89, 2005d). Moreover, the “regeneration” of the group is
a cognitive act, consisting of persons’ mental representations of other per-
sons’ mental representations; what he called “collective consciousness” thus is
shared knowledge and can be studied from the cognitive psychological perspec-
tive (see Pyysiäinen 2005f ).
Belief in supernatural agents forms an obvious recurrent pattern in vari-
ous cultural traditions. As some of these beliefs help people express, process,
and justify their central values and norms, we can try to combine the Tylo-
rian and Durkheimian legacies in explaining how beliefs about supernatural
agents are used in organizing a society (see Boyer 2000b, 2002; Pyysiäinen
2005f ). Typical examples of supernatural agents are gods, spirits, ghosts, angels,
demons, and so forth. The category of supernatural agents may seem so het-
erogeneous that one may well ask whether it is a genuine category at all. Some
might want to claim that it is a pseudocategory just like “mysticism,” “ritual,”
or even “religion” itself (see, e.g., Fitzgerald 1996, 1997; Penner 1983). I think
this view is overly pessimistic. If we can find some theoretical depth to the ideas
of agency and counterintuitiveness, the idea of supernatural agents might be
operationalized for research. We need a theoretical frame of reference in which
the various emic distinctions between different kinds of supernatural agents are
seen as subdivisions within the general, etic category of supernatural agency.
Using concepts at a mediating level between cultural particulars and the gen-
eral category of supernatural agents, we may then conceptualize the various
recurrent patterns within that category.
1.3 There Must Be Somebody Out There
I use quite consciously the expression “supernatural agents” instead of “super-
natural beings.” By agents, psychologists and philosophers mean organisms
whose behavior can be successfully predicted by postulating conscious beliefs
and desires (Dennett 1993, 15–17) or entities whose behavior is caused by their
mental states (Bechtel 2008, xii). I use the concept in the sense of an organ-
ism to which animacy (liveliness, self-propelledness) and mentality (beliefs and
desires) are (correctly or incorrectly) attributed.
I follow Barrett (2008) in distinguishing between two components of
agency: animacy and mentality. The domain of animacy is characterized
by such things as self-propelledness and goal-orientation: organisms tacitly
postulate animacy for other organisms when they feel that those organisms
move by themselves and seem to be moving toward some goal (see Boyer and

28. religion: her ideas of his ideas of their ideas . . . 13
Barrett 2005). When one also attributes to the organism a conscious intention
and begins to simulate its mental states or to try to theorize about its beliefs
and desires, one moves on to “mentalizing ” or “mind reading ” (Nichols and
Stich 2003). This ability is often called a “theory of mind” (ToM), in the sense of
folk-psychological theories about other minds (Carruthers and Smith 1996).7
I distinguish three overlapping cognitive mechanisms that contribute
to agentive reasoning. The first is hyperactive agent detection (HAD, Barrett
2000): the tendency to postulate animacy—this mechanism is triggered by
cues that are so minimal that it often produces false positives, for example, we
see faces in the clouds, mistake shadows for persons, and so forth. Second is
hyperactive understanding of intentionality (HUI): the tendency to postulate
mentality and to see events as intentionally caused even in the absence of a visi-
ble agent. Third is hyperactive teleofunctional reasoning (HTR): the tendency
to see objects as existing for a purpose.
1.3.1 Agency
Barrett (2000) coined the acronym HADD ( hyperactive agent detection device)
to refer to the cognitive processes that help us recognize agents (actually, ani-
macy) and distinguish them from nonagents. This mental “device” is hyperactive
or hypersensitive, in that under certain conditions, it is triggered by very mini-
mal cues.8 We see faces in the clouds and detect predators in rustling bushes
because such ambiguous perceptions easily trigger the postulation of agency.
According to Barrett, a normally functioning HADD is hyperactive by its very
nature—hyperactivity is not something exceptional. From an evolutionary point
of view, this is plausible, insofar as the costs of false positives that an overreact-
ing detector produces are lower than the benefits it brings (Atran 2006).
The following most common direct cues for agency have been suggested
(Blakemoore et al. 2003; Boyer and Barrett 2005; see Heider and Simmel 1944):
1. Animate motion that has as its input such things as nonlinear changes
in direction, sudden acceleration without collision, and change of
physical shape that accompanies motion (e.g., caterpillar-like crawling)
2. An object reacting at a distance
3. Trajectories that only make sense if the moving entity is trying to reach
or avoid something, which leads to goal-ascription
4. An entity appearing to be moving by conscious intention to an
apparent end result (intention-ascription)
5. The experience of joint attention (for which we develop a capacity
between nine and twelve months of age)
These cues trigger the feeling or intuition of agency spontaneously and auto-
matically, in the sense that this intuition can neither be rationally controlled
nor initiated or terminated at will (see Bargh 1994; Pyysiäinen 2004c; see

