Animal cognition and animal minds

Abstract

Psychology, according to a standard dictionary definition, is the science of mind and behavior. For a major part of the twentieth century, (nonhuman) animal psychology was on a behavioristic track that explicitly denied the possibility of a science of animal mind. While many comparative psychologists remain wedded to behavioristic methods, they have more recently adopted a cognitive, information-processing approach that does not adhere to the strictures of stimulus-response explanations of animal behavior. Cognitive ethologists are typically willing to go much further than comparative psychologists by adopting folk-psychological terms to explain the behavior of nonhuman animals. This different attitudes of many scientists presupposes a distinction between cognitive and mental state attributions that is not commonly articulated. This paper seeks to understand that distinction.

Full text

Page 1

Allen, page 1
Animal cognition and animal minds
Colin Allen, Texas A&M University
colin.allen@tamu.edu
Psychology, according to a standard dictionary definition, is the science of mind and behavior.
For a major part of the twentieth century, (nonhuman) animal psychology was on a behavioristic
track that explicitly denied the possibility of a science of animal mind. While many comparative
psychologists remain wedded to behavioristic methods, they have more recently adopted a
cognitive, information-processing approach that does not adhere to the strictures of
stimulus-response explanations of animal behavior. Cognitive ethologists are typically willing to
go much further than comparative psychologists by adopting folk-psychological terms to explain
the behavior of nonhuman animals.
The theoretical terms of cognitive and folk-psychological approaches are intentional, in
Brentano’s sense. Contra Brentano, there is considerable optimism among philosophers that it
will be possible to provide a naturalistic theory of intentionality, although there is plenty of
disagreement about the form such a naturalistic theory will take. Given the various deflationary
accounts of intentionality that philosophers are offering these days it is not hard to convince most
philosophers and scientists (staunch behaviorists excepted) that intentional terms are respectable
ingredients in scientific explanations of the behavior of many different kinds of organisms.
Great apes aside, many scientists are skeptical of claims about animal minds even though
they are willing to attribute internal, representational (and hence intentional) states to the same
organisms. To illustrate: Gould (1986) has argued that individual honeybees form internal maps
of their spatial environments that represent features such as the locations of large bodies of water

Page 2

Allen, page 2
relative to the hive. These maps are postulated to explain several phenomena, including that bees
are not recruited by dances that indicate a direction and distance to a food source in the middle of
a lake. (The dances were cleverly produced by exposing foraging bees to a boat with a food
source on board that was gradually moved into the middle of a lake.) Challenges to Gould’s
hypothesis (e.g. Dyer 1991) do not typically start from the premise that bees are incapable of
having cognitive maps; in other words, it is generally considered a reasonable (but perhaps false)
hypothesis that bees utilize internal representations that integrate various spatial features of their
environments. In contrast, many people would reject out of hand the claim that bees have minds,
or that they have beliefs about the locations of bodies of water--thus rejecting mentalistic
explanations of bee behavior. The specific challenge to mentalistic attributions that I have in
mind might, therefore, be raised by someone who rejects strict behaviorism and accepts that
cognitive, intentional explanations are appropriate for explaining some animal behavior. Such a
person would accept that more or less complex interactions between internal states with
intentional content explain some animal behavior, but would deny that the mental states
recognized within folk psychology are appropriately attributed to (most) nonhuman animals.
This attitude presupposes a distinction between cognitive and mental state attributions that
is not commonly articulated. Indeed the terms "mental" and "cognitive" are frequently used
interchangeably by both philosophers and scientists. Among philosophers, when discussion is
focused on intentionality the conflation is unsurprising since both mental states and cognitive
states are widely considered to be intentional. When coupled with Brentano’s thesis that
intentionality is the distinguishing characteristic of the mental the conflation is even easier to
understand. The conflation is also evident in Griffin’s (1976, 1984, 1992) exhortations to
ethologists to become cognitive ethologists by paying more attention to animal minds. And it is

