Published 8 Jan, 2008
The philosophical issues that relate to research on animal cognition can be categorized into three groups: questions about the assumptions on which the research is based; questions that arise about the methods used in the research programs; and questions that arise from the specific research programs themselves.
While the study of animal cognition is largely an empirical endeavor, the practice of science in this area relies on theoretical arguments and assumptions regarding the nature of mind and rationality. If nonhuman animals don't have beliefs, and if all cognitive systems have beliefs, then animals wouldn't be the proper subject of cognitive studies. If animals aren't agents because their behavior isn't caused by propositional attitudes, and if all cognitive systems are agents, we get the same conclusion. While there are arguments against animal minds, the cognitive scientists studying animals largely accept that animals are minded, cognitive systems. Animal consciousness, however, is a different matter; like the topic of human consciousness, it is an area that some scientists are less willing to engage with.
Many of the research programs investigating particular cognitive capacities in different species raise philosophical questions and have implications for philosophical theories, insofar as they impose additional empirical constraints for naturalistically minded philosophers. Traditional research paradigms in animal cognition are similar to those in human cognition, and include an examination of perception, learning, categorization, memory, spatial cognition, numerocity, communication, language, social cognition, theory of mind, causal reasoning, and metacognition.
These different research programs are not always investigated using the same methods. For example, social cognition could be studied by a developmentalist who documents mutual gaze in mother-infant dyads across primate species (e.g. Matsuzawa 2006b), an ethologist interested in free-ranging canid social play behavior (e.g. Bekoff 2001), an experimental psychologist testing theory of mind in an adult symbol-trained chimpanzee (e.g. Premack & Woodruff 1978), an anthropologist observing social games in capuchin monkey communities (e.g. Perry et al. 2003), or a cognitive neuroscientist investigating neural basis of gaze-following in primates (e.g. Emery 2000). Different methods for investigating a phenomenon are based in part on whether the study is in captivity or in the field, and whether developmental stage, social relations, ecology, and past history are taken into account.
1. What is Animal Cognition?
Animal cognition is the study of the processes used to generate adaptive or flexible behavior in different animal species. As a part of cognitive science, research in animal cognition aims to uncover the different cognitive mechanisms at play across species in order to understand the varieties of cognition, the similarities between humans and other species, and the evolution of cognitive processes.
As with human cognition, there are competing theories about the structure of animal cognition at play. After the cognitive revolution many animal cognition researchers believed that cognition is a general purpose computational system. This was especially true of those who advocated the Social (or Machiavellian) Intelligence Hypothesis (Humphrey 1978; Jolly 1966; Byrne & Whiten 1988; Dunbar 1998). According to the hypothesis, the relatively sophisticated general problem-solving capacities of social animals are due to challenges arising from social living. This view has been challenged by those who claim that the research on other species suggests that cognition is more modular than the traditional view allows. Some argue that individuals cannot use one mechanism to solve all cognitive tasks, since there are different rates of learning and different computational processes implicated in different domains. Because of these differences, cognition must be thought of as a number of special-purpose computational modules rather than one general purpose processor (Hauser & Carey 1998; Shettleworth 1998; Cosmides & Tooby 1994). Other researchers examine animal cognition from within the framework of embodied cognition (Barrett & Henzi 2005), or dynamical systems theory (King 2004; Shanker & King 2002).
Animal cognition research has historically accepted a close relationship between affect and cognition. Early experimental psychologists manipulated subjects' motivation (e.g. by withholding food) and the pain associated with shocks were assumed to cause an unpleasant affective response. Today there is growing interest in the emotional responses of animals, as will be discussed in Section 4.3. In this sense, animal cognition research has anticipated the recent interest in emotion in cognitive science.
2. Theoretical Issues
Philosophical discussion about animal cognition has traditionally focused on the metaphysics and epistemology of mind in creatures who do not have language. While the traditional debate about the existence of animal minds is problematic given the lack of clarity about the nature of mind, recent discussions of animal beliefs and rationality help to make the discussion less muddy.
The early history of western philosophy reflects a tendency to see animals as lacking rationality. Aristotle defined "human" as "the rational animal", thus rejecting the possibility that any other species is rational (Aristotle Metaphysics). Aquinas believed that animals are irrational because they are not free (Aquinas Summa Theologica). Centuries later Descartes defended a distinction between humans and animals based on the belief that language is a necessary condition for mind; on his view animals are soulless machines (Descartes Discourse on the Method). Locke agreed that animals cannot think, because words are necessary for comprehending universals (Locke Essay Concerning Human Understanding). Following in this tradition, Kant concludes that since they cannot think about themselves, animals are not rational agents and hence have only instrumental value (Kant Lectures on Ethics).
However, there were dissenters. Voltaire criticized Descartes' view that humans but not animals have souls and hence minds, by suggesting there is no evidence for the claim (Voltaire Philosophical Dictionary). Hume was downright dismissive of the animal mind skeptics when he wrote "Next to the ridicule of denying an evident truth, is that of taking much pains to defend it; and no truth appears to me more evident than that beasts are endowed with thought and reason as well as man. The arguments are in this case so obvious, that they never escape the most stupid and ignorant" (Hume Treatise of Human Nature, 176).
2.1 Animal Minds
Despite Hume's judgment about their worth, much ink has been spilled developing arguments for the existence of animal minds. The two standard arguments are extensions of the two arguments for other minds: the argument from analogy and the inference to the best explanation argument. The argument from analogy for animal minds can be formulated as:
All animals I already know to have a mind (i.e. humans) have property x.
Individuals of species y have property x.
Therefore, individual of species y probably have a mind.
Versions of the argument differ as to what they choose as the reference property x, and how they defend the choice of reference property. The reference property might refer to some general capacity (e.g. problem solving), some specific ability (e.g. language, theory of mind, tool-use), or it might consist of some set of properties.
The argument from analogy for animal minds is in one sense stronger than the argument for other minds insofar as the reference class is larger; rather than starting with the introspection of one's own mind and then generalizing to all other humans, the argument for animal minds takes as given that all humans have minds and generalizes from the human species to other species. In another sense, the analogical argument for animal minds is weaker, since the strength of the argument is a function of the degree of similarity between the reference class and the target class. Humans might be more similar to one another than they are to members of another species.
The second standard argument for animal minds rests on the claim that the existence of animal minds is a better explanation of animal behavior and physiology than those offered by other hypotheses. This argument can be formulated as:
Individuals of species x engage in behaviors y.
The best scientific explanation for an individual engaging in behaviors y is that it has a mind.
Therefore, it is likely that individuals of species x have minds.
The inference to the best explanation argument justifies the attribution of mental states to animals based on the robust predictive and explanatory power that is gained from such attributions. As the argument goes, without such attributions we would be unable to make sense of animal behavior. This argument makes use of standard methods of scientific reasoning; of two hypotheses, the one that better accounts for the phenomenon is the one to be preferred. Those who offer this sort of argument for animal minds are claiming that behaviorist or other mechanistic explanations for animal behavior fail to account for the diversity and flexibility of behavior in at least some species of animal. Critics of this argument offer alternative explanations for the relevant behaviors.
2.2 Rationality, Beliefs, and Concepts
Though philosophers and psychologists seem to generally accept that other species have minds, there is widespread disagreement about what exactly that means. For some, it merely means that some species can feel pain (and hence there are codes of conduct for working with animals). For others, it means that some species are rational agents who have reasons for action. The theoretical investigation into animal cognition examines whether any other species are rational, have beliefs, or have concepts. There are corresponding epistemological questions about how we might ever know the content of an animal's belief or reason for action, given the difficulty of attributing content to a creature who doesn't use language. Whether animals are rational is related to philosophical questions having to do with the moral status of animals. For example, arguments in favor of personhood for great apes are made on the basis of the rational capacities of these species (Cavalieri & Singer 1994), as are arguments for duties toward animals (Skidmore 2001), and arguments given in favor of moral agency (or proto-moral agency) (de Waal 2006; Hauser 2006).
Discussions of animal rationality are confounded by the lack of consensus on what is required for rationality. Because there are many different kinds of rationality (e.g. practical vs. theoretical, process vs. product), disagreements about what sorts of cognitive mechanisms are implicated in rationality (e.g. linguistic processing, logical reasoning, causal reasoning, simulation, biases and heuristics), and disagreements about the extent to which different kinds of normativity are implicated in rationality (e.g. biological fitness or reason-respecting propositional attitudes), there is no straightforward way to answer the question about whether other species are rational agents. Some philosophical theories of rationality are based on an initial acceptance of rationality across species, given evolutionary considerations. For example, on Fred Dretske's view, even some simple learned behaviors, such as a bird's avoiding eating a monarch butterfly, can be construed as minimally rational. Because monarchs who eat toxic milkweed become toxic to birds and other predators, when a bird learns not to eat monarch butterflies after having formed an association between eating monarchs and vomiting, it has a reason for its avoidance behavior. The birds also have a reason to avoid eating a viceroy, given that it is visually almost indistinguishable from a monarch, though not poisonous. The behavior in both cases is explained by the content of the bird's thought (or "thought"), and for Dretske this is sufficient for the bird to count as a minimally rational agent (1988, 2006). Other theories of rationality that take evolutionary considerations into account include those of Daniel Dennett (1995, 1987), Ruth Millikan (2004, 2006), and Joelle Proust (1999, 2006). Another method used to develop theories of rationality is to base it on the human model, and then attempt to extend it to other species. This approach is exemplified by José Bermúdez (2003, 2006). A third class of theories try to stake a middle ground between these two strategies (e.g. Hurley 2003a, 2003b, 2006). For a collection of essays on rationality across species see Hurley & Nudds (2006).