29. 14 human agency
the appendix). However, I suggest that the first three cues only trigger ani-
macy assumptions, while the latter two trigger HUI. It may also be misleading
to call HADD a (single) “device,” as several systems contribute to the percep-
tion that one is facing an agent (see Boyer and Barrett 2005). Once people
start reasoning about the beliefs and desires of a postulated agent, they use
mind reading: the capacity to make inferences about the beliefs and desires of
others and to explain their behavior on that basis (see Carruthers and Smith
1996; Frith and Frith 2005; Nichols and Stich 2003; Premack and Woodruff
1978; Saxe and Baron-Cohen 2006; Tomasello et al. 2005; Tremlin 2006, 75).
Mind reading is often understood to be innate—not a learned ability but rather
something triggered in the course of normal development (Leslie 1996; Well-
man and Miller 2006, 28). It has been seen as an innate modular algorithm,
(e.g. Leslie), as an innate body of knowledge (e.g. Pinker and Sperber), and as
modular in all three senses of the concept of a “module” (Baron-Cohen) dis-
cussed in the appendix (see Gerrans 2002, 308). It can also be understood to
be based on a nonmodular conceptual competence ( Wellman et al. 2001; see
Yazdi et al. 2006).
The standard psychological test for ToM is the so-called false belief task
( Wimmer and Perner 1983). In this test, children are shown a sketch in which
a boy called Maxi first puts a chocolate bar in container A. Then, without Maxi’s
knowing, his mother moves it to container B. The subjects are then asked
where Maxi would look for the chocolate bar when he returns. Children around
three years of age tend to say (incorrectly) that Maxi would look in container B
(where the bar was hidden). Only in the fourth year does the tendency appear
to say that Maxi would look in container A. It thus seems that younger chil-
dren cannot understand that Maxi remains ignorant of his mother hiding the
bar in container B. These results have been obtained in many replications of
the experiment and across many variations of the original design. Scholarly
opinions differ on what these results really tell about the ways children think
(see Baron-Cohen 2000; Baron-Cohen et al. 1985; Fodor 1992; Perner 1995;
Wimmer and Perner 1983).9
Simon Baron-Cohen, Alan Leslie, and Uta Frith (1985) used the same task
to test autistic children’ ability to impute mental states to others. They tested
twenty autistic children, fourteen children with Down syndrome, and twenty-
seven clinically normal preschool children. Two dolls were used, Sally and
Anne. First, Sally placed a marble in her basket, and then left the scene. Anne
then transferred the marble into her own box. When Sally returned, the experi-
menter asked: “Where will Sally look for her marble?” Children with Down
syndrome as well as normal preschool children answered correctly by point-
ing to the basket where Sally had put the marble. The autistic group consis-
tently answered by pointing to the box where the marble really was; the autistic
children did not seem to get the difference between their own and the doll’s
knowledge. In the authors’ words, they “failed to employ a ToM”—a failure

30. religion: her ideas of his ideas of their ideas . . . 15
consisting of an “inability to represent mental states” (Baron-Cohen et al. 1985,
43). Autism thus might seem to involve a lack of a ToM. The inability to make
inferences about what other people believe to be the case in a given situation
prevents one from predicting what they will do (Baron-Cohen 2000; Baron-
Cohen et al. 1985, 39; see Carruthers and Smith 1996, 223– 73; Frith 2001).
However, failure in the test does not necessarily indicate a lack of a ToM,
because having a ToM does not necessarily require the ability to reason about
false beliefs (Bloom and German 2000; see Wellman et al. 2001). Reasoning
about false beliefs is just one way of using ToM (see Stone 2005). Philip Ger-
rans and Valerie Stone also argue that there is no hardwired ToM as a dedicated
module at all and that belief attribution is not supported by a domain-specific,
modular mechanism (see appendix). Instead there is interaction between such
low-level, domain-specific mechanisms as tracking gaze and bodily movement,
joint attention, and so forth and higher-level domain-general mechanisms for
metarepresentation, recursion, and executive function. The output of the low-
level mechanisms serve as input for the higher-level, domain-general processes
(Gerrans 2002, 306, 311; Stone and Gerrans 2006a,b).
Perner et al. (2006), as well as Saxe et al. (2006) present neuroscientific
evidence for the view that in the right and left temporo-parietal junction of the
brain there is a specialized, domain-specific neural mechanism for reasoning
about beliefs. Inhibitory control of mental contents, together with response
selection, is mediated by domain-general mechanisms, while the domain-
specific mechanism of the temporo-parietal junction is only recruited for
processing beliefs. And reasoning about beliefs is faster than following domain-
general rules in reasoning about other things than beliefs (Saxe et al. 2006; see
Saxe and Baron-Cohen 2006).
Leslie et al. (2004, 2005) argue that ToM is a partly modular learning
“mechanism” that only “kick-starts” the attribution of beliefs and desires. Just
as color vision provides us with color concepts, ToM introduces belief and desire
concepts. Reasoning about the contents of beliefs takes place through a selec-
tion process (SP) with inhibition. Developing slowly from the preschool period
onward, SP acts by selecting a content for an agent ’s belief and an action for the
agent ’s desire. In everyday thinking, taking a belief to be true is the default, so
success in the false belief task requires an ability to inhibit the default attribu-
tion and to select the erroneous belief for the character looking for the hidden
object.
The mentality of organisms that ToM processes is very early on under-
stood as separate from the physical body. Kuhlmeier et al. (2004), for example,
show that five-month-old infants apply the constraint of continuous motion to
inanimate blocks but not to persons. They thus do not seem to view human
agents as material objects. In the experiment, twenty infants watched a film in
which a woman was standing on a stage that contained two large red screens
separated by 1.21 meters. The woman then went behind the first screen; her

31. 16 human agency
identical twin sister in identical clothing had been hiding behind the second
screen and now emerged from behind it. What the infants saw was a woman
moving behind a screen and then suddenly emerging from behind the other
screen without being visible in the space between the two screens. The infants
revealed that they were not surprised by this apparently miraculous event,
although they were surprised when they watched mere physical objects (as dis-
tinct from humans) behaving in the same way. ( They exhibited surprise by
looking longer at the action they were witnessing.) This suggests that they do
not consider agency to be constrained by the physical body.
Even as adults, people do not feel themselves to be bodies but instead feel
that they occupy bodies (Bloom 2005, 191, 2007; Merleau-Ponty 1992; see
Bronkhorst 2001, 402). Routinely, people spontaneously attribute their own
agentive properties to the homunculus called the “self,” which is not merely
the sum of a set of lower-level, “dumb[er]” homunculi (Nichols and Stich 2006,
9–10). To the extent that the agentive properties are detached from the physical
body, it is perfectly natural for people to have intuitive beliefs about disembod-
ied agents such as spirits. However, disembodiment does not mean a complete
lack of bodily form; mentality may instead be attributed to various forms of
“subtle” or otherwise nonstandard bodily forms.
1.3.2 Intentionality
There are also indirect cues that lead people to interpret an event as inten-
tionally caused even when they do not perceive an agent. Hyperactive under-
standing of intentionality produces an automatic conclusion that an event or
structure must have been caused or designed by an intelligent agent, even
when no trace of an agent is evident (Barrett 2004b, 34). For example, nor-
mally, when one sees an artifact, one knows that it must have been designed
and made by an agent, even though this agent is not present and one does not
know her or his identity (see Czachesz 2007, 86).
This kind of reasoning contributes to beliefs about supernatural agency
when it is extended to such cases as, for instance, seeing the image of Jesus
falling off the wall right at the moment when someone says something blas-
phemous. The coincidence of the words uttered and the picture falling creates
the feeling that the falling was intentionally caused, despite the fact that no one
is present and one has no idea of the mechanism by which an agent could have
caused the picture to fall (see Atran 2002a, 59 –63; Bering 2006; Bering and
Parker 2006). Bering (2003) suggests that there may even be a specific cogni-
tive mechanism devoted to processing apparently intentional events (“existen-
tial meaning”) in the absence of a physical agent.
Such inferences are fast and automatic intuitions (see the appendix).
The low-level intuitions about agency are dedicated to combining perceived
movement with the perceiver ’s reflective ideas of agency and free will. When