Page 3

Allen, page 3
present when Gould and Gould (1994) use the terms "cognitive map" and "mental map"
interchangeably (searching their index for the former will locate instances of the latter). If we
grant that whatever representation a bee has of its spatial environment may reasonably be called a
"map", the further question of whether it is appropriately called a "mental map" or a "cognitive
map" may seem to be a mere terminological quibble. But I believe that there is a substantive issue
lurking and I hope by the end of this paper it will be clear what that substantive issue is. The fact
that it is not obviously incoherent to accept that bees have cognitive maps of their environments
while simultaneously denying that bees have minds suggest that there is a distinction to be drawn
here.
Just what, however, is being denied when it is denied that bees have minds? There is no
simple answer to this question. We have something like the following state of affairs: many
philosophers are convinced that a naturalistic theory of mind is in the offing because they have
cracked the nut of intentionality. However, in doing so they have given accounts of intentionality
that seem to apply to many organisms and devices that do not, prima facie, possess minds. One
reaction to this state of affairs is to reject the prima facie mindlessness of such entities: thus, for
example, McCarthy (1980) argues that it is appropriate to attribute beliefs to thermostats. A
second reaction is to reject the proffered accounts of intentionality (e.g. Searle 1980, 1992). A
third is to attempt to say what else must be added to an organism before we say it really has a
mind. This latter task has been tackled by several philosophers. (Bennett 1964/1989 specifically
runs the thought experiment on honeybees; see also Fodor 1986; Dretske 1986; Millikan 1993). I
shall take a suggestion of Searle’s on board (even if he would not approve), but what follows is
my attempt to add to the third tradition--which we might call "theoretical comparative

Page 4

Allen, page 4
psychology"--of attempting to say what besides intentionality is relevant to mentalistic
attributions.
Spacetime limitations necessitate passing rather quickly over some controversial
assumptions. First, the question of whether mentalistic (folk-psychological) explanations an
adequate theoretical framework for explaining nonhuman behavior takes on its greatest
significance against a background of the presumed adequacy of using such terms to explain
human behavior (see Lycan, this volume). This is a huge presumption, but nonetheless one I shall
make because I wish to open just one can of worms at a time. By making this assumption, I do
not mean to be claiming that mentalistic terms alone are adequate for explaining human behavior.
My claim is only that their use is not fatally flawed. Given this assumption, the issue at hand is
whether there special reasons for thinking that mentalistic terms are inappropriate for explaining
the behavior of nonhuman animals. Second, I take a realist stance to explanation: an explanation
is genuinely explanatory only if the statements comprising its explanans are true. Thus,
mentalistic explanations are genuinely explanatory only if the mentalistic attributions involved are
literally true of the subjects whose behavior is to be explained, and the question of explanatory
adequacy is linked to the question of whether or not animals possess minds. Third, I hope it
would go without saying that we are seeking a naturalistic and empirically tractable theory of
mental state attributions. In particular, the extent to which questions about animal minds can be
brought into the contemporary scientific domain depends on the extent to which it is possible to
get beyond the reliance on anecdote that is so characteristic of popular discussion of this topic,
and that turned out to be the downfall of early Darwinian comparative psychology (Burghardt
1985; see also Gerald Massey’s commentary following this paper).

Page 5

Allen, page 5
Concern with the folk psychological notions also serves to bring philosophers back in
touch with their scientific colleagues. The reason for this is consciousness: the mental states
recognized by folk psychology are paradigmatically conscious states. Despite the recent
explosion of philosophical and neuropsychological work on consciousness, much remains
puzzling about the notion, including even whether there is a single notion here at all. Suspicion
about consciousness is, therefore, partially responsible for suspicion about attributing believing,
desiring minds to nonhumans. Although Freudian ideas about subconscious beliefs and desires
have infiltrated folk psychology, these do not provide reassurance for attributing mental states to
nonhumans because subconscious beliefs and desires have explanatory value within a Freudian
model of the mind only in the context of competition between the subconscious mind and
conscious ego.
Aside from the deliverances of folk psychology, there are various reasons for thinking that
the topic of consciousness should not be avoided in context of thinking about animal mentality.
Here are two; one is historical, the other conceptual:
(1) Griffin, who coined the term "cognitive ethology", has made consciousness a central
issue in his writings (Griffin 1976, 1984, 1983). Critics of Griffin’s work have been quick to pick
up on this issue and use it to question the scientific status of cognitive ethology (Bekoff and Allen
1996; Allen and Bekoff 1996). For instance, Blumberg and Wasserman (1995, p.133) write: "We
submit that it is this very goal of investigating animal consciousness that, although grand and
romantic, falls far outside the scope of a scientific psychology that has struggled for the better
part of the past century to eschew such tantalizing, but ultimately unsubstantiable, analyses of
subjective mental experience." Consequently, whether proponents of mentalistic explanations of
animal behavior like it or not, to ignore the issue of consciousness with respect to mentalistic