Given the lack of theoretical consensus on the nature of rationality, empirical research projects are not designed to examine rationality directly. Instead, researchers investigate various capabilities that may be associated with rationality. For example, tool-use has long been considered to be an indicator of rational thought. Because tool use involves finding or constructing an object that is utilized as an extension of the body to achieve a goal, it is thought that tool use requires identifying a problem, considering ways of solving the problem, and realizing that other objects can be used in the manipulation of the situation. Early experimental research on chimpanzee problem solving by the German psychologist Wolfgang Köhler had chimps constructing tools to acquire out of reach objects; it was reported that chimpanzees would stack boxes or put together tubes to form a long rod in order to reach bananas hung overhead (Köhler 1925). Given this behavior, Köhler suggested that chimpanzees solve some problems not by trial-and-error or stimulus response association, but through a flash of insight. (But see Povinelli (2000) for a critique of the contemporary interpretation of Köhler's research). Since the days of Köhler, tool use in the wild has been discovered in a number of different taxa, including all great apes, some monkeys, some birds, sea otters, and cetaceans.
Some theoretical arguments about animal rationality identify rationality with other properties, such as having beliefs or concepts. Donald Davidson has offered an argument against animal rationality based on an association between concepts, beliefs, and language. On Davidson's view, believers must have the concept of belief, because to have a belief they must recognize that beliefs can be true or false, and one cannot understand objective truth without understanding the nature of beliefs. In order to develop an understanding of objective truth, one must be able to triangulate with others, to talk to others about the world, and hence all believers must be language users. Since other species lack language, they do not have beliefs (Davidson 1982). Davidson also argues against animal beliefs based on the claim that having a notion of error is necessary for being a believer (Davidson 1975).
A different argument against animal belief has been presented by Stephen Stich, who argues that we cannot attribute propositional attitudes to animals in any metaphysically robust sense, given our inability to attribute content to an animal's purported belief (Stich 1978). On Stich's view, if attribution of belief to animals is understood purely instrumentally, then animals have beliefs. However, if attribution of beliefs to animals requires that we can accurately describe the content of those beliefs, then animals don't have beliefs. Given the second sense of having belief, Stich argues that because "nothing we could discover would enable us to attribute content to an animal's belief" (Stich 1978, 23), we are unable to make de dicto attributions to other species, and we cannot make de re attributions because this would violate the truth-preserving role of attribution. Hence we can make no attribution, and if we can't say what an animal's belief is about, it makes no sense to say that an animal has a belief. The worry here is similar to the worry about anthropomorphism; when we use our language to ascribe content to other species, we may be attributing to them more than is appropriate. Stich is concerned that when we say "Fido believes there is a meaty bone buried in the backyard" we are attributing to Fido concepts he cannot possibility have, concepts like "backyard" which are only comprehensible if one has corresponding concepts such as "property line", "house", "fence", and so on. Stich's argument can be formulated as:
In order for something to have a belief, it must have a concept.
In order to have a concept, one must have particular kinds of knowledge, including knowledge of how the concept relates to other concepts.
Non-human animals don't have this knowledge.
Therefore, non-human animals don't have beliefs.
While animal cognition researchers agree that we ought to be careful about the concepts that we attribute to other species, many deny Stich's claim that empirical research cannot help us learn anything about the conceptual organization of other species. One of the earliest attempts to examine animal concepts came out of a series of experiments with pigeons. The subjects were shown photographs, and were trained to peck at the pictures that contained a target object, such as a tree, and not respond to the pictures that didn't contain the target object. After the training period, the pigeons were able to generalize to new photographs, pecking only at those that contained trees just as in the training set. It was suggested that this sorting ability demonstrates that the pigeon has a concept of the target object (Herrnstein 1979; Herrnstein et al. 1976).
Many reject the idea that being able to sort objects is sufficient for having a concept corresponding to the sortals. For one, some think language is necessary for concept acquisition (Chater & Heyes 1994). Others think that while concept acquisition is independent of language, sorting behavior alone doesn't demonstrate having a concept, because humans can be trained to sort objects while lacking the corresponding concept. As Colin Allen and Marc Hauser put it, "It is possible to teach a human being to sort distributors from other parts of car engines based on a family resemblance between shapes of distributors. But this ability would not be enough for us to want to say that the person has the concept of a distributor" (Allen & Hauser 1996, 51).
Rather than identifying concept acquisition with sorting behavior, Allen and Hauser suggest alternative methodologies for identifying concepts in other species. For example, they offer a possible (though, they admit, ethically untenable) test for a death concept among vervet monkeys (Allen & Hauser 1996). Vervet mothers are capable of recognizing the alarm cries of their infants, and when they hear such a cry the mother will look towards her infant, and the other females will look towards the mother. Allen and Hauser suggest that playing a recording of a recently deceased infant's alarm cry would help to determine whether vervets have a concept of death. If the mother responds to these recordings in an atypical fashion, unlike the usual response made to a living infant, that response provides evidence of a death concept. According to Allen and Hauser, having a concept permits different responses to identical stimuli. The actual sound of the infant's alarm cry during life is identical to the sound played back after death. If the response to this stimulus is different, this is evidence that there has been a conceptual change associated with the stimulus. Allen presents the general strategy for attributing concepts to animals as follows: "An organism O may reasonably be attributed a concept of X (e.g. TREE) whenever:
O systematically discriminates some Xs from some non-Xs; and
O is capable of detecting some of its own discrimination errors between Xs and non-Xs; and
O is capable of learning to better discriminate Xs from non-Xs as a consequence of its capacity" (Allen 1999, 37).
One way to study the conceptual structure of other species is to use the same methods that are used to study concepts in another group that lacks language, namely human infants (Hauser et al. 1996; Hauser & Carey 1998; Bermúdez 2003; Gómez 2005). The preferential looking time paradigm, also known as the dishabituation paradigm, is used to study human infants' understanding of the physical and social world (Baillargeon & DeVos 1991; Spelke 1991). Dishabituation experiments are thought to help us understand what kinds of predictions infants make about their word, and this information can help us determine how they see the world. The methodology is simple; an infant is repeatedly shown a stimulus, and after becoming habituated to the stimulus the infant becomes disinterested. At this point, a new stimulus is shown. If the infant sees the new stimulus as different from the target stimulus, or impossible given the target stimulus, the infant will look longer at the new stimulus. If the infant takes the new stimulus to be similar to the target stimulus, then she will not show any additional interest. The idea is that by comparing responses among groups of individuals, a researcher can learn something about how that group conceptualizes the world.
In one study using this method, Marc Hauser and colleagues investigated numerical concepts in different primate species, including rhesus monkeys (Hauser et al. 1996) and cotton-top tamarins (Uller 1997). The researchers tested the monkeys' ability to keep track of individual objects placed behind a barrier. They found that like human infants, the monkeys look longer at impossible outcomes. For example, in one test condition the rhesus monkeys were shown two eggplants serially placed behind a screen, and then the screen was lifted showing only one eggplant. The monkeys looked longer at the one eggplant than they did when shown the expected two eggplants, suggesting that they represent the eggplants as distinct sortals.
Another way we might learn how different species organize the world is to teach individuals a symbolic communication system. For example, the biologist Irene Pepperberg trained an African Grey parrot named Alex to sort objects using meta-level concepts that categorize other concepts. Alex was able to sort objects according to color, shape, and matter, and he was able to sort sets of objects according to number. In addition to sorting, Alex could report which feature makes two objects similar or different. For example, when presented with a red block and a red key, Alex responded to the question "What's same?" by uttering "color." He could also report similarities and differences in shape and matter. Pepperberg claims that her studies demonstrate Alex's understanding of categorical concepts, and reveal the classifications that Alex devised (Pepperberg 1999). However, one might be worried that the concepts exhibited by symbol-trained animals are an artifact of the communication system, and not typical of the species.
When researchers attribute mental states to other species, they open themselves to the charge of anthropomorphism. The term "anthropomorphism" refers to the act of attributing uniquely human traits to other animals; the traits in question are usually psychological states. In recent years, there have been a number of theoretical discussions about the charge of anthropomorphism itself (including the essays in Mitchell et al. 1997; Daston & Mitman 2005; and work by Fisher 1990, 1991; Kennedy 1992; Crist 1999; Rivas & Burghardt 2002; Keeley 2004; Andrews forthcoming).
In response to charges that psychological and agential attributions are examples of anthropomorphic attribution, some have argued that the charge of anthropomorphism is a charge that the attributor is making a category mistake, rather than merely a false attribution. It is a claim that the attribution must be logically false, because members of the target species are not the sorts of things to which the term can apply (Fisher 1991; Keeley 2004). However, if the charge of anthropomorphism is the charge that the attributer is making a category mistake, then the charge is being made on conceptual, rather than empirical grounds; hence the worry that refusing to attribute so-called anthropomorphic properties without first examining whether they might be held by members of any other species is unscientific (Fisher 1990; Asquith 1997; Keeley 2004). Thus, one response to the charge of anthropomorphism is continued research, for one wouldn't know whether a property is anthropomorphic until after the relevant research has been completed. As Sober puts it, "The only prophylactic we need is empiricism" (Sober 2005, 97).