32. religion: her ideas of his ideas of their ideas . . . 17
action consistently follows prior thought and when apparent causes of action
other than somebody’s thought are excluded, we have the experience of voli-
tional action. Volition, in turn, presupposes an agent. Torsten Nielsen (1963)
showed in the 1960s that when subjects were asked to put one hand in a box
and draw a straight line with it while looking into the box through a tube, they
could be fooled into believing that the experimenter ’s hand seen in the box
through a mirror was their own (which they did not see). When both hands drew
a straight line, the subjects perceived the alien hand as their own. When the
experimenter ’s hand drew a curve to the right, the subjects still experienced the
hand as their own, and now perceived it as making involuntary movements.
Similarly, Ramachandran and colleagues arranged an experiment in which
a subject with an amputated arm was asked to put his real arm and what he
experienced as his phantom limb in a box with a vertical mirror in the middle;
the reflection of the real arm then appeared in the mirror where the phantom
would have been if it were real. When the subject moved his real arm, he felt
he was moving his phantom limb. Then the experimenter put his arm inside
the box so that it appeared in the place of the phantom. Observing his real arm
and the experimenter ’s arm in the box made the subject feel as though his
phantom arm was moving (Ramachandran and Blakeslee 1999, 46–48; see
Wegner 2002, 40 –44).
Wegner (2002, 44) thus argues that one can think movement is intentional
by watching any body move where one thinks one’s own body is. But there are
also cases in which one perceives a movement as guided by an intentional will
not one’s own. When the picture of Jesus Christ suddenly falls off the wall, the
default explanation is not necessarily that one did it oneself (although this is
possible). One can come to think that the movement is intentional by combin-
ing one’s perception of it with a representation of an invisible agent.
Wegner (2002, 44) suggests in passing that the experience of another ’s
movement as willful is mediated by the so-called mirror neuron system in the
brain. This system mediates such intuitive reasoning as, for example, feeling
disgust when witnessing someone drink a glass of milk with a face contracting
in an expression of disgust. It cannot mediate such reflective ToM reasoning as
in, for instance, trying to figure out what gift would please a foreign colleague
(Bargh 1994, 3; Keysers and Gazzola 2007). Mirror neurons were first found
in monkey brains in premotor area F5 and parietal area 7b. They are called
mirror neurons because they are activated during both the execution of pur-
poseful, goal-related hand actions (grasping etc.) and the observation of simi-
lar actions performed by conspecifics (see Gallese et al. 1996; Rizzolatti et al.
1988; Rizzolatti et al. 2000). A mirror neuron system has also been found in
the human brain (although no special type of neuron seems to be involved).
Implicit, automatic, and unconscious simulation processes establish a link
between an observed agent and the observing agent, thus making imitation
possible (Bremmer et al. 2001; Gallese and Metzinger 2003; Hari et al. 1998).

33. 18 human agency
It has been suggested that social cognition is based on the mirror neuron
system, which mediates observations about other agents’ apparently intentional
behavior (see Hurley 2008a). When two agents interact socially, the mirror
neuron system is activated and creates a shared neural representation (Becchio
et al. 2006, 66 –67). It thus helps people understand the intentions of others
and to relate them to their own (see Gallese and Goldman 1998; Gallese and
Metzinger 2003; Gallese et al. 2004). In being restricted to motor processes
and intuitions, though, the mirror neuron theory alone may not be sufficient
to account for social cognition (see Bargh 1994; Hurley 2008b; Keysers and
Gazzola 2007). Social cognition also involves complicated reflective processes
related to understanding of intentionality.
Recognizing consciously goal-directed action involves recognizing inten-
tionality in three senses: intention recognition, attribution of intention to its
author (HUI), and understanding motivation behind the intention (HUI, HTR)
(see Becchio et al. 2006). Intentionality as the mark of the mental is usually
understood as “directedness” or “aboutness” (Brentano 1924; Dennett 1993, 67,
1997, 66; Husserl 1950a,b). Mental states are beliefs about the world, desires
for things, and so forth ( Wellman and Miller 2006, 34–35). Following Jaakko
Hintikka (1975), I argue that it might be better understood as intensionality,
however (see the appendix, section 2.1).
Intensionality relates to such things as meanings, properties, property-
based relations, and propositions. When we describe a set intensionally, we
list the properties the members of the set have. We may, for instance, try to list
the various gods of the religions of world using the following formula: {x:x is
a god}. An extensional listing, in contrast, is made by just listing the members:
{Allah, Quetzalcoatl . . . Zeus}. These two lists are extensionally equivalent
but intensionally distinct. This distinction is based on the fact that terms and
expressions have an intension or meaning (Sinn) and an extension or reference
(Bedeutung ) (Frege 1966). For example, someone’s beliefs about the Evening
Star are not necessarily beliefs about the Morning Star, because the person in
question may not know that these two are the same planet.10 The terms have
different intentions but the same reference. Mere extensional concepts are not
enough for conceptualizing agency because agency is defined by the mental act
of directedness (instead of mere reactivity, Leslie 1994).
The idea of intentionality as intensionality brings to the fore what seems to
be crucial in intentionality in a psychological sense. According to Hintikka, “a
concept is intentional if and only if it involves the [logically] simultaneous con-
sideration of several possible states of affairs or courses of events.” Intentional-
ity as intensionality thus means that for an act to be intentional, it must involve
a conscious choice made between several alternatives (1975, 195; see also 212–13,
1982). Intentionality as intensionality means that intentional systems involve
the consideration of possible worlds as logically alternative scenarios (2006a,
21–24, 2006b, 557–58; see Dennett 1993, 122, 174– 75). Thus, ToM depends