Page 6

Allen, page 6
explanations of animal behavior is not a practical option. Those philosophers (including myself)
who have turned their attention explicitly to cognitive ethology have tended to focus on
intentionality and have had rather little to say about animal consciousness. It is time to restore the
balance.
(2) Deflationary, naturalistic accounts of intentionality do not provide a complete account
of mentality. For example, Millikan’s theory allows that trees are producers and consumers of
intentional icons (Allen and Hauser 1993; Allen 1995). Despite disagreeing with Searle’s
Brentano-like equation of mentality with intentionality, I think there is something worth pursuing
in his suggestion that mentality requires "some awareness of the causal relation between the
symbol and the referent" (Searle, 1980, p.454; my emphasis). Searle’s point cannot quite be
correct; thought can occur even when there is no actual causal connection between a symbol
token and its referent--for example, when the intended referent does not exist. I shall therefore
consider relations between symbol and referent more generally. It is also unfortunate that Searle
says little about what gives rise to awareness of such relations. His gesture towards
neurophysiology is not particularly helpful, and by rejecting behavioral tests (exemplified by his
rejection of the Turing Test for artificial intelligence) it is hard to see his suggestion as
empirically tractable for ethologists and others who study behavior directly. Nonetheless, I think
there is an important idea that is worth pursuing in Searle’s suggestion. Searle is concerned that
purely formal systems have no access to their semantics. So, rather than put the point in terms of
causal relations, we might rather say that mentality involves consciousness or awareness of the
(semantic) contents of internal states. Interestingly, Stich (1978), in arguing explicitly for a
distinction between mental (belief) states and other cognitive (subdoxastic) states comes to a
similar idea. He writes that the abilities of typical adult human believers to report and answer

Page 7

Allen, page 7
questions about their beliefs "are themselves associated with an ability to become aware of or to
be conscious of the contents of one’s beliefs." (Stich 1978, p.504) Stich, rather deliberately, has
little to say about what such consciousness consists in. Although neither Searle nor Stich might
approve of the use to which I intend to put their ideas, I will take up the challenge of trying to
characterize consciousness of content in a way that makes it amenable to behavioral investigation.
This talk about consciousness of content is, admittedly, so far pretty vague, but in what
follows I will try to make it less so. Before going on to give the positive account, I wish state my
neutrality on the question of whether the phrase "consciousness of content" picks out qualitative
properties associated one-to-one with specific belief contents. Stich writes somewhat
ambiguously about this issue, viz. (1978, p.504): "in typical cases of belief a subject will have a
certain sort of characteristic conscious experience when his attention is suitably directed to the
content of the belief." In the context of Stich’s discussion, it is unclear whether he thinks this
"characteristic conscious experience" is characteristic of occurrent beliefs generally, or specific to
the belief’s content. In fact, I shall sidestep entirely the issue of "what it’s like" to be a particular
organism (Nagel 1974), or other puzzles about qualia. Like Dennett (1991), I think there’s plenty
left to consciousness even if we end up Quining qualia.
The state of play so far is this: I am assuming that we possess a naturalistic theory of
intentionality (and its corollary, a naturalistic theory of intentional content); I am assuming that
many organisms who possess internal states with intentional content are nonetheless "mindless" in
the sense that these states lack what Rosenthal (1993) calls "state consciousness"; the task
remains to specify what else is needed to have a mind; and I have identified a proposal, albeit
vague, that consciousness of content is a missing ingredient. (Not the only missing ingredient,
mind you.)

