On Crovidae cognition: An analysis of intelligence behavior in the family Corvidae

This paper was originally written as a final project for my Neurobiology and Behavior course.


Bird species of the family Corvidae live in complex social groups and demonstrate cognition and intelligence superior to other avian species. Indeed, Corvidae rival (and in some cases outperform) nonhuman primates in many measures of intelligence. Research is uncovering the extent and nature of Corvidae cognition. In this paper I discuss recent findings that show species in this family can purposefully manipulate objects for tool fabrication and use as well as engage in social learning. Furthermore, some Corvidae species can infer the mental state of their partner and pass the mirror-test – two abilities that are believed to precede the development of Theory of Mind.


In many folk stories, the Corvidae are depicted as clever, intelligent birds. But is this reputation warranted? Many recent studies seem to indicate that, yes, Corvidae are intellectually superior to other avian species and even rival many nonhuman primates (Emery, 2004).

The study of intelligence is difficult. It is a complex, context-dependent phenomena influenced by one’s weltanschauung – or world view. Nevertheless, intelligence can be defined and then rigorously studied.

In this paper I use the definition provided by Nathan J. Emery that intelligence depends on “causal reasoning, flexibility, imagination, and prospection” (2004) to highlight four types of Corvidae intelligence: 1) tool use, 2) social learning or cultural transmission, 3) inferring mental states, and 4) self-recognition.

Tool use

Perhaps the most well known example of Corvidae intelligence come from recent findings that document innovative and novel tool use. Tool use, defined by Beck, is “the external employment of an unattached environmental object to alter more efficiently the form, position, or condition of another object, another organism, or the user itself when the user holds or carries the tool during or just before use and is responsible for the proper and effective orientation of the tool” (Beck, 1980 as cited in Bird, 2009).

Because tool use – and to a greater extent, tool fabrication – requires the ability to problem solve, plan ahead, and understand and manipulate forces, it has been suggested as an indication of intelligence. While indicative, such indications do not fully demonstrate the “causal reasoning, flexibility, imagination, and prospection” that defines complex cognition and intelligence (Emery, 2004). Tool use could result from cognitive specialization molded by environmental pressures and as such, would merely reflect clever adaptation. In order for tool use to serve as an intelligence assay, it would need to demonstrably satisfy the above definition.

In 2002 a New Caledonian (Corvus moneduloides) crow named Betty, spontaneously and without prior training (though she had previously used a metal hook tool) bent a piece of wire and used it to collect a food reward. Intentional manipulation of objects to solve specific, novel problems is rare – even for tool using animals (Weir, 2002). Indeed, unless shown the technique, chimpanzees fail to unbend piping (exactly the opposite task spontaneously accomplished by Betty) to retrieve a food reward (Povinelli, 2000 as cited by Weir, 2002).

The novel and spontaneous fabrication and use of a hook tool by the New Caledonian satisfies the definition of complex cognition outlined by Emery. However, it is worth noting that the New Caledonian is a native tool user, known to utilize tools in the wild. Could spontaneous tool use be demonstrated in a species that does not natively use tools?

Rooks (Corvus frugilegus), despite having physical intelligence that rivals primates, do not use tools in the wild (Bird, 2009). Because Rook tool use is not observed in the wild, the ability to do so in captivity would indicate generalized intelligence not merely an adaptation to specific evolutionary pressures.

Researchers presented a completely novel apparatus that contained a worm on an out of reach platform that could only be retrieved by collapsing the platform and thereby dropping the worm to an accessible position. During a very limited training phase (5 trials), a stone was placed near the opening of the tube (positioned above the platform). Rooks presumably nudged the stone accidently the first time. However, having learned this, Rooks were able to solve nine different tool-use problems (Bird, 2009). A few examples are highlighted here.