However, Sober also argues that the empirical methodology of psychology places a different burden of proof on animal cognition and human cognition research. This is because the null hypothesis in the animal cognition research is that there are different cognitive mechanisms at work in humans and animals. Given that type 1 errors (reporting a false positive and rejecting a (possibly true) null hypothesis) are taken to be more serious errors than are type 2 errors (reporting a false negative and not rejecting a null hypothesis when it should be rejected), the practice of science results in a bias against attributing psychological traits to animals (Sober 2005). The debate about how to interpret the results of animal studies as compared to human studies may be seen as a debate about an inconsistent application of Morgan's Canon. Morgan's Canon states: "In no case may we interpret an action as the outcome of the exercise of a higher psychical faculty, if it can be interpreted as the outcome of the exercise of one which stands lower in the psychological scale" (Morgan 1894, 53). Though this is a longstanding rule of thumb in animal cognition research, sometimes referred to as the "principle of conservatism," it is not a principle commonly used in human cognition research. To complicate matters, attempts to determine what exactly Morgan's Canon instructs a researcher to do have raised worries about its meaningfulness (Sober 2005; Allen-Hermanson 2005).
Despite the defenses given for attributing mental states to animals, worries about anthropomorphism remain. Kennedy claims that the arguments for attributing mental properties to animals often rest on a false dichotomy: either animals are stimulus-response machines, or they are agents with beliefs and desires. Since animals are not stimulus-response machines, they must be psychological agents (Kennedy 1992). The problem with this argument is that not all machines implement stimulus-response functions; some machines are complex and indeterministic, and if animals were machines, they would be machines of that sort (Barlow 1990; Kennedy 1992).
Other critics rely on arguments much like those of Stich and Davidson discussed above. If we have no good scientific methods for attributing mental states to creatures without language, then we should not make such attributions. Since we are barred from making the attributions, scientific psychology ought not engage in analyzing animal mentality (e.g. Keeton 1967; Kennedy 1992; Blumberg & Wasserman 1995). Anthropomorphism is seen as a human tendency that must be overcome in order to do good science.
It has been noted that such arguments are about the proper methods of science, the scope of science, and how to interpret data (Keeley 2004; Bekoff & Allen 1997). As such, the argument is not an empirical one, but a theoretical one. This can be seen in the way the debates sometimes result in an impasse. Those opposed to attributing mental properties to animals are accused of begging the question (Griffin 1992), by committing "reverse anthropocentrism" (Sheets-Johnstone 1992) or "anthropodenial" (de Waal 1999). The charge of begging the question goes both ways. Kennedy argues that arguments for animal mentality are grounded in human intuition or introspection and that introspection is itself anthropomorphic and ought not be taken as objective evidence (Kennedy 1992). As both sides accuse the other of begging the question, some conclude that the debate is not fecund, and ought to be replaced with empirical work in comparative biology and psychology (Keeley 2004; Sober 2005; Andrews forthcoming).
The concerns about anthropomorphism appear to be largely limited to western scientists. It has been argued that researchers from countries with a Buddhist rather than Christian orientation are not culturally encouraged to see a categorical distinction between humans and nonhuman animals (Asquith 1991; Sakura 1998; Matsuzawa 2003; de Waal 2003). Unlike Christianity, Buddhist doctrine does not claim that humans but not animals have immortal souls, and it does not allow humans to use animals for their own purposes in the ways Christianity does. The Buddhist tradition sees a connection between humans and other animals, and allows that humans can be reborn as animals. De Waal argues that the difference in cultural attitudes toward animals led to an early rejection of Japanese methods and findings in primatology, and that it is only recently that some of those ideas, such as Kinji Imanishi's claim that primates display cultural differences within species, have made their way into western scientific discourse (de Waal 2003).
3. The Science of Animal Cognition
Scientific interest in animal minds and cognitive capacities grew as a result of Charles Darwin's theory of evolution by natural selection. In The Descent of Man, Darwin introduced many of the issues that motivate the research programs in animal cognition today, including tool use, reasoning, learning, concepts, consciousness, the social sense, and the moral sense. He was also interested in animals' aesthetic judgments and whether they believe in the supernatural, issues that haven't been taken up by contemporary researchers. In addition, Darwin anticipated current interest in implicit reasoning with his comment "The savage would certainly neither know nor care by what law the desired movements were effected; yet his act would be guided by a rude process of reasoning, as surely as would a philosopher in his longest chain of deductions" (Darwin 1974, 75)
Darwin advocated the continuity of the mental across species; just as some morphological characteristics are homologous across species living in similar environments, we should expect psychological and behavioral similarities as well: "the difference in mind between man and the higher animals, great as it is, certainly is one of degree and not of kind" (Darwin 1974, 122). This view was also advocated by Darwin's contemporary, the naturalist George Romanes, who in his book Animal Intelligence writes "there must be a psychological, no less than a physiological, continuity extending throughout the length and breadth of the animal kingdom" (Romanes 1970, 10).
3.1 Anecdotal Method
The method that Darwin, Romanes, and other contemporaries used to investigate these questions could be described as the anecdotal method. Stories about animal behavior were collected from a variety of people, including military officers, amateur naturalists, and layfolk, and were compiled and used as evidence for a particular cognitive capacity in that species.
The anecdotal method as practiced by Darwin and Romanes has been criticized for a number of reasons. The "evidence" gathered was often a story told about an event witnessed by a single person, usually not a trained scientific observer. In addition, these stories were often acquired second- or third- hand, so there were worries that the reports had been embellished or otherwise altered along the way. These problems were recognized early on, and in response Romanes developed three principles for accepting anecdotes in order to avoid some of these problems:
Never accept an incident report as fact without considering the authority or respectability of the observer.
If the observer isn't known, and the incident report is sufficiently important, consider whether the observer may have reason or cause to make an inaccurate report.
Look for corroborations of the observation by examining similar or analogous observations made by other independent observers (Romanes 1970).
The third principle was the one he most relied on, writing "This principle I have found to be a great use in guiding my selection of instances, for where statements of fact which present nothing intrinsically improbable are found to be unconsciously confirmed by different observers, they have as good a right to be deemed trustworthy as statements which stand on the single authority of a known observer, and I have found the former to be at least as abundant as the latter" (Romanes 1970, ix).
Despite Romanes' attempts, the method remained problematic insofar as it didn't provide any statistical information about the frequency of such behaviors; selection bias would lead people to report only the interesting intelligent behaviors and ignore the frequency of behaviors that might serve as counterevidence. Thus, the anecdotal method as practiced by Darwin and Romanes fails many of the virtues associated with good scientific method.
The legacy of Darwin and Romanes for animal cognition can be summed up in the oft-quoted phrase "The plural of 'anecdote' is not 'data'." Though 19th century anecdotalism has been rejected, defenses of the use of observations (often now referred to as "incidents" or "qualitative reports" rather than "anecdotes") come from ethology. Among ethologists, collecting behavioral incidents is part of the standard methodology, and fieldwork by trained scientists has resulted in greater knowledge of animal behavior. For example, when Jane Goodall reported seeing chimpanzee hunting and lethal intergroup aggression, the scientific image of chimpanzees had to be significantly revised (Goodall 1986).
A middle ground in the use of qualitative reports takes the collection of incidents as a research tool. To be useful as a research tool, the concerns about anecdotalism must be addressed. One major concern has to do with the interpretation and description of the behavior as it is reported. Like concerns about attributing content to animal minds, the worry is that there will be an over-attribution due to the bias of the scientist. In addition, if anecdotes are presented as stories, and the story structure is used to determine truth, there is a worry that such stories will be elaborated so as to remove inconsistencies or fill gaps (Mitchell 1997). The psychologist Richard Byrne defends the scientific use of rare events as a useful tool, writing, "careful and unbiased recording of unanticipated or rare events, followed by collation and an attempt at systematic analysis, cannot be harmful. At worst, the exercise will be superseded and made redundant by methods that give greater control; at best, the collated data may become important to theory" (Byrne 1997, 135). It is important to note that Byrne is not talking just about data acquired by the ethological method, but also incidents observed by scientist in the field or lab who are well-versed in the baseline behavior of the species.
3.2 Experimental Method
Before Watson and Skinner promoted behaviorism in human psychology, similar ideas were being espoused among animal cognition researchers who were disappointed in the lack of rigorous methods for studying animal minds. The psychologist C. Lloyd Morgan's early text on comparative cognition was critical of the anthropomorphism in the anecdotes reported by Darwin, Romanes, and others. In An Introduction to Comparative Psychology (1894), Morgan famously proposed what is now called Morgan's Canon; recall that Morgan's Canon is a principle of conservatism that instructs researchers to interpret behavior as caused by the lowest possible "psychical faculty" (Morgan 1894, 53). In his text Morgan suggested that many of the seemingly cognitively sophisticated behaviors of animals could be explained by associative learning.