Page 8

Allen, page 8
The game plan for the rest of the paper is as follows: I shall describe a framework for
understanding naturalistic approaches to mentality. I shall focus on a particular functional aspect
of a particularly narrow aspect of consciousness, that I have labelled "consciousness of content"
and I shall show how it might be accommodated within this framework. I shall then consider the
consequences of this approach for attributing mental states to nonhuman animals.
The framework for naturalizing the mind that I shall adopt involves the notion of
implementation. The project, according to this conception, is to show how to implement or
instantiate mental properties given nonmental parts. Because of multiple realizability, an account
of implementation or instantiation does not imply reduction of the properties at one level to those
at another, any more than showing how to implement a radio receiver from transistors constitutes
a reduction of the property of being a radio receiver to that of being a particular organized
collection of transistors. Within this framework, it is reasonable to take a hierarchical approach, a
crude version of which is shown in Figure 1.
Place Figure 1 here
The first step within this hierarchical approach (implementation theory 1) is to provide a
theory of content that shows how intentional properties can be instantiated by processes described
in the nonintentional terms of a natural science (for want of a better term, I refer to these as
"mechanical processes"). For the purposes of this paper I am assuming that we possess a
satisfactory, but deflationary, account of intentionality along these lines--Fodor’s (1990) causal
theory, Millikan’s (1984) biofunctional theory, and Dretske’s (1986) "indicator" semantics are all
candidates. Our concern is to specify what else is needed for mentality. Specifically, the
suggestion contained in Figure 1 is that mental processes are implemented by intentional
processes.

Page 9

Allen, page 9
There is much to quibble about the model in Figure 1. One might object to the suggestion
that all mental processes are implemented by intentional processes. The objection may be correct,
but I wish to see how far we can get by ignoring it. Even so, the model is crude in at least two
respects. First, a successful strategy for naturalizing mental phenomena is likely to involve more
than two theories of implementation spanning three layers. Secondly, the model is probably too
linear--that is, it may be necessary to naturalize different features of mentality separately and then
show how mental processes are instantiated by combining several different kinds of process. (A
more sophisticated model would therefore appear as a graph containing at least one
pathway--through various implementation theories--from every higher level node down to nodes
representing mechanical processes.) Nonetheless, the crude model provides a useful starting
point for discussion.
Dretske (1981, 1986, 1988) provides the most explicit example of the two level strategy in
Figure 1. The lower-level implementation theory is developed in Dretske (1981) from the
mathematical theory of information. Turning his attention to belief, Dretske says (p.175): "The
first order of business is to understand how higher-order intentional structures can be
manufactured out of lower-order intentional states." This is at the level of implementation theory
2, and Dretske’s account of mental states is given in terms of their etiology--causation by simpler
indicator states--and their executive role in behavior. Particularly important to his theory at this
level is the role that learning plays in fixing the content of an internal state and in establishing and
maintaining causal connections between internal states and behavior. He also writes (1981, p.
171): "We have provisionally used the concept of belief to distinguish genuine cognitive systems
from mere processors of information." Note the coupling of belief and cognition in this quote,
thus conflating mentality and cognition. In a later work, Dretske (1986) returns to the question of

Page 10

Allen, page 10
distinguishing beliefs from the kind of "primitive representational capacities" (p.32) exemplified
by bacterial magnetosomes. Focusing on the idea that beliefs can be false, Dretske explicitly
attempts to show how the "clear and unambiguous capacity for misrepresentation" (p.32) can be
implemented by simpler intentional processes. According to Dretske, an internal representation R
can be said to misrepresent the occurrence of an external condition F only if, as shown in Figure
2, there is more than one pathway from F to R (through different proximal stimuli s1 and s2 and
internal responses to those stimuli I1 and I2) and (not shown in Figure 2) the system is capable of
associative learning with respect to stimuli that typically (but not necessarily) occur in the
presence of F. Given such a set up, it is possible for a particular stimulus to trigger R even when
F is not present, thus (omitting some details) R can misrepresent the presence of F.
Place Figure 2 here
Let us grant that this provides the robust notion of misrepresentation that Dretske desires,
and that the capacity for misrepresentation is a necessary condition for an organism to have
beliefs. Nonetheless (and I expect that Dretske would agree) it is not yet sufficient to lay to rest
worries about belief attributions to nonhuman organisms. Associative learning is a very
widespread phenomenon; its neural basis has, for example, been studied in the marine snail
Aplysia californica. If we could say nothing more about beliefs than that they involved the
capacity for misrepresentation, then we might be faced with having to bite the bullet and attribute
beliefs to Aplysia. However, the number of scientists or philosophers willing to do this is
probably vanishingly small, and although this is not an issue to be settled by taking a poll, I do
think that the intuition that Aplysia do not have beliefs is sound.
The distinction suggested earlier between cognition and mentality suggests a slightly more
complicated model for naturalizing the mind than that shown in Figure 1.