34. religion: her ideas of his ideas of their ideas . . . 19
on the ability for counterfactual reasoning about contrastive states of affairs
or courses of events (see also Buller 2005a, 194). Directedness is involved, but
in an intensional sense: a mental state is directed toward something, but with
the awareness that it might just as well be directed toward something else. In
intentionality, as distinct from mere reactions, a course of action is selected
from among multiple mentally represented alternatives.
If mind reading or ToM operates through a selection process with inhi-
bition, understanding intentionality as intensionality fits it much better than
the idea of intentionality as simply directedness. For example, understanding
another ’s beliefs and desires in the false belief task involves precisely consid-
eration of possible worlds as logically alternative scenarios. This entails under-
standing that the mind is a representational device and that there is no reason
that all represented propositions should be true (Perner 1993). One has to be
able to metarepresent another ’s beliefs, that is, to embed them in one’s own
beliefs, as in “I believe that ‘she thinks that he is wrong.’ ” Valerie Stone (2005)
argues that it is precisely the ability of metarepresentation that enables children
to succeed at explicit false belief tasks: the child must be able to understand
that various agents represent the location of the chocolate bar in various ways.
According to Bering, it is around six or seven years of age that children become
capable of understanding such third-order intentionality in which “he thinks
that she believes that he wants,” for example. At that point in their develop-
ment, they are also able to regard random events as symbolic and declarative of
a supernatural agent ’s mental states (Bering and Johnson 2005, 134–36).11
Degrees or orders of intentionality is an idea of Dennett (1993, 243–46). It
has been argued that normal adult humans are capable of fourth- or fifth-order
intentionality (when no external memory stores are used) (Dunbar 2003, 170,
2006, 171– 72). “I know12 that John wants Mary to understand that Bill believes that
Linda loves him” is an example of fourth-order intentionality. Such metarepre-
sentation of intentional attitudes requires a capacity to understand recursive13
structures (Stone 2005; see Hintikka 1975). To the extent that we are interested
in how ToM works in social cognition, only embedded mental states count, not
mere embedded observed facts. For example, the sentence “John says Mary to
claim that Bill wrote her that Linda sent him a love letter ” is not a case of degrees
of intentionality. Even an autistic person incapable of mind reading might be
able to understand this sentence as reporting four interrelated facts.
The recursive embedding of metarepresentations seems to be an exclu-
sively human capacity; apes, for example, do not understand recursive struc-
tures, though the evidence here is somewhat ambiguous (Dennett 2006, 111;
Dunbar 2003, 170). Premack and Premack (2003, 149–53), however, claim
that chimpanzees are only capable of first-order intention ascription and that
even this requires much conscious effort from the chimp, whereas a ten- to
twelve-month-old human infant does this spontaneously. Human social cogni-
tion starts where the chimpanzee’s cognitive ability ends. As Tomasello and

35. 20 human agency
Carpenter (2007, 122) observe, “from a very early age human infants are moti-
vated to simply share interest and attention with others in a way that our nearest
primate relatives are not.” Joint attention, which is one of the low-level inputs
of ToM, starts to develop in human infants as early as nine months, when the
learning child begins to look at an object and at the parent, trying to draw the
adult ’s attention to a shared object (see Moore and D’Entremot 2001; Tomasello
and Carpenter 2007; Tomasello et al. 2005).
Sperber (1997) argues that recursively metarepresenting a belief within
another belief can provide a validating context for the embedded belief. It is
then accepted as true because of certain second-order beliefs about it (Boyer
1994b, 120; Sperber 1996, 69– 70, 89–97). To the extent that these second-
order beliefs seem compelling, the first-order belief embedded in them is also
plausible.14 In, for example, “ ‘Jesus is our redeemer ’ is true because it says so
in the Bible,” “Jesus is our redeemer ” derives its plausibility from the validating
embedding “It says so in the Bible” (see Pyysiäinen 2003c).
Metarepresentation thus makes it possible to make inferences about prop-
ositions with an undecided truth-value—as in, for instance, “ ‘If it was John
who stole my wallet,’ then he should be arrested.” Here the judgment of John
being guilty must be temporally suspended in order to make a provisional
inference (see Cosmides and Tooby 2000, 59 –60; Pyysiäinen 2003c; Sperber
1996, 71– 73). It is important to distinguish between what Sperber calls “half-
understood” propositions and propositions with an undecided truth-value. It is
possible to generate conditional inferences from premises with an undecided
truth-value, but it is not possible to do so from incomprehensible premises.
These can only be used as quotations that receive a meaning from the validat-
ing context in which they are metarepresented.
Failure to understand the nature of metarepresentation leads to accepting
all beliefs as one’s own (see Sperber 1994, 2000a; Pyysiäinen 2003c). A psy-
chiatrist may, for example, entertain the metarepresentation “Joe believes ‘I am
Jesus.’ ” If she is not able to use the metarepresentation “Joe believes,” then
“I am Jesus” becomes her own belief (see Corcoran et al. 1995; Frith 1995). Meta-
represented beliefs are decoupled from reality, and their semantic relations are
suspended, in the sense that one cannot directly reason from “Ann believes
‘John is a spy’ ” that John really is a spy (Cosmides and Tooby 2000). This would
only follow in the case that the metarepresentational context is automatically
validating. For a Christian believer, the metarepresentation “It says so in the
Bible” often is such a validating context, for example (Pyysiäinen 2003c). It is
beliefs about beliefs that validate beliefs.
1.3.3 Teleofunctional Reasoning
Humans are prone to see things as existing for a reason or purpose. This ten-
dency to view entities as existing for a purpose may derive from children’s early