When the apparatus was presented without a stone tool, Rooks left the testing environment, entered the aviary, collected a stone, and returned to complete the task. Because the tube size varied between small and large, it was necessary for Rooks to remember what size tube was provided. Rooks selected small stones (4.1 ± 0.7 g) and large stones (7.6 ± 0.8 g) according to tube presentation (Bird, 2009).

How well the Rooks adopt a new tool in accomplishing the task indicates cognitive flexibility. When presented with a stick in lieu of a stone, all subjects immediately used the stick to solve the task. Thick, heavy sticks were used in the same fashion as stones (dropped into the tube). But, when provided with a thin, light stick, the Rooks maintained their hold on the stick and used it to push down to collapse the platform. The use of a tool by a novel mechanism (pushing vs. dropping) suggests comprehension of the problem and goal-directed action rather than a conditioned response (Bird, 2009).

Rooks also identified functional tool vs. nonfunctional, demonstrated metatool use (use of one tool to acquire a second tool, which is then used to collect food), as well as tool modification. But, could the Rook spontaneously fabricate tools as did Betty? Similar to Betty, the Rooks had been tested with a metal hook tool but never trained to bend it. When presented with a straight wire all four subjects manufactured a hook and used it successfully. Three of the four Rooks did so on their first trial (Bird, 2009).

Tool use, once thought to distinguish Homo sapiens from other animals, is no longer a definitive assay of intelligence. However, specific instances of tool use such as metatool tests, tool modification and novel tool fabrication show forethought, planning, and imagination that satisfy Emery’s definition of intelligence.

Social intelligence hypothesis

Considering that most Corvidae live in complex social groups, their intelligence ought to be considered through the lens of the social intelligence hypothesis. The social intelligence hypothesis claims that enlarged brains and intelligence evolved in response to challenges associated with social complexity (Holekamp, 2007). While some examples of Corvidae intelligence, such as tool use, do not seem related to social behaviors, many are.

Recognizing faces

Because of their successful adaptation to North American cities, the American Crow (Corvus brachyrhynchos) is exposed to a wide range of human-associated dangers. Crows have adapted behaviors – or perhaps developed de novo behaviors – in response to new environmental hazards. Further, these behaviors appear to be socially learned (Cornel, 2011). Indeed, the ability to learn and teach within a social or cultural system demonstrates a type of higher order cognition seen in Corvidae.

The benefits of identification and recognition of a conspecific individual are obvious: avoiding conflicts, identifying mates and kin, and maintaining social structures. While less obvious, engaging in heterospecific identification appears to confer benefits to species with close inter-species relationships, which explains the ability of domestic animals to do so (Marzluff, 2010). Crows, too, are able to recognize and remember individual human faces, apparently remembering them for long periods of time (Marzluff, 2010).

While human facial recognition is not unique to crows, it does provide an opportunity for researchers to test individual vs. social learning. Certainly by undergoing an aversive experience an individual crow will learn to avoid the conditioned stimulus. However, social learning – learning from other crows to avoid the aversive stimulus – is a distinct ability.

Crows demonstrate two types of social learning: horizontal (learning from crow peers) and vertical (learning from parents) (Cornell, 2011). Crows subjected to an aversive experience (capture, banding, release) by a researcher wearing a mask designated as “dangerous” learned to associate the mask with the experience. Learning was demonstrated by aggressive behaviors (e.g. mobbing, scolding, agitated wing/tail flicking) toward the mask.

With the aid of naïve, blinded volunteers, researchers tested crow learning by walking around the area of capture (2-4 km) wearing either the “dangerous” or a control mask and recording response behavior. Testing occurred two weeks, 1.25 and 2.7 years after trapping.

Over time scolding behavior involved more crows over a larger area. At two weeks 26% of crows encountered scolded the “dangerous” mask (consistently at one location within 3060 m2). By 1.25 years 30.4% of crows scolded at two locations (covering an area of 6000m2). By 2.7 years 66% of crows scolded consistently at seven locations (covering an area of 10922 m2) (Cornell, 2011). Prior to the experiment only 3% of the 167 crows scolded the mask.