The Clever Hans scandal of 1904 demonstrated Morgan's Canon in use; Hans was a famous Russian trotting horse who charmed crowds by appearing to calculate mathematical problems, as well as to read German and musical notation, simply by tapping his hoof (Candland 1993; Pfungst 1965). After much investigation, the experimental psychologist Otto Pfungst found that Hans wasn't counting or reading language; rather he was reading his owner von Osten's bodily motions. Von Osten was unconsciously cuing Hans to start and stop tapping his foot at the correct time, and Hans had merely leaned to associate von Osten's movements with the correct behavior. Today, the legacy of Clever Hans can be seen in the controls used during experimental tests of an animal's ability. For example, researchers who know the correct response will wear a welder's mask, blackened goggles, or some other device to keep the subject from being cued by eye gaze or facial expressions, or naive trainers are used during testing.
One of the leading animal cognition researchers during this time was Edward L. Thorndike (1874-1949). Thorndike argued for the necessity of experimental study of animal intelligence, writing, "most of the books do not give us a psychology, but rather a eulogy of animals. They have all been about animal intelligence, never about animal stupidity" (Thorndike 1911, 22). Experiments will help us learn both what animals can do and what they fail to do, thus giving us a better overall understanding of animal cognition. One of Thorndike's projects focused on the problem solving abilities of housecats. He placed cats in a variety of puzzle boxes, and observed the strategies the cats used to escape. When first put in a new box, the cats took a long time to find the solution, but after experience with the box they were able to escape much more quickly. Thorndike found that the cats improved by ignoring the ineffective actions and performing the useful ones. This suggested to Thorndike that cats learn through trial and error, and his conclusion helped to reinforce the belief that animal behavior can be fully explained in associative terms.
While the behaviorists succeeded in introducing much-needed rigor into the study of animal minds, there was some concern that they had gone too far, that the methods were too stringent, and that the drive for repeatable and controlled experiments could not be used to uncover all there is to know about the function of animal minds. For example, the ethologists thought that to understand animal behavior, animals must be observed in their natural environment. As sterile laboratory experiments stripped of social and environmental context, some considered the behaviorists' studies ecologically invalid.
While today some experimentalists defend these methods (Povinelli & Vonk 2004), other experimentalists agree with the criticisms, and in response have developed methods that are sensitive not only to environmental concerns, but also to development and social relations. For example, research coming out of Kyoto University's Primate Research Institute (PRI) is based on a three-part research program (Matsuzawa et al. 2006). First, chimpanzee physical, cognitive, and social development is taken into account in the design of experiments, and subjects are raised by their mothers rather than by human caregivers or unrelated animals. In addition, lab work and fieldwork is synthesized; field observations are used to develop experiments, and experiments are conducted both in the field and in the laboratory. Finally, the method includes analysis of the physiological and biological features of the species that could be related to cognitive abilities.
The experimental research at PRI uses what they call the "participant observation" method, which is based on the triadic social relationship between mother, infant, and experimenter. When testing chimpanzees in the lab, they are never taken from their natural social environment; rather, the experiments are brought into the social environment. As a researcher becomes a member of that social environment, she can run experiments that are woven into normal daily activities. At PRI, a different researcher is bonded with each mother-infant dyad, and the relationship is expected to last a lifetime. This close relationship between human and chimpanzee is thought to offer many benefits. It makes the chimpanzees more willing to engage in the research activities, so researcher can better understand of what chimpanzees can and cannot do (rather than what they will and won't do). In addition, Matsuzawa claims that the participant observation method is better at investigating species typical social cognition than are isolated experiments on single subjects, because the PRI chimpanzees are not integrated into a human social environment, but the researchers adapt to the chimpanzee social environment (Matsuzawa 2006a). Finally, the bond between researcher and subject allows the human to interact with his chimpanzee "research partners" at a younger age, given the trust between researcher and mother. Mother and infant can be taught a task together, which can help to illuminate developmental differences in particular abilities. For example, Inoue and Matsuzawa have recently reported that infant chimpanzees are better able to recall strings of numerals in order than are adult chimpanzees and humans (Inoue & Matsuzawa 2007).
While the behaviorists were performing laboratory experiments, ethologists such as Oskar Heinroth, Konrad Lorenz, Nikolaas Tinbergen, and Karl von Frisch were following animals into the field to observe naturally occurring behavior. The ethological method is based in biology, and takes into account not only the behavior, but also the context of behavior, the environment, and the physiology and evolutionary history of the animal.
Lorenz and Tinbergen were interested in analyzing the complex and rigid set of movements that make up a single act. They postulated that such movements, which they called fixed action patterns, are innate and caused by the existence of a releasing mechanism that responds to some external sensory stimulus. Ethologists study such acts at various organizational levels, e.g. at the individual, dyad, family group, and species level (Menzel 1969). To explain behavior, ethologists follow Tinbergen's suggestion that we can distinguish between explanations in terms of proximate causes, such as mechanism or function, and ultimate causes, such as ontogeny (the maturational processes involved in the behavior) and evolution (Tinbergen 1963).
But before an explanation for some behavior can be found, the behavior must be well understood in the context of species normal behavior. Thus, the ethologist will begin the study of a species by constructing an ethogram from field notes taken after many hours of observation (Brown 1975; Lehner 1996). An ethogram is a thorough catalogue of the characteristic behavioral units of a species, and each unit is given a verbal description and perhaps an image. Theoretical debates arise over the issue of how an ethogram should label and describe behavioral units. Behaviors can be described formally, and describe the action at the level of muscular contractions, or patterns of bodily movements (e.g. beak pecks ground). On the other hand, an ethogram could describe behaviors functionally, and place the behavior in a larger context by referring to the purpose or consequence of the behavior (e.g. eat) (Hinde 1970).
There are criticisms of both functional and formal methods of description. Formal descriptions may leave out important aspects of an animal's behavior, whereas functional descriptions are subject to over-interpretation and may lead to anthropomorphism, and they may conflate explanations in terms of ultimate and proximate reasons (see Allen & Bekoff 1997 for a discussion). Millikan (1993) and Allen and Bekoff (1997) provide philosophical defenses of relying on functional descriptions in ethology. While Millikan has claimed that ethologists should only be concerned with behavior as functionally described, Allen and Bekoff argue that the choice between a functional and formal description will vary on the context, depending on which is more useful. In many cases, functional descriptions will be preferred because of the advantages Hinde (1970) has identified. For one, behavior described functionally will result in fewer data sets, making for more robust data analysis. In addition, descriptions in terms of function are more informative than formal ones, given that they include information about the cause of the behavior or its consequence. Finally, behavioral changes can be described in terms of environmental changes.
However an ethologist decides to describe behaviors, the question of how to individuate a behavior arises (Skinner 1935; Russon et al. 2007). Descriptions of behavior can be finely grained, and refer to the specifics of a behavior, e.g. using a stone (or a leaf cup, or chewed leaves, or a hand, or fur, etc.) for drinking water from a river. On the other hand, behaviors can be roughly grained into larger behavioral units, e.g. using a drinking tool. If behaviors should be categorized to reflect the way the species organizes its behavior, then identifying behaviors requires first knowing what the species' internal organizational scheme is (Byrne 1999; Russon et al. 2007).
In addition to classical ethology, there is a field of study called cognitive ethology, founded by psychologist Donald Griffin. Cognitive ethology purports to study animal consciousness (Griffin 1984, 1985, 1992), which is often identified with the study of beliefs, intentions, self-awareness, deception, and theory of mind (Shettleworth 1998). Griffin's use of the word "consciousness" belies his greater interest in cognition, given that most of the topics he discusses (e.g. perception, memory, spatial cognition, language, tool use etc.) are cognitive (Griffin 1992). While "cognitive ethology" is often used to describe animal cognition research that uses the methodologies of classical ethology, it appears that few researchers welcome the label (Allen 2004).
3.4 Comparative Cognitive Neuroscience
A more recent methodology for studying animal cognition results from recent technological advances in brain imaging, and many current discussions of cognitive evolution relies on data from comparative neuroscience.
As neural correlates for cognitive abilities have been found in humans, neuroscientists have looked for parallel structures in other species. For example, while language is considered uniquely human, a distinct cortical area in the macaque monkey is thought to be a precursor to Broca's area in humans. Broca's area is necessary for proper speech production in humans, and the analogous area in macaques is used for controlling facial muscles. It is hypothesized that this area has the function of regulating control over facial expressions related to communication (Petrides et al. 2005).
Others argue that the mirror neuron system found in monkeys and humans may be partially homologous to Broca's area (Rizzolatti & Craighero 2004). The mirror neuron system becomes active both when the subject engages in a behavior and when the subject observes another engage in a similar action, and has been suggested as a neuronal basis for abilities such as theory of mind, empathy, and imitation (Goldman 2006).
Comparative cognitive neuroscience plays a significant role in discussions of the Social Intelligence Hypothesis. If large executive brains evolved in response to social challenges, then larger executive brains ought to be positively correlated with increased intelligence (Dunbar 1998). One problem with testing this thesis has been determining what counts as the executive brain. Comparative studies of brain size, relative brain size, encephalization quotient, and brain capacity have met with counterexamples (for example, the shrew's brain is much larger than the human brain, relatively speaking). It has been argued that the number of cortical neurons, and the thickness and density in the cortex is positively correlated to a greater information processing capacity (Roth and Dicke 2005).