36. religion: her ideas of his ideas of their ideas . . . 21
emerging ability to think that there are hidden intentions behind everything
and to interpret agents’ behavior as goal-directed (Johnson 2000, 188, 208;
Kelemen and DiYanni 2005, 6). Children’s understanding of how agents use
objects as means to achieve goals may provoke a rudimentary teleofunctional
view of entities: agents’ intentions are understood as being intrinsic properties
of the objects themselves (Kelemen 1999a,b; Kelemen and DiYanni 2005, 6).
There is experimental evidence for a “promiscuous teleology”: British and
American elementary schoolchildren are prone to generating teleofunctional
explanations of the origins and nature of living and nonliving natural entities
and endorsing intelligent design as the source of both animals and artifacts
(Evans 2000, 2001; Kelemen 2004; Kelemen and DiYanni 2005).
Kelemen and DiYanni (2005, 7) point out that this view about purpose does
not necessarily derive from agentive intuitions. Teleo-functional reasoning may
also be an independent characteristic of causal reasoning. However, Asher and
Kemler Nelson (2008) provide evidence for the interpretation that three– and
four–year-old children can adopt the intentional stance (Dennett 1993) and do
understand the true functions of artifacts to be the designed functions. It thus
is also possible that intuitions about purpose and design and intuitions about
agency derive from a common source.
Lewis Wolpert (2007, 27–33, 67–82) argues that all causal reasoning in
humans derives from the manufacturing and using of tools that typify the spe-
cies, not from social interaction (cf. Dunbar 1993, 2002, 2003, 2006). The
making of first tools created a new kind of selection pressure, to which causal
thinking is an adaptation: the making of tools required an ability to understand
the ideas of means and ends and of causality and intentionality. Although non-
human primates can distinguish the animate from the inanimate, they do not
view the world in terms of intermediate and often hidden underlying causes,
reasons, intentions, and explanations. ( They understand animacy but not
mentality.)
Wolpert neither develops this idea carefully nor contrasts it in detail with
competing accounts. It does not have to be incompatible with, for example,
Dunbar ’s view of human sociality as having provided the selection pressure for
the large neocortex of our species. Tool use and social interaction are not two
mutually exclusive phenomena but seem to have developed together. Thus, we
can see Wolpert ’s idea as an important addition or qualification to other argu-
ments that may be able to help explain beliefs about supernatural agents.
The hyperactivity in HADD, HUI, and HTR means that they produce many
false positives: people perceive agents where there are none, ascribe intentional-
ity to events and structures that are purely mechanical, and use teleofunctional
reasoning to explain the natural world. As natural-born tool users, humans see
intelligent agency and design everywhere. As humans are the prototype of an
agent, humans attribute at least some humanlike features also to the agents
HADD postulates (Boyer 1996c; see Guthrie 1993; Richert and Barrett 2005).

37. 22 human agency
However, supernatural agent concepts are not derived from these false
positives. First, ambiguous perception and overextended attribution of inten-
tionality and design in themselves do not contain enough information for
forming a persisting agent concept. Second, the supernatural agent concepts of
religious traditions are abstractions from a large number of individual mental
representations that are communicated among persons. Therefore, we can-
not explain these concepts only by referring to ambiguous individual percep-
tion; we have to explain why they have become widespread in populations (see
Barrett 1998, 617, 2004b, 41, 43; Boyer 1994a,b).
The notions of HADD, HUI, and HTR can help explain why certain kinds
of supernatural agent concepts are easier to adopt than others; this ease then
explains, partly at least, why these concepts are found all over the world both
in ancient and modern times. Because there is a natural place for these con-
cepts in the human mind and in human practices, they are contagious and
have become widespread (Barrett 2004b, 33–39). Mind and culture thus are
not so much two different levels as the endpoints on a scale from individual
to public and “shared” (see Sperber 2006). The folklore scholar Lauri Honko
(1962, 93–99) tried to explain the actualization of traditional beliefs in casual
encounters with the spirits by referring to the psychology of perception; we can
now try to do much the same with the help of cognitive and developmental psy-
chology and evolutionary theory. That effort may turn out to have far-reaching
consequences for the study of religion and culture. The next thing for me to do
with this purpose in mind is to provide some criteria for calling some concepts
counterintuitive.
1.4 Out of the Ordinary
1.4.1 Intuitive Ontology and Counterintuitiveness
Animacy, mentality, and teleology are, of course, not all that the human mind
processes intuitively. Some of the basic domains of intuition are enumerated
in the list that follows (see Barrett 2000, 2008; Boyer 1994b; Keil 1979, 1996;
Sommers 1959). We intuitively understand that things exist in space and time;
some things also have physicality and solidity, and some of them belong to
the folk-biological domain of living kinds (plants and animals). Animacy is
a mediating category between folk biology and mentality, in the sense that it is a
subcategory within folk biology and may be the basis on which mind reading
has developed: the idea of beliefs and desires in a way completes the understand-
ing that something is goal-directed. Social position, for its part, is the domain
of such social relationships as “being someone’s father ” or “being the king,”
and so forth. Such phenomena as cooperation and reciprocal altruism require
one to have intuitions about the kind of social relationship they entail. Finally,
we must add the category of time—though temporality is a neglected theme