These data show a positive relationship between the number of opportunities to observe scolding, but not trapping, and the number of scolding crows. This is consistent with social learning and transmission of the aversion to the “dangerous” mask throughout the population. Crows do not learn aversion solely via individual experience. Rather, they are capable of recognizing a “dangerous” face and learning, from parents and peers, to express hostility toward it.

State attribution

In addition to socially dependent predator-avoidance learning, Corvidae have also demonstrated complex cognition in other social contexts such as courtship. Indeed, the Eurasian Jay (Corvus Glandarius) engages in behaviors that suggest a level of cognition somewhat reminiscent of Theory of Mind.

One form of social reasoning that, in humans appears to develop by 18-months, is the ability to infer or predict the desire of another individual (Repacholi, 1997). This desire reasoning, also known as state-attribution, requires that the child recognize another individual’s desire-state and respond accordingly. It is believed to developmentally precede Theory of Mind. It is possible that state-attribution is an evolutionary precursor of this ability (Ostojić, 2012).

The study of state-attribution requires manipulation of desire states. This can be done using specific satiety, a research paradigm in which an organism becomes satiated by a specific food thereby decreasing preference (presumably, desire) for the same. When satiated by one food source, an organism is much less likely to choose the satiating food when presented with a novel food. An organism capable of state-attribution can predict the preference (satiating vs. novel food) of another satiated individual.

Because the Eurasian Jay engages in food-sharing courtship behavior, manipulating the desire-state of male and female Jays allows for the study of state-attribution in the Corvidae family. If male Jays are capable of state-attribution, he must recognize that the female has a distinct desire-state, subjugate his desire-state for hers, and modify sharing behavior (Ostojić, 2013).

In one study male-female pairs of Jays were separated into two cages with a wire window separating them. During pre-feeding the window was covered by a transparent screen (seen condition) or an opaque screen (unseen condition). Both screens prevented food-sharing. The Jays were pre-fed either maintenance diet (MD), moth larvae (W), or mealworm larvae (M). During testing, the screens were removed, food W and M were presented to males, and food-sharing was observed.

In the unseen condition, the male Jay shared foods as would be predicted by his desire-state (induced by specific satiety). In other words, the male was more likely to consume and share the novel food over the satiating. Sharing preference was independent of the female pre-feeding source. This result suggests that the male’s desire-state influences food-sharing behavior (the food-source was not random) and that the female does not communicate her desire-state to the male.

Male Jays in the seen condition similarly consumed food according to their desire-state. However, in contrast to the unseen condition, they preferentially shared food according to the female’s desire-state. In doing so, the males alter food-sharing behavior, but not consumption, from an ego- to partner-centric preference.

This evidence suggests that the male Eurasian Jay is capable of state-attribution and modifies courtship behavior accordingly (Ostojić, 2013). This complex reasoning, absent in human infants, shows that the Jay can observe behavioral and environmental contexts, infer their influence on another’s mental state, and predict the behavior and preference of another individual.


In light of the Eurasian Jay’s ability to infer mental states, a precursor to Theory of Mind, it is worthwhile to briefly mention another cognitive ability associated with the development of Theory of Mind: self-recognition.

Though not without its controversy (Heyes, 1994; Heyes 1998 as cited by Prior, 2008), a well-known test of self-recognition is the mirror-test. Subjects can demonstrate emerging self-recognition if, when presented with a mirror the individual understands that the reflection does not represent another individual but oneself.

The Magpie (Pica pica) is the first non-mammalian species to exhibit self-recognition in the mirror-test (Prior, 2008). Initial interaction with the mirror elicited social behaviors (aggressive and submissive displays, jumping toward the mirror, and transient courtship-like behavior). However, within a few trials most of the social behavior vanished.

An increase of self-directed behaviors (e.g. attempting to reach the mark with the beak or touching the mark with the foot) accompanied the placement of a mark on the Magpie’s breast. Such behavior occurred in the presence of a mirror and ceased once the mark was removed (Prior, 2008).