4. Research Programs
The research programs in animal cognition are too numerous to thoroughly cover here; good introductory texts can be found that introduce a variety of topics (e.g. Shettleworth 1998; Roberts 1997; Pearce 1997). What follows is a brief introduction to three areas of research that have been of interest to philosophers: communication, theory of mind, and empathy. While these areas are those that have been discussed by philosophers, the choice of these paradigms is not meant to undermine the philosophical importance of other areas of research, such as concept acquisition, perception, causal reasoning, memory, culture, imitation, innovation, and so forth. And a final caveat: these three areas are characterized primarily by research on great apes (chimpanzees, bonobos, gorillas, orangutans, as well as humans), but this emphasis does not imply that philosophical issues don't arise in research with other taxa, or that these cognitive capacities only occur in great apes.
To communicate, an individual must be capable of engaging in some behavior that transmits information to others, and the individual must have some control over whether the behavior is exhibited. Communication is distinguished from signalling; while signaling, like communication, can cause a change in another's behavior, signaling only accomplishes this in an inflexible and rote way. Research on communication focuses both on the natural communication systems of animals in the wild, and on attempts to introduce novel communication systems to animals in a laboratory setting.
4.1.1 Artificial Symbolic Communication Studies
In the 20th century there was great interest in teaching symbolic communication systems to other species. The earliest forays into this area were with chimpanzees, and focused on teaching spoken language to chimpanzees raised as human children (Kellogg & Kellogg 1933; Hayes & Hayes 1951). With the realization that chimpanzees lack the vocal apparatus needed to form human words, research shifted to teaching chimpanzees American Sign Language and artificial symbolic communication systems. The first such study, Beatrix and Allen Gardner's Project Washoe, was initially reported to be a success. Using explicit training methods, including shaping, molding, and modeling, the researchers were able to train the infant Washoe to form at least 132 ASL signs. Focus was on production of gestures, rather than comprehension, and the Gardners' stated intention was to train Washoe (and later, other chimpanzees) in a social setting, mimicking the language-learning environment of children as much as possible. The Gardners claim that, "[Washoe] learned a natural human language and her early utterances were highly similar to, perhaps indistinguishable from, the early utterances of human children. Now, the categorical question, can a nonhuman being use a human language, must be replaced by quantitative questions: how much human language, how soon, or how far can they go?" (Gardner & Gardner 1978, 73).
While the Gardners' claims about ape language were being echoed by others working on ape language (e.g., Premack 1971; Patterson 1978), not everyone agreed. The psychologist Herbert Terrace, who used the methods of the Gardners to teach ASL signs to an infant chimpanzee named Nim Chimpsky, argued that the apes were not using the signs to communicate. Terrace concluded that some of the results achieved by the Gardners could be explained by associative learning rather than comprehension of the semantics of the symbols. In his study, he tried to control for associative learning, and his focus on syntax had him attend to symbol order in multi-symbol strings. While early results of this study seemed promising, after watching videos of Nim's symbol use he noticed that what had been initially seen as spontaneous utterances were often imitations of utterances just made by his trainers. Terrace reviewed films of Washoe's utterances, and found similar patterns: the teacher initiates the signing, and the chimpanzee mimics the teacher's signs. He also noted that the give and take rhythm of child-adult communication was not mirrored by the chimpanzee-trainer conversations, and took this difference in pragmatics as further evidence that the chimpanzees were not using language (Terrace et al. 1979).
Though the Gardners defended their studies against Terrace's critiques (Gardner & Gardner 1989), other researchers tried to control for alternative interpretations of their results. Premack, for example, relied on transfer tests as evidence that Sara understands the symbols she was taught (Premack 1971). In a transfer test, a new symbol is taught only in the context of a subset of the subject's vocabulary. Once the subject reaches criterion on the teaching set, a formal test is conducted using novel strings of symbols.
The post-Terrace research on symbolic communication has expanded to include different species, such as the other apes, dolphins, parrots, and sea lions. In addition, the focus of some studies has shifted from syntax to semantics, and from production to comprehension.
Advocates of this research program argue that the studies uncover something about the relationship between language and mind, the evolution of human language, and the roles played by development and scaffolding in human language (Lloyd 2004). However, to the critics, these studies are just more evidence of the power of association, and the ability of humans to train animals to do anything. There is a huge literature on these studies, with critics (Pinker & Bloom 1990; Pinker 1994; Chomsky 1980) as well as defenders (Lloyd 2004; Greenfield 1991; Savage Rumbaugh et al. 1998).
One area of contention has to do with whether animals who successfully use some aspect of human language are using it qua language, or are instead engaged in symbolic communication. At least three different demarcations between language and other symbolic communication systems have been offered. According to Noam Chomsky's original linguistic program, to use language is to embody certain structural principles, and all language users are able to produce a potentially infinite number of grammatical strings via recursive embedding (Chomsky 1968). The linguistic anthropologist Charles Hockett identified up to seventeen design features that occur in every human language, including semanticity, discreteness, and arbitrariness (Hockett 1977). More recently Hauser, Fitch, and Chomsky (2002) argue that the mechanism that allows for recursive thinking is the central cognitive requirement for language, and is a feature of human communication systems not found in other species.
Chomsky was a vocal critic of early animal language studies, especially of the claims made by some researchers that the apes had acquired language. For Chomsky, language requires syntax, something that is lacking in all the communication systems of the apes. Furthermore, to train an ape to use symbols is a laborious process, whereas children learn language effortlessly. Language is innate, according to Chomsky, so if apes had the capacity for learning language, they would speak without human intervention (Chomsky 1968). Chomsky often states his criticism as an a priori argument against animal language: "if an animal had a capacity as biologically sophisticated as language but somehow hadn't used it until now, it would be an evolutionary miracle, like finding an island of humans who could be taught to fly" (cited in Lloyd 2004, 585).
Another argument Chomsky has offered against animal language is based on the dissimilarity between animal communication systems and human language. He writes, "The question of whether other systems are 'like' human language is a question about the usefulness of a certain metaphor" (Chomsky 1980, 434), and he argues that the structural principles, manner of use, and ontogenetic development of ape symbol use is so different from human language that any analogy between the two would be very weak. Those who defend the animal symbolic communication system as language take Chomsky to task on this point, and stress the similarities between the two systems of communication.
Given recent findings in genetics, the biological capacity for language may be more accurately described as a collection of biological capacities, some of which we share with other species. The FOXP2 gene is found to play a role in speech production, and some claim that it was instrumental in the development of language in humans. The FOXP2 gene is also expressed in the same part of the brain in zebra finches, and it has been reported that finch fledglings with reduced FOXP2 are impaired in their ability to learn to sing (Haesler et al. 2007).
There appears to be less concern about describing animal symbol use as communicative, though many theories also portray communication as carrying large cognitive demands. For example, according to H.P. Grice's classical pragmatics view of communication, for an individual to make a meaningful utterance, the speaker must intend to change the belief of the audience, and the audience must recognize the speaker's intention (Grice 1969). This strong requirement for communication requires metacognition on the part of communicators, including a theory of mind, or the ability to attribute mental states such as beliefs and desires to another. As we will see in section 4.2.1, claims about theory of mind in nonhuman animals are cause for much skepticism. A weaker reading of Grice's criteria might limit the requirement of communication to a recognition of intentionality, something which is rather less controversial.
4.1.2 Natural Communication
In The Descent of Man, Darwin claims that communication can be seen in many species: in the alarm cries of monkeys, the distinctly meaningful barks of dogs, and the songs of birds that that are learned from adult experts (Darwin 1871). Darwin's evidence took the form of anecdotes, but soon formal studies of natural animal communication systems were attempted. One of the earliest projects was that of Richard Lynch Garner, who, convinced that primate vocalizations are linguistic in nature, intended to learn the natural language of the apes. Garner traveled to central Africa in 1892 and lived in a cage in the forest, to protect himself from the chimpanzees and gorillas he was observing. Since the time of Garner's failed research, ethologists, biologists, and psychologists have documented the natural communication systems of hundreds of species.
Though these systems are described as communication, the central theoretical questions are whether the communicative utterances are referential and whether the utterers are mentally representing the referent, that is, whether the utterance is meaningful from the perspective of conspecifics. These questions are applied just as well to the symbolic communication research discussed above.
One possibility is that vocalizations such as alarm calls could be the result of an involuntary reaction to stimuli, an emotional display, or an indication of the severity of the threat. On the other hand, alarm calls might be meaningful signals can be distinguished from involuntary behavioral responses. Those who argue that animal communication systems and human language are homologous (functionally similar due to common evolutionary origin) or analogous (functionally similar with different evolutionary origins) to one another have attempted to demonstrate that some animal signals are referential. One test for referential communication is to see if the behavior is flexible by determining whether there is only a probabilistic relationship between the stimulus and response. For example, if there are different responses to different utterers (e.g. infant vs. adult, dominant vs. submissive), this is thought to demonstrate flexibility in the behavior that is suggestive of referential understanding (Evans 2002; Tomasello & Zuberbühler 2002). In addition, it is thought that referential calls will encode specific information about the predator and that animals who hear the alarm call perceive that encoded information (Evans et al. 1993a, Evans 1997). Marler et al. (1992) offer two criteria that must be met for a signal to be functionally referential. The production criterion requires that all the stimuli that elicit the signal can be said to belong to one category, either a general category such as "aerial predators" or more specific one such as "eagle." The perception criterion states that utterance of the referential signal is alone sufficient to elicit the same behavior as would be elicited by perceiving the referent (Marler et al. 1992).