38. religion: her ideas of his ideas of their ideas . . . 23
in cognitive neuroscience—because humans experience everything as existing
in time and have intuitions about temporal relationships such as “before” and
“after ” (Lloyd 2002, 2004, 260 – 73, 310 –25).
basic domains of intuition
Spatiality: Location in space
Physicality: E.g., liquids
Solidity: Natural objects, artifacts
Living kinds: Plants and animals
Animacy: Liveliness, goal-directedness
Mentality: Beliefs and desires
Social position: Social relationships
Events: Events and acts
Temporality: Everything exists in time
The systems of HADD, HUI, and HTR, though triggered by domain-
specific sensory or conceptual inputs, probably must also consult such other
systems as the physicality module, for example (Carruthers 2006, 6). The mind
can be viewed as a collection of modular reasoning systems, but the modules
are only functionally specified systems, not encapsulated entities; they can even
share parts with each other (see Barrett and Kurzban 2006; Carruthers 2006;
the appendix). Although the problem of modularity is far from being solved,
I will use “methodological modularity” as a reasoning strategy ( leaving the
question of evolution somewhat open; see the appendix and Panksepp 1998,
2007). For example, it is obvious that a face-like arrangement of two dots and
a straight line has the power to trigger the idea of a face; one simply fills in the
gaps in perception, drawing from a tacit and domain-specific database. As face-
like arrangements thus reliably trigger the idea of a face and not something
else, some kind of functional modularity is the most economical explanation.
But as modules are not encapsulated, some face-like arrangements might fail
to trigger the idea of a face in some persons under some circumstances.
Interestingly, we can also manipulate the concepts and images triggered in
the mind, creating new concepts that do not respect the boundaries of an “intu-
itive ontology” (Boyer 1994a,b, 2007). In this way, we create what Boyer calls
“counterintuitive” representations. In my previous writings, I have thus sug-
gested that the notion of “supernatural agents” be replaced by the concept of
“counterintuitive agents.”15 I am now forced partly to change my mind, because
spontaneous attribution of agency to physically unidentified sources does not
seem to be counterintuitive (see Atran 2002a, 65; Barrett 2004b, 77–88; Bloom
2005, 2007; Lindeman and Aarnio 2007). The HADD, HUI, and HTR model
actually imply that people perceive disembodied agency as a natural category
(see Barrett 1998, 617). Yet it is also possible to form counterintuitive repre-
sentations of agency by adding some extra features or by denying something

39. 24 human agency
that is natural (as shown by the plus and minus symbols in the following list).
Moreover, it is not always clear in practice whether a representation is or is not
counterintuitive. The cross-domain formation of counterintuitive representa-
tions is presented in “Counterintuitive Combinations” below.
Counterintuitive representations are formed by tacitly assigning the repre-
sented entity to an intuitive ontological category, while recognizing that it contains
elements that contradict intuitive expectations concerning that category. Typical
intuitive ontological categories are persons, animals, plants, artifacts, and natural
objects (Boyer 1994b, 101, 1996c, 84, 1998, 878, 2000b, 198, 2000c, 280).
Boyer claims that a catalogue that uses these basic intuitive ontological cat-
egories and includes only breaches of physical and cognitive properties associ-
ated with those categories virtually exhausts cultural variation in this domain
(Boyer 2000a, 103; cf. Franks 2003).
combinations of counterintuitive representations
category possible violations (–, +)
Spatiality – Location of a nonexisting thing
+ Spatiality and mentality
Physicality – A liquid that is nowhere
+ A liquid that understands
Solid objects – Solid object that is nowhere
+ Artifact with mentality
Living kinds – Plant or animal without physicality
+ Plant or animal with high-level mentality
Animacy –Animacy without spatiality
+ Animacy with mentality (without spatiality)
Mentality – Mentality without spatiality or biology
(+ Omniscience)
Social position – Denial of intuitive relationships
+ Adding a relationship (“son of God”)
Events – Intuitively expected event does not take place
+ Intuitively not to be expected happens
Temporality – Immortality (no death)
+ Immortality ( boundless afterlife)
This is not an exhaustive list of intuitive domains or of all the possible violations
of intuitive expectations; it is meant merely to convey the general idea. I will
try to provide some flesh for the bare bones here. For spatiality, it is difficult to
find an example of a representation of an existing thing that yet is nowhere;
it is difficult or impossible to separate mere “existence” from “existing some-
where.” A combination of spatiality and mere animacy might be represented
by the idea of the will-o’-the-wisp, for example ( Thompson 1955–58, 3:131).

40. religion: her ideas of his ideas of their ideas . . . 25
When we combine spatiality, animacy, and mentality ( but without biology),
we get the typical idea of a ghost as a mist-like, animated entity (see Simon-
suuri 1999; 2:419–81; 3:289). The angel of the Lord who appears to Moses
“in flames of fire from within a bush” (Exod. 3:2) also counts as an example
(cf. Wyatt 2005, 13–17).
A solid object that is nowhere comes close to being a contradiction in terms.
Perhaps a layperson who asks where the space-time of scientific cosmology
exists is trying to represent something like a solid object that exists but is
nowhere ( because there is nothing outside of the space-time). In folk tradi-
tions, such an idea is unknown. On the other hand, natural objects and arti-
facts that have mentality form a recurrent theme in folk traditions. Examples
range from magic wands and amulets to things such as the image of the Black
Madonna of Czestochowa, Poland, which is believed to work miracles (Cruz
1994). Some related examples might be cases of physicality and animacy
(without biology). A case in point might be the Arthurian sword, embedded in
an anvil that was in turn embedded in a great stone, with the words “Whoso
pulleth out this sword of this stone and anvil, is rightwise king born of all
England” (Malory 1998, 8). The idea here seems to be that the sword was
somehow alive yet only something physical, without biology.
The category of living kinds includes, for example, invisible plants and ani-
mals or animals capable of miraculous transformations (see Thompson 1955–
58, vol. 3). Another kind of example is animals that talk or think like humans.
Animacy without spatiality may be difficult to find, but nonspatial ani-
macy combined with mentality might be represented in certain very abstract
conceptions of God: God exists but is nowhere.
Such representations also exemplify mentality without spatiality (but men-
tality without animacy is difficult to conceive). Another example is “pure
consciousness” understood as a kind of “knowledge by identity,” or know-
ing something by virtue of being that something (Forman 1990, 1993). In
Buddhism, we find the idea that the supposed objective support of cognition
(vijñānālambana, roughly: material things) is “nothing but idea” (vijñaptimātratā)
or “mind only” (cittamātratam) (Samdhinirmocanasutra 8.7–9). The idea of the
Buddha before he was born (see Pyysiäinen 1993, 143) exemplifies mentality
without biology, as does Paul’s idea of a resurrection body: “Flesh and blood
cannot inherit the kingdom of God . . . the dead will be raised imperishable,
and we will be changed” (1 Cor. 15:50 –52). An interesting question is whether
it is possible to add something counterintuitively to mentality. Boundless men-
tality, such as omniscience, may or may not count, depending on whether a
category violation really is or is not involved.
Social positions can also be counterintuitively represented, as when people
imagine someone to be the child of elves, of an evil spirit, or of God ( Thomp-
son 1955–58, 5:140 –83). I remain undecided on whether this really is a case of
counterintuitiveness.