Page 11

Allen, page 11
Place Figure 3 here
In terms of this model, I would argue that Dretske’s account of misrepresentation belongs
to implementation theory 2a for two reasons. First, as I made explicit at the beginning, I am
interested in those uses of mentalistic language that apply only when state consciousness is
involved, but associative learning in the kind of system shown in Figure 2 can plausibly occur
unconsciously. Second, misrepresentation in Dretske’s robust sense appears at the "merely"
cognitive or subdoxastic level. For example, the familiar Müller-Lyer illusion (Figure 4) involves
misrepresentation of the relative lengths of two lines. Susceptibility to this illusion seems to
depend on experience with rectangular three-dimensional objects. Visual patterns of the form
<--> are associated with protruding, hence closer and smaller edges between two surfaces, and
patterns of the form >--< are associated with receding, hence more distant, larger edges.
Place Figure 4 here
Illusions such as this are extremely interesting for understanding a difference between mere
cognition and full blown belief. Our visual systems are such that, given typical exposure to
buildings and other rectangular objects, we cannot help but see one line as longer than the other
when presented with the Müller-Lyer figures. There is no choice in this matter. But at the level of
conscious understanding we are capable of representing that the lines are really the same length
even though they do not appear that way. Put more generally, we are capable of recognizing that
our visual systems have misrepresented the world.
I want to use this idea of a capacity for detecting one’s own misrepresentations to flesh out
the notion of consciousness of content. There are two separate projects here: one is to argue that
the functional capacity to detect misrepresentation is necessary (although perhaps still not
sufficient) for consciousness of content; the other is to provide an account of how such a capacity

Page 12

Allen, page 12
might be implemented. I shall concentrate on the first in the hope that a reasonable functional
characterization is an important first step towards an implementational theory.
Organisms act in ways that are appropriate to the contents of their representations. Attributions
of misrepresentation are explanatorily useful, because they can help to explain why an organism
does something that is not appropriate given the actual state of affairs. Consider a worker ant
confronted with a dead nestmate already beginning to decompose. Responding to oleic acid, a
chemical byproduct of decomposition, she goes into an internal state R that causes the invocation
of a remove-from-nest routine. Experiments show that ants will engage in the remove-from-nest
routine even when the oleic acid is emanating from something other than a dead ant, such as a
piece of paper, or even a moving live ant. I do not wish to argue against the view that, as
described so far, R is a meaningless intervening variable because this behavior is amenable to a
strict stimulus-response explanation according to which the stimulus of oleic acid invariably
produces the remove-from-nest response. But suppose that there are organisms similar to the ants
for which there are other pathways to R that do not involve the same proximal stimulus and that
these organisms are capable of learning to associate other stimuli with dead nestmates. That is,
these hypothetical organisms conform to the structure laid out by Dretske (Figure 2) and their
tokens of R are capable of misrepresenting the external condition of the presence of a dead
nestmate (F). For such an organism, it would be appropriate to explain her removal of a live
nestmate or a piece of paper on the grounds that she had misrepresented these objects as dead
nestmates.
If one identified the capacity for belief with the capacity for misrepresentation, one might
loosely say that our imagined organism removes the corpse, paper, and live nestmate because she