Given these criteria, Marler and Evans examined the anti-predator behavior of bantam chickens, and found that the chickens reliably give different alarm calls to aerial predators and ground predators (Evans et al 1993b; Evans & Marler 1995). Because they also behave differently toward the two different predators, Marler and Evans suggest that the alarm cries functionally refer to the kind of predator approaching. When a chicken emits a scream after seeing a hawk, they claim the chicken is referring to the hawk, rather than expressing fear of the hawk, or ordering conspecifics to take cover, crouch, and look up to the sky.
Alarm calls and other communicative vocalizations that fulfill these requirements are found in many species. Gunnison's prairie dogs, for example, give different alarm calls to humans, hawks, and dogs/coyotes. In response to the hawk alarm call, only the prairie dogs that are in the flight path of the hawk respond, running into a burrow. The human alarm call elicits a community wide flight into the burrows, whereas the dog/coyote alarm call leads all individuals to run to the edge of the burrow and stand erect (Kiriazis & Slobodchikoff 2006). Vervet monkeys also give alarm calls for different predators. Following up on the field observations of zoologist Thomas Struhsaker (1967), Dorthy Cheney and Robert Seyfarth used playbacks of prerecorded alarm calls to demonstrate that when a leopard alarm is sounded, the vervets run into trees, where they are safe from the leopards due to the monkeys' agility in jumping from tree to tree. When an eagle alarm is sounded, monkeys look up and run into bushes. When the snake alarm is sounded, the monkeys stand bipedally and peer into the grass around them (Cheney & Seyfarth 1990). Other species found to have different alarm calls and different behavior for different species of predator include Diana monkeys (Zuberbühler 2000), Campbell's monkeys (Zuberbühler 2001), and meerkats (Manser 2001; Manser et al. 2001). In addition, ground squirrels (Owings & Hennessy 1984), tree squirrels (Green & Meagher 1998), and dwarf mongooses (Beynon & Rasa 1989) are all known to have alarm calls that distinguish between terrestrial and aerial predators.
The emphasis on alarm calls isn't to suggest that this is the only area in which other species are said to use referential communication. Many other calls and gestures are thought to involve referential communication. For example, the food calls of chimpanzees are thought to indicate not only the presence of food, but also the location or quality of food (e.g. Slocombe & Zuberbühler 2005, 2006). Similar findings have been reported for chickens (Evans & Evans 2007). Bottlenose dolphins are said to refer to themselves via their signature whistle (Janik et al. 2006). In addition, we shouldn't expect that all communication proceeds through calls or other vocalizations; in some species olfaction or echolocation may be more salient. Allen and Saidel suggest that empirical work can be done to determine the kinds of referential communication different species can engage in, and that such research can help us to better understand the correct description of the mechanisms subsuming human language (Allen & Saidel 1998).
4.2 Theory of Mind and Metacognition
Metacognition is generally understood as cognition about cognitive processes, i.e., thoughts about thoughts. One can engage in metacognition when thinking about one's one thoughts, desires, etc. or by thinking about the mental processes of others. Theory of mind, or mindreading, is associated with the latter ability. Since the late 1970s, philosophers and scientists alike have been pursuing this research program.
4.2.1 Theory of Mind
Like humans, many species are social animals who, in addition to navigating a physical world, must also navigate a social world. In the 1970's it was suggested that in order to succeed in a social world, individuals would benefit from having some understanding of the mind of others. The psychologist Nicholas Humphrey was one of the first to propose that social knowledge requires theoretical knowledge of the causes of behavior (Humphrey 1976, 1978). Humphrey wrote, "...I venture to suggest that if a rat's knowledge of the behaviour of other rats were to be limited to everything which behaviourists have discovered about rats to date, the rat would show so little understanding of its fellows that it would bungle disastrously every social interaction it engaged in; the prospects for a man similarly constrained would be still more dismal" (Humphrey 1978, 60).
The term "theory of mind" was introduced by psychologists David Premack and Guy Woodruff the around the same time, and they make the same assumption regarding social cognition. The specific question Premack and Woodruff were interested in was whether the chimpanzee attributes beliefs and desires in order to predict and explain behavior, something they assumed that humans do. In effect, Premack and Woodruff wanted to know whether a chimpanzee is a Humean action theorist who understands the behavior of others as being caused by propositional attitudes. Thus, they defined theory of mind as "the ability to predict and explain behavior by attributing mental states." Premack and Woodruff attempted to determine whether Sarah, the same chimpanzee from Premack's symbolic communication project, has a theory of mind. To examine whether Sarah understands what others believe, they used the following paradigm: Sarah was shown videotapes of humans trying to solve certain tasks (e.g. acquiring out of reach bananas, warming up a cold room by lighting a heater) and she was supposed to choose from an array of photos to pick the solution (Premack & Woodruff 1978). Because Sarah picked the correct photograph at an above-chance level, Premack and Woodruff concluded that she has a theory of mind. They claimed that Sarah must have been attributing "at least two states of mind to the human actor, namely, intention or purpose on the one hand, and knowledge or belief on the other" (Premack & Woodruff 1978, 518).
In commentary on this study, it was pointed out that Sarah could have used other methods to solve the problems. She could, for example, have attended to the goal of the actors, as opposed to their mental states (which is the interpretation that Premack now endorses (Premack & Premack 2003)). Most of the commentators were unconvinced by the design of the study, and several suggested alternative methodologies for examining the question. One suggestion was to require the subject to solve a coordination problem. To succeed in a coordination problem the subject would have to alter his own behavior in expectation of what another will do (e.g. Bennett 1978; Dennett 1978). Dennett suggests that a good coordination problem might require that the subject considers another's false belief, so that the behavior that is predicted will be an unusual one, such as a behavior that would only be taken if the actor had a false belief. A false belief coordination problem would avoid alternative interpretations having to do with identifying the actor's goal, or making associations from similar situations in the past. The behavior performed by an actor who has a false belief will not achieve the actor's goal, and will probably not be something the subject has witnessed previously. The main problem with this suggestion, Dennett notes, is how to determine the content of the predictions a chimpanzee might make.
Given the difficulties associated with developing a good nonverbal test for theory of mind (Dennett 1983), Dennett's suggestion was taken up by researchers interested in studying theory of mind in children (Wimmer & Perner 1983). Wimmer and Perner accepted Premack and Woodruff's definition of theory of mind and asked when the small child gains a theory of mind. To answer that question, they designed the false belief task, which was to become a standard test for theory of mind. Children watched a show in which a puppet named Maxi put away a piece of chocolate in a box before leaving the room. While Maxi was out, his mother found the chocolate and moved it to a cupboard. Maxi returns to the scene, the show is stopped, and children are asked to predict where Maxi will go to look for his chocolate. If the child says Maxi will look in the cupboard, she shows that she doesn't have a theory of mind. If the child says Maxi will look in the box, she passes, and is thought to have a theory of mind because the child demonstrates that she can attribute mental states and use them to predict Maxi's behavior.
The theory of mind research program was closely associated with a debate on folk psychology between folk psychology as theory (the standard view that human knowledge of other minds is theoretical in nature), and folk psychology as simulation (the view that our knowledge of other minds relies on using our own mind as a model). In the late 90's there was a growing acceptance that both theory-theory and simulation theory were partially right and partially wrong, which culminated in a general acceptance of some sort of hybrid theory (e.g. Nichols & Stich 2003; Goldman 2006). These arguments make use of empirical data from both the developmental and the animal cognition literature.
During this time, there were various attempts to uncover theory of mind in animals using nonverbal paradigms, without much success (Heyes 1998). Given the subsequent theoretical and definitional disagreements, some researchers have concluded that "the generic label 'theory of mind' actually covers a wide range of processes of social cognition" (Tomasello et al. 2003b, 239). The theory of mind research paradigm in animal cognition subsequently shifted from attempts to come up with a nonverbal false belief task, toward more specific questions about cognitive capacities like understanding others' perceptual states (Hare et al. 2000), goals (Uller 2004), or intentionality (Tomasello 2005).
Like belief states, other's perceptual states are opaque, and require the attributor to make a distinction between oneself and another. And like knowing someone's beliefs states, knowing another's perceptual state can lead to predictions about future behavior. Ethological evidence that chimpanzees monitor gaze and modify their behavior when they are visible to others (e.g. Plooij 1978; Whiten & Byrne 1988; Goodall 1986) was taken as some evidence that chimpanzees can attribute perceptual states to others, and experimental researchers decided to design studies to determine whether chimpanzees understand seeing.
The results of early laboratory studies were mixed; David Premack's research suggested that chimpanzees do understand seeing (reviewed in Premack & Premack 2003), whereas studies by Povinelli and Eddy (1996) challenged that conclusion. Later studies suggested that chimpanzees understand both seeing and intentionality (Hare et al. 2000; Hare et al. 2001). In Hare et al.'s experimental set-up, a subordinate and a dominant chimpanzee are released in a room baited with food. Normally, if both animals can see the food, or saw one another witness the baiting, a subordinate animal will avoid the food and allow the dominant access. However, in these experiments, when the food is occluded from the dominant's view, the subordinate will approach it. Only if the dominant can see the food or the baiting will the subordinate avoid it. The animals are across the room from one another, so the subordinate has to consider the visual perspective of the dominant in order to judge correctly whether he can see the food or not. Because it seems that the subordinate is able to make different judgments about whether to seek out the food based only on whether it is visible to the dominant, this study is thought to indicate that the apes understand the mental state of seeing.