41. 26 human agency
Counterintuitive events are typified by miracles such as a man walking
on the water, a dead person rising physically from the dead, and so forth (see
Pyysiäinen 2002a, 2008b).
Finally, there are violations of temporality such as immortality. This can
be conceived of as either a denial of biological death or as adding life to the
condition after death. This is usually regarded as a violation of intuitive biology,
but in addition, the temporal dimension is here represented counterintuitively.
Exceptional longevity and a very fast growing up also belong to this category.
The buddhas, for example, supposedly take on the semblance of being old,
even though in reality they have overcome old age (Mahāvastu 1.133).
Everything that contradicts category-specific, intuitive expectations is coun-
terintuitive, but counterintuitive is not the same as false, funny, or deluded.
Just as people’s everyday intuitions do not necessarily describe reality correctly,
counterintuitive concepts need not necessarily be erroneous. They merely vio-
late one’s intuitive expectations. In modestly counterintuitive representations,
one minor “tweak ” has either added a counterintuitive feature or deleted an
intuitive one.16 The idea of a cat that talks, for example, is counterintuitive in
the sense that we do not normally expect animals (and not just cats) to talk.
Representations are “cognitively optimal” when one counterintuitive feature
makes them attention-grabbing while the retained intuitive features make
them easy to process in mind (Atran 2002a; Atran and Norenzayan 2004; Bar-
rett 2004b; Boyer 2001; Norenzayan and Atran 2004).
A familiar example of a cognitively optimal representation is the idea of an
otherwise ordinary statue that yet hears prayers. Having the agentive capacity
to hear is a violation of expectations in the artifact category. The idea of a statue
that hears prayers from afar would involve two violations (an artifact hears and
hearing from afar) and thus be massively counterintuitive (Boyer 2001, 86; see
Atran 2002a, 95–107; Atran and Norenzayan 2004, 722).
Boyer and Ramble (2001) as well as Barrett and Nyhof (2001) found in
that when subjects had to recall lists of items and events they had read about in
the context of a short story, their recall was better for modestly counterintuitive
representations than for intuitive ones. Modestly counterintuitive representa-
tions also survived best in artificial transmission to others over three “genera-
tions.” Boyer and Ramble’s experiments carried out in France were replicated
in Nepal with Buddhist monks and in Gabon with laypersons with very much
the same results. In addition, Anders Lisdorf (2001) shows that in Roman
prodigy lists from the first three centuries BC, 99 percent of counterintuitive
representations were cognitively optimal.
Scott Atran and Ara Norenzayan’s (2004, 721–23) experiments seem to
suggest that modestly counterintuitive representations are initially not recalled
any better than intuitive ones but in the long run degrade at a lower rate
(162; Norenzayan et al. 2006). Atran and Norenzayan also argue that when

42. religion: her ideas of his ideas of their ideas . . . 27
counterintuitive representations complement bodies of intuitive information,
they help people recall chunks of intuitive information better (2004, 157–59;
Norenzayan et al. 2006). Atran (2002a, 105) observes that in laboratory exper-
iments, the nature of the task may make subjects pay special attention to non-
natural representations and intuitive representations thus lose their privileged
status. When counterintuitive representations are not presented in the con-
text of an exciting, science-fiction-like narrative, intuitive representations are
recalled the best (Norenzayan and Atran 2004, 159).
Studies conducted by Lauren Gonce, Afzal Upal, and colleagues seem to
confirm that when subjects are presented mere lists of concepts, without a
narrative context, they indeed recall intuitive representations better than mod-
estly counterintuitive ones. Modestly counterintuitive representations are better
recalled when the narrative context creates an expectation for counterintuitive
concepts that persons thus may experience as intuitive. As different types of
discourse activate different kinds of background knowledge, persons can, for
example, expect the attack of aliens in a science fiction movie but not in such
real life situations as, for example, listening to the radio news (Gonce et al.
2006; Tweney et al. 2006; Upal 2005; Upal et al. 2007).
Gonce et al. (2006) showed that massively counterintuitive concepts have
the poorest recall in both immediate and delayed recall conditions, regardless of
the presence or absence of context. There were no significant differences in the
recall rates of intuitive and modestly counterintuitive concepts except that after a
one-week delay, subjects recalled modestly counterintuitive concepts presented
in a narrative context significantly better than the intuitive ones (also Barrett
and Nyhof 2001). Context also seems to play a crucial role in item recall. In real
life, concepts are usually embedded in a context, and it is in this context that
their counterintuitiveness is evaluated (Gonce et al. 2006; Tweney et al. 2006).
Think of the following embedding of the counterintuitive concept of a “flying
cow ”: “Looking through the kitchen window, I saw the flying cow. The twister
had lifted the animal 50 feet above the ground” ( Tweney et al. 2006, 486).
One who reads only the first sentence is led to interpret the flying cow
as a modestly counterintuitive concept, whereas the second sentence provides
a context in which the flying cow becomes an intuitive concept. Thus, context
can make an apparently counterintuitive concept intuitive. But even when
counterintuitiveness is retained, a relevant context can make it something to be
expected. Actually, Boyer recognized this all along. Persons can simply become
so routinized in using some counterintuitive concepts that their counterintu-
itiveness is no longer consciously recognized (Boyer 1994a, 394, 1994b, 120;
Sperber 1994, 62).
This fact also helps explain why people often regard their own religious
beliefs as perfectly natural while regarding those of other faiths as “supersti-
tious” (see D. Martin 2004; Pyysiäinen 2004d, 90–112). Violations of intuitions