Bird species of the family Corvidae, while evolutionarily distinct from primates, are capable of complex cognition and ape-like intelligence. Indeed, in some tasks, Corvidae out perform primates (Emery, 2006). It has been shown that species in this family can purposefully manipulate objects for tool fabrication and use (Weir, 2002 and Bird, 2009); share information about dangerous human faces (Cornell, 2011); infer the mental state of their partner (Ostojić, 2013); and pass the mirror-test (Prior, 2008).

Taken together, the literature on Corvidae cognition shows a well-developed intelligence in a non-primate brain. Like flight in birds, bats, and insects, convergent evolution of intelligence, shows that intelligence does not require mammalian cortical structures.

Obviously the Corvidae represent a family of extremely intelligent species whose cognition developed via an evolutionary trajectory distinct from our own. The implications of such evolution are exciting. For one, it shows that human (or primate) intelligence is not the only possible mechanism for the development of intelligence. Prior to studies in Corvidae and Cephalopods (octopus and cuttlefish), one could only confirm the emergence of intelligence in one small group: a handful of mammals. As someone who harbors hope of one day confirming that we are not alone in this universe, the evidence of convergent evolution of intelligence is, well, beyond exciting.


Beck B. B. (1980). Animal Tool Behaviour (Garland Press, New York).

Bird C. D., Emery, N. J. (2009). Insightful problem solving and creative tool modification by captive nontool-using rooks. PNAS, 106 (25) 10370-10375.

Cornell H. N., Marzluff J. M., Pecoraro S. (2011). Social learning spreads knowledge about dangerous humans among American crows. Proceedings of the Royal Society B: Biological Sciences, 279(1728):499-508.

Emery N. J. (2006). Cognitive ornithology: the evolution of avian intelligence. Phil. Trans. R. Soc. B, 361, 1465, 23-43.

Emery N. J., Clayton N. S. (2004). The Mentality of Crows: Convergent Evolution of Intelligence in Corvids and Apes. Science, 306 (5703), 1903-1907.

Heyes C. M. (1998). Theory of mind in nonhuman primates. Behav Brain Sci 21: 101–114.

Heyes C. M. (1994). Reflections on self-recognition in primates. Anim Behav, 47: 909–919.

Holekamp K. E. (2007). Questioning the social intelligence hypothesis. Trends in Cognitive Sciences, Volume 11, Issue 2, 65-69.

Marzluff J. M., Walls J., Cornell H. N., Withey J. C., Craig D. (2010). Lasting recognition of threatening people by wild American crows. Anim. Behav. 79, 699–707

Ostojić L, Shaw R. C., Cheke L. G., Clayton N. S. (2013). Evidence suggesting that desire-state attribution may govern food sharing in Eurasian Jays. PNAS, 110 (10) 4123-4128

Povinelli D. J., (2000). Folk physics for apes. Oxford: Oxford Univ. Press.

Prior H., Schwarz A., Güntürkün O. (2008). Mirror-Induced Behavior in the Magpie (Pica pica): Evidence of Self-Recognition. PLoS Biol 6(8): e202.

Repacholi B. M., Gopnik A. (1997). Early reasoning about desires: Evidence from 14- and 18-month-olds. Developmental Psychology, 33, 12–21.

Weir A. A. S., Chappell J., Kacelnik A. (2002). Shaping of Hooks in New Caledonian Crows. Science, 297 (5583), 981.

Animal Models in Research Are Necessary and Ethical

Read my opinion piece published in The Daily here.

Two of the most frequently repeated claims of the animal rights movement are that animal models are not actually useful in science and that there are more effective, humane ways to engage in research. While appealing, both statements are wrong.

If you are interested in learning more about the use of animals in research there are many reputable sources that discuss the scientific and ethical justification for such models. I list a few here:

In addition to general information collected from the above websites, I cite the following primary sources.