Povinelli and Vonk (2004) criticize the Hare et al. studies, suggesting that the ecological nature of the study (using food competition behavior from the subject's natural repertoire) is a weakness of the study, not a strength as Hare et al. believed. Povinelli and Vonk argue that the subordinate chimpanzee need not have any understanding of the dominant's mental state, but could instead be operating on a theory of behavior. In response, the authors claim that they have accounted for all possible alternative explanations for the subordinate's behavior, making an inference to the best explanation argument that the subordinate understands what the dominant sees (Tomasello et al. 2003a, 2003b). Further evidence is provided by a study of chimpanzee hiding behavior (Hare et al. 2006). The debate is an epistemological one regarding how best to determine from evidence whether an animal postulates a mental state (Andrews 2005).
While the research on understanding what conspecifics can see has focused on chimpanzees and to a lesser extent other great apes, there are a few studies on other species. Research on scrub-jay caching behavior shows that individuals who have pilfered another's cache in the past will privately recache food when a conspecific observed the original caching, but not if the original caching was unobserved (Emery & Clayton 2004). Naive scrubjays did not recache. Emery & Clayton suggest that the jays who do recache are engaging in what they call "experience projection. . .they relate information about their previous experience as a pilferer to the possibility of future stealing by another individual, and modify their recovery strategy appropriately" (Emery & Clayton 2004, 1905). Experience projection, they suggest, relates to theory of mind.
Studies on perceptual understanding were also done with rhesus monkeys (Flombaum & Santos 2005). Like the chimpanzee and the scrubjay studies, these experiments set up a naturalistic competitive situation in which the subject had to predict the behavior of a competitor. In one version of this study, rhesus macaques from the island of Cayo Santiago were pitted against human competitors in a foraging task; two experimenters would approach a lone monkey, and each would situate himself differently so that the monkey was visible to one experimenter but not the other. Both experimenters had one grape. Flombaum and Santos found that monkeys were more likely to steal grapes from the experimenter who couldn't see them. They found similar results for audibility; when given the choice of stealing a grape in a transparent box covered with bells, or a grape in a transparent box that was free of noisemakers, the monkeys preferred the silent food when no one was looking at them. However, when it was obvious that the monkey was observed, there was no preference for stealing quiet over noisy grapes (Santos et al. 2006).
Other studies in this area focus on an understanding of intentionality and goal directed behavior. For example, Claudia Uller used a dishabituation paradigm designed to test for understanding of goal directed behavior in human infants to test infant chimpanzees (Uller 2004). Recall that dishabituation paradigms involve showing the subject a stimulus until it no longer holds the subject's attention, and then showing them one of two alternative stimuli. If one of the new stimuli gets more attention than the other, it is thought that the subject sees that stimulus as different from the original target stimulus. Gergely and colleagues used this method to study goal attribution in human infants: 12-month-old infants were habituated to a video of a small ball leaping over a barrier to reach a large ball (Gergely et al. 1995). After habituation, the infants were shown either the identical behavior with the barrier removed (e.g., the small ball moving toward the big ball and then jumping in the air before meeting the large ball) or a different behavior in which the small ball moved directly toward the large ball. Children looked longer at the first condition, and the authors concluded that infants perceive the small ball's action as goal-directed. Using the same paradigm, Uller found that chimpanzee infants respond at the same rate as human infants, but she suggests more evidence is required before claiming that chimpanzees understand intentions (Uller 2004).
Another recent study of understanding intentionality also compared the behavior of chimpanzees and human children (Warneken & Tomasello 2006). In a naturalistic social setting, subjects were nonverbally requested to help the experimenter achieve goals, such as picking up a dropped sponge or opening a box. They found that 18 month-old children and chimpanzees both respond to simple requests (e.g. picking up a dropped object), but that children are also able to respond to more complex requests. Earlier studies by Call et al. also demonstrated that chimpanzees are able to distinguish between a person who is unwilling to perform a task, and one who is unable (Call et al. 2004).
While chimpanzees show some sensitivity to intentions and goals, domestic dogs may be even more attuned to the intentions of humans. Dogs are able to use the gaze of a human in order to determine where food is hidden, an ability not demonstrated in the chimpanzee (Hare et al. 1998; Hare & Tomasello 1999; Miklosi & Topal 2004; Brauer et al. 2006). Dogs appear to be sensitive to eye gaze in humans, and often make eye contact before initiating play. One explanation for dogs' social acuity is that in selecting for traits that make dogs better human companions, humans inadvertently bred dogs who are better able to pass theory of mind tasks (Hare et al. 2002).
Research on metacognition aims to explore what individuals know about their own minds. One area of much attention has been mirror self-recognition (MSR). In this paradigm, developed by psychologist Gordon Gallup, subjects are surreptitiously marked and then given a mirror. "Passing" MSR involves touching the mark more frequently when there is a mirror available than when there is not. Gallop theorized that passing MSR entails that the animal has a concept of self (Gallup 1970), though others dispute this claim. While it was once thought to be a rare behavior, limited to some of the great apes, today many species have been studied and at least some positive results have been reported for the following species:
Many other species failed to show mirror directed behavior, including some monkey species, which suggests to some that there is a corresponding cognitive mechanism that the above species, but not others, enjoy. However, it has been pointed out that there are ecological and biological constraints on this test; not all species are visually oriented, and some find eyes aversive (this was the explanation for studies that failed to show MSR in gorillas). For a discussion of these issues, see the collection of articles in Self-awareness in Animals and Humans (Parker et al. 1994).
Uncertainty monitoring is another area of research aimed to investigate understanding of one's own mental state (for a review, see Shettleworth & Sutton 2006). Subjects who know what they do and do not know demonstrate metacognition about their epistemic states, and several nonverbal tests for uncertainty monitoring have been developed for use with different species. The paradigm might go as follows: subjects are trained to indicate whether a stimulus is the same as or different from a sample. When the subjects get the answer correct, they are rewarded with food, but food is taken away when they get the wrong answer. Once the subjects are trained on this task, the paradigm is modified to introduce a "bail out" key with the function of starting a new trial without supplying either a reward or punishment. Interspersed with the easy stimuli are ambiguous stimuli that the subject is unable to accurately categorize above a chance level. If the subjects learn to chose the "bail out" key when they are uncertain, it is thought to indicate that the subjects are aware of their epistemic state. It has been reported that many species choose the "bail out" key in such a way as to maximize rewards, including dolphins (Smithe et al. 1995), rhesus monkeys (Hampton 2001), great apes (Call & Carpenter 2001), human infants (Call & Carpenter 2001). Mixed results have been reported with pigeons (Sole et al. 2003).
Though such tests have been designed to test for metacognition, Peter Carruthers argues that animals can come to solve the problems without engaing in second-order reasoning. He suggests that the animal could be operating over beliefs and desires of different strengths, and that standard practical reasoning systems can be used to output different responses to the different permutations of weak and strong beliefs and desires (Carruthers forthcoming).
4.3 Emotion and Empathy
Social cognition isn't limited to knowing the reasons another has for acting; it also involves understanding the emotions of others. While out of vogue for some time, researchers are again attending to emotions in different animal species, and the ability of some species to perceive the emotions of others. Studying emotion in other species is an important part of studying cognition because knowledge of another's emotional state allows an actor to respond differently in otherwise similar situations. Among social species, awareness of others' emotions is thought to play an important role in regulating social interactions, coordinating behavior, forming bonds between mothers and infants, as well as in forming short-term coalitions and long lasting relationships.
At least since Darwin's The Expression of the Emotions in Man and Animals, facial expressions have been of interest because they can indicate individuals' affective states. Research from the field and the lab suggests that different facial expressions have the same kinds of functions for chimpanzees and humans (e.g. van Hoof 1967, 1972; Goodall 1986; Parr 2001). Just as Paul Ekman argued for universality in emotional expressions among humans across cultures (Ekman et al. 1969), animal researchers have argued that at least some human emotions are also found in chimpanzees, and that chimpanzee facial expressions are homologous to human facial expressions in morphology and function. For example, van Hoof has argued that the bare-teeth display of chimpanzees is homologous to the human smile (van Hoof 1973).
Lisa Parr's research demonstrates that chimpanzees, like human infants, are able to categorize facial expressions associated with different emotional responses. Using a match-to-sample paradigm, Parr and colleagues have shown that chimpanzees recognize at least five different facial expressions: bared-teeth display, scream, pant-hoot, play face, and relaxed-lip face (Parr et al. 1998). Given the salience of these facial expressions to chimpanzees, Parr and colleagues have argued that facial expressions are important behaviors for regulating social relations (Parr & de Waal 1999; Parr et al. 2000). Current research in Parr's lab is focused on the development of ChimpFACS (Facial Action Coding System ' see Other Internet Resoures) modeled after Ekman's work in emotion in human facial expressions. They are using the ChimpFACS to construct models of chimpanzee expressions in order to determine the configuration of muscle movements that the chimpanzees find salient in their perception of emotion.