43. 28 human agency
immediately strike one as odd, when one has not had the chance to get routin-
ized in processing them. On the other hand, getting routinized in one type of
counterintuitiveness might also make it easier to accept other counterintuitive
claims as well; there is experimental evidence that persons who score high on
tests of paranormal beliefs are prone to accept confusions of core ontological
assumptions (that is, counterintuitiveness) as quite natural. Confusion between
basic ontological category boundaries distinguishes between superstition and
skepticism better than, for instance, intuitive (versus reflective) thinking or
emotional instability (Lindeman and Aarnio 2007).
1.4.2 Counterfactuals and Counterintuitiveness
Counterintuitive representations presuppose the ability to form counterfactu-
als, in the sense of contrastive representations of states of affairs: “ What if f
instead of c?” or “Maybe f instead of c” (see Garfinkel 1981; Ylikoski 2001). Boyer
has recently suggested that each cognitive module, especially the mind-reading
one, has an inbuilt capacity to produce counterfactuals within its domain.17
Modules would thus not work in a reflex-like manner but instead allow for
reflective reasoning and alternative scenarios.18 Either this capacity is inherent
in the modules, or counterfactual thinking is a domain-general mechanism
that has access to any type of mental contents (see McNamara et al. 2003; Rev-
lin et al. 2003). The fact that we use counterfactuals more readily in mind read-
ing, as compared to intuitive physics and folk biology, could then be explained
by either the nature of the modular computational processes of mind reading
or the fact that domain-general counterfactual reasoning is easier to apply to
data about beliefs and desires than to physical or biological data. (McNamara
et al. 2003 used solely counterfactuals from the domain of folk psychology.)
It seems that imagination and counterfactual thinking cannot be accounted
for by such processes as off-line simulation of experience, or the embodiment
of knowledge gained through practice (see Barsalou et al. 2005, 20 –22; Grush
2002). A better candidate is the received view in cognitive science that con-
ceptual knowledge is somehow transduced from sensory representations of
experience in the brain. In contrast to the various sensory modalities (vision,
audition, etc.), conceptual knowledge is amodal and represented in the form
of abstract symbols that can be manipulated. The problem here is that so far
no account of the possible mechanism of this transduction has been specified.
Yet the capacity to recombine concepts arising from components of experience
and to imagine counterfactual scenarios seems to require some such transduc-
tion (cf. Barsalou et al. 2005, 17).
Counterfactual reasoning is also partly responsible for the production of
counterintuitive ideas, although counterfactuality is by no means the same as
counterintuitiveness—contrary to what the vocabulary of Atran and Norenzayan
(2004) seems to suggest (cf. Barrett 2004a). Counterintuitive representations

44. religion: her ideas of his ideas of their ideas . . . 29
require counterfactuals because such representations involve a what-if con-
trast. Counterintuitiveness violates intuitive expectations and thus creates rep-
resentations of the type “It should be A, but it seems to be B” or “It could be B
instead of being A.”
As the examples quoted show, it is not always easy to categorize counter-
intuitive representations because all necessary information is not always avail-
able. The angel in the burning bush, for example, seems to have animacy and
spatiality, but we do not know for sure whether he also has mentality because
this is not said. Maybe he had physicality, too? And many cases do not quite
fit the Boyerian scheme (see Thompson 1955–58). What about cases such as a
dog vomiting gold, oil bursting from the ground when a saint is made bishop,
and flowers bursting from a saint ’s mouth when he speaks, for example (1:391,
5:456)? It thus is neither possible nor meaningful to try to catalogue all dif-
ferent supernatural agent representations in the world’s religious traditions.
Instead, I try to describe in the following chapters in more detail the case
examples of Finnic and related soul conceptions, the Jewish-Christian God,
and buddhas and bodhisattvas, attempting to show how cognitive structures
shape thinking about supernatural agents.
Apparently, some odd combinations are easier to process in mind than
others and thus are more contagious and easy to spread. Some are modestly
and some massively counterintuitive, but some may be quite intuitive. A com-
bination of spatiality and animacy, as in a cloud that chases you, naturally trig-
gers HADD; yet apparently nothing much follows unless HUI is triggered, too.
When that happens, mentality is attributed to the spatial entity, which results in
a concept of a ghost (Barrett 2008). But some ghosts may also have physicality
in the sense of solid boundaries.
Persons may attribute animacy to physical objects, as in the case of a solid
object that hovers unsupported in midair; but in this case things get far more
interesting when animacy is accompanied by mentality as well (e.g. Strong
2004, 141). This seems to be the case with all kinds of amulets, magic wands,
and so forth. Weeping statues and bleeding icons also belong to this category.
A more intriguing example comes from the narrative motive of a dead evil-
doer’s coffin being so heavy that no one is able to lift it (see Simonsuuri 1999;
Thompson 1955–58, 2:440). The idea here seems to be that it is the sins the
dead person committed that make either the corpse or the coffin heavy. To the
extent that sins are past actions that are heavy, this is a case of transferring
physicality to an abstraction. On the other hand, this is also a case of abstrac-
tions making a physical object heavy.
Mentality without spatiality is the most intriguing category. It seems that
disembodied agency is not counterintuitive; on the other hand, it does not
indicate pure mentality but refers to beliefs about souls leaving and reenter-
ing bodies at will (see Cohen 2008). Agency is independent from the body, in
the sense that mentality directs the body and can be separated from a specific