Page 13

Allen, page 13
believes them all to be dead nestmates. Alternatively one might say that she removes these
objects because they appear to her to be dead. Someone might respond to this loose talk by
asking "Which is it--does she believe those things to be dead, or do they just appear dead to her?"
This question presupposes a distinction between appearance and belief that makes sense only
when applied to an organism that is capable of making a distinction between the way things
appear and the way they are. As described so far, there is nothing to suggest that ants or our
imagined ant-like organisms are capable of making such a distinction for they are entirely locked
into responding as if R has true content. In contrast, consider our own responses to the
Müller-Lyer illusion. We are not locked into responding to the deliverances of our perceptual
systems as if they had true content, but we are capable of comparing the contents of those
perceptual states to other representations of the same states of affairs. Thus we can believe that
the lines are the same length even though they appear to have different lengths.
There are at least two reasons for thinking that such an ability to detect representational
error is an ingredient of consciousness of content. First, misrepresentation occurs when the
normal relationship between symbol and referent is absent. Thus, the ability to detect
misrepresentation constitutes an ability to detect the failure of this normal relationship. This ties
in rather closely with Searle’s idea of awareness of the relation between symbol and referent.
Second, if we take seriously the claim that this capacity involves drawing distinctions between
appearance and reality, then we are required to take seriously that there is a way that things
appear to the organism. (Note that this does not entail that the organism is capable of making the
appearance reality distinction. For that, it is necessary to have the concepts of appearance and
reality. The capacity under discussion here requires only the ability to detect specific cases of
misrepresentation, not the ability to think about misrepresentation generally.) Although we might

Page 14

Allen, page 14
be able to speak of the way things appear to an organism that cannot detect its own
misrepresentations, the theoretical motivations for doing so are much weaker than when the
organism itself makes the distinction. Appearance is a key component of consciousness, thus the
present characterization helps to explicate a role for consciousness in mental life.
As I mentioned earlier, I am not going to delve into implementational details, which are
likely to be complicated. However, it is possible to identify some necessary elements (in no way
do I maintain that these elements are sufficient). In addition to the Dretskean capacity for
misrepresentation, for an organism to be conscious of a specific content that F has occurred, it
must be have alternative ways of representing the occurrence of F and the non-occurrence of F.
Consider the Müller-Lyer illusion once more. Detecting the illusion depends on at least having
the visual representation that one line is longer than the other while simultaneously being capable
of representing the relative length of the lines nonvisually (in humans, this is plausibly
accomplished linguistically). If one was capable only of the visual representation, one would be
bound to react as if the lines were of different length. One can detect the visual misrepresentation
only because one is capable of representing simultaneously and nonvisually that the lines are of
the same length.
The capacity for representing instances of misrepresentation is minimally a second-order
capacity because a state that represents such a misrepresentation involves intentional notions in its
content. It is important, however, to realize that this capacity does not entail either self-awareness
or a conscious theory of mind. For just as an organism may have states with first-order
intentional contents yet fail to be conscious of those contents, so too may an organism have states
with second-order contents and fail to be conscious of those contents. Having a belief in virtue of

Page 15

Allen, page 15
being conscious of the content of some intentional state is not tantamount to having a belief about
a belief.
We are now at the last stage of my plan, which is to draw out some of the implications of
these ideas for the attribution of mental states to animals. Of prime importance, I believe, is that
the idea of sensitivity to representational error is an empirically tractable notion that can be
assessed in many ways. Reactions such as surprise, embarrassment, and rapid learning (often
involving just one or a few experiences with the conditions that caused the error) are all
characteristics that might be shown by organisms who have epistemic access to their own errors.
It is an empirical question to what extent various organisms can detect and respond to their own
errors, and there is plenty of room for matters of degree. There are various promising areas to
look. One such area is social play, which involves responding to behaviors that would elicit
different responses in nonplay contexts (there is also an interesting developmental connection in
human children between the concepts of pretense and appearance; see Flavell et al. 1987).
Another area is social communication, particularly where signals are used deceptively or withheld
in conditions where they would be appropriate (see Allen and Hauser 1993 for extensive
discussion of an interesting case of signal-withholding by rhesus macaques).
The present approach helps to account for why some features that are often suggested as
evidence of nonhuman mentality (or the lack thereof) are, indeed, relevant. Three such features
are behavioral flexibility, the integration of information from multisensory sources, and language
abilities. Behavioral flexibility is relevant to mentalistic attributions because it is connected to an
organism’s monitoring of its own performance. An organism that cannot detect when its states
misrepresent its environment will be limited to adjusting its behavior only when the proximal
causes of those states are removed. Merely cognitive systems are typically inflexible in this way;