Another area of study in animal emotion focuses on stress. As a response to potentially dangerous situations, stress is thought to be an adaptive emotion in the short term for humans and other animals. Stress is measured physiologically via levels of glucocorticoid hormones such as cortisol. In humans cortisol levels are correlated with stress levels, and researchers have studied stressors such as dominance and status ranking in a number of different species (Sapolsky 2005; Abbott et al. 2003). For example, among baboons stress in the form of elevated glucocorticoid levels has been documented in females for about a month after the death of a close relative, in nursing mothers when a potentially infanticidal immigrant male arrives, in females when the female ranking system is undergoing instability, and in males when the male ranking system is unstable (Cheney & Seyfarth 2007). However, baboons do not appear to experience stress in the face of another's stress. Females are not concerned about male rank instability, and the death of a cohort's infant does not raise stress levels in anyone but the mother. It is suggested that for the baboon, stress is personal and egocentric, and the lack of sensitivity to the stress of others might be indicative of a lack of empathy as well (Cheney & Seyfarth 2007).
4.3.2 Empathy and Morality
Some research in moral psychology suggests that empathy is a necessary component of moral agency, and animal cognition researchers have been examining whether any other species share this ability with humans. Empathy is thought to require the same cognitive sophistication as does understanding another's mental state or intention, but in addition it requires an affective response to that mental state. While the terms 'empathy' (sharing the mental state of another) and 'sympathy' (a friendly feeling in response to another's mental state) have distinct meanings in the history of psychology, folk psychology, and ethical theory, in the animal cognition research the senses are often blurred and the terms used interchangeably.
Early reports of chimpanzee empathy came from Russian comparative psychologist N. N. Ladygina-Kohts, who raised a chimpanzee named Joni in the early 20th century (Ladygina-Kohts 2002). More recently, Sanjida O'Connell analyzed thousands of qualitative reports of primate responses to the distress of others, and her results suggest that apes give complex responses in the face of others' emotions, compared to the responses of monkeys in similar situations (O'Connell 1995). Studies of chimpanzee behavior performed by Frans de Waal and his colleagues suggest that chimpanzees understand emotions, and respond to different emotional states with different behavior, e.g. consoling the looser of a fight, helping, etc. De Waal takes these behaviors to be evidence of empathy in chimpanzees (de Waal 2006).
Much of the research on empathy in other species examines helping behavior. Helping is ubiquitous among humans, even when it requires that the actor suffer a cost, and even when the recipient is a stranger. But because helping can be engaged in without empathy, and empathic helping requires additional cognitive resources (e.g. knowledge about how to help someone achieve their goal and the motivation to act on that knowledge), it is important to understand the limitations of these studies.
Reports of naturalistic animal behavior suggest that many nonhuman animal species engage in prosocial behavior that may be empathic or proto-empathic in nature (de Waal 1996). De Waal often presents the famous example of Binti-Jua, a female gorilla at the Brookfield zoo in Chicago, who made the news when she rescued a 3 year-old human boy who fell into her habitat. Binti-Jua cradled the boy in her arms before handing him to a zookeeper. However, critics of the prosocial interpretation of Binti-Jua's behavior suggest that given her early exposure to doll play, an associative learning explanation is also possible.
Others claim that what may look like prosocial behavior may instead be a way of eliminating aversive stimuli. For example, research on rats and rhesus monkeys has shown that both species will cease eating when doing so causes shocks to a conspecific in an adjoining cage (Masserman et al. 1964). Masserman reports that one rhesus monkey almost starved himself to death to avoid shocking another. Alternatively, helping behavior among kin may be explained noncognitively as biological altruism. To determine whether other species engage in helping behavior that cannot be explained by other mechanisms, researchers have developed paradigms to determine whether chimpanzees display helping behavior to unrelated individuals. Chimpanzees are thought to be an especially good species to examine, given the range of cooperative behaviors they naturally perform, such as hunting (Boesch 2002), border patrolling (Mitani 2002), and coalition building (de Wall 1982). Cooperation among chimpanzees (Hirata 2003; Melis et al. 2006) and bonobos (Hare et al. 2007) has been demonstrated in a food-sharing task, but chimpanzees are thought to cooperate only when the dyads are generally tolerant of one another (Hare et al. 2007).
Another set of experiments suggests that chimpanzees are indifferent to the desires of others, and will not help another access food even when it requires no additional effort on the actor's part (Silk et al. 2005). The experimental setup allowed the actor to pull one of two ropes, one of which delivered food to only the actor, and the other of which delivered food to an adjacent chimpanzee as well as the actor. No preference was found for pulling the rope that rewarded both animals, even though the actor could see and hear the other chimpanzee. There was no significant correlation between another animal being present and which rope was pulled. Silk and colleagues conclude that "The absence of other-regarding preferences in chimpanzees may indicate that such preferences are a derived property of the human species, tied to sophisticated capacities for cultural learning, theory of mind, perspective taking and moral judgment" (Silk et al. 2005, 1359).
In contrast to Silk and colleagues' findings, several studies coming from the Max Planck Institute indicate that chimpanzees may be willing to help others. In one set of studies Warneken and Tomasello compared the helping behavior of 18 month-old human infants and chimpanzees, and found that while both subjects helped an adult human male retrieve dropped or out of reach objects, the infants were more likely to help in other contexts as well. The authors explain the disparity between Silk et al.'s results and their own by suggesting that the chimpanzees are distracted by the presence of food; they suggest other-regarding preferences in chimpanzees are overridden by the opportunity to gain food for themselves (Warneken & Tomasello 2006). A second study led by Warneken demonstrated that when a chimpanzee wasn't able to receive food for himself, he would reliably help another chimpanzee acquire food by opening a door (Warneken et al. 2007).
Insofar as these studies indicate empathy in a chimpanzee, one would expect to find a number of related cognitive capacities, including a sophisticated social cognition that allows for the reading of goals or emotions, and a causal reasoning capacity that allows the animal to offer assistance. Whether the current findings demonstrate that chimpanzees display altruisim, as is sometimes claimed (de Waal 2006), or whether having a full fledged theory of mind is a necessary part of our moral psychology (Hauser 2005), are issues that require conceptual analysis as well as empirical investigation of prosocial behavior in humans and other animals.
5. Animal Cognition and Philosophy: What Next?
Today we are in a kind of golden era when it comes to animal cognition research. Different species are being studied in the field and in the lab, and the results of these studies may be relevant to areas of philosophy including action theory, agency, belief, concepts, consciousness, culture, epistemology, ethics, folk psychology, imagery, language, memory, mental causation, mental content, modularity of mind, perception, personal identity, practical reason, rationality, and so forth. It seems that every day a new report is released, and many of these may have some theoretical implications.
Of course, scientific reports must be examined carefully to distinguish between methodologically solid findings and unwarranted interpretations, and it goes without saying that popular media reports of these studies are sometimes misleading. The epistemology of animal cognition research is ripe for investigation. Animal cognition is studied by psychologists, anthropologists, biologists, zoologists, neuroscientists, and ecologists, among others, and while some textbooks aim to integrate these different disciplinary approaches (e.g. Shettleworth 1998), and many extremely clever research methods have been used to test hypotheses, there is no unifying principle that brings together all those working under the umbrella of animal cognition. For example, psychologists tend to endorse the old adage: "It's not what you do, it's how you do it" to describe their interest in mechanism, while the ecologists' focus on adaptive problems may be less interested in mechanism, since solving adaptive problems such as acquiring food and avoiding predators might involve several cognitive processes.
Philosophy of animal cognition as sub-field of philosophy of science is one place where these methodological questions can be examined. For further reading in this area, see Colin Allen and Marc Bekoff's classic book Species of Mind: The Philosophy and Biology of Cognitive Ethology. For an introduction to an evolutionary psychology approach to studying animal cognition, see another classic, Sara Shettleworth's Cognition, Evolution, and Behaivor. And one more classic, from ethology, is Philip Lehner's Handbook of Ethological Methods.
Abbott, D. H., Keverne, E. B., Bercovitch, F. B., Shively, C. A., Mendoza, S. P., Saltzman, W., Snowdon, C. T., Ziegler, T. E., Banjevic, M., Garland, T., Jr., & Sapolsky, R. M. (2003). "Are Subordinates Always Stressed? A Comparative Analysis of Rank Differences in Cortisol Levels Among Primates." Hormones and Behavior, 43: 67-82.
Allen, C. & Hauser, M. (1996). "Concept Attribution in Nonhuman Animals: Theoretical and Methodological Problems in Ascribing Complex Mental Processes." In M. Bekoff & D. Jamieson (Eds.), Readings in Animal Cognition. Cambridge MA, MIT Press. 47-62.
Allen, C. & Bekoff, M. (1997). Species of Mind: The Philosophy and Biology of Cognitive Ethology. Cambridge MA, MIT Press.
Allen, C. & Saidel, E. (1998). "The Evolution of Reference." In D. Cummins & C. Allen (Eds.), The Evolution of Mind. New York, Oxford University Press. 183-202.
Allen, C. (1999). "Animal Concepts Revisited: The Use of Self-Monitoring as an Empirical Approach." Erkenntnis, 51(1): 33-40.
Allen, C. (2004). "Is Anyone a Cognitive Ethologist?" Biology and Philosophy, 19: 589-607.