Penn et al's reassessment of claims for the human-like nature of nonhuman cognition is both timely and welcome, and I agree wholly with its sentiments. On the one hand, a rejection of the anthropocentric (and often anthropomorphic) emphasis of current comparative research strategies serves to increase our recognition of, and receptiveness to, the potential diversity of psychological mechanisms that exist across the animal kingdom, and it draws us away from the evolutionarily impoverished view that other species' cognition will merely represent some or other variant of our own. On the other hand, and despite their rejection of “classical” physical symbol system (PSS) models, Penn et al. continue to rely heavily on a computational model of cognition that places all the interesting work to be done solely inside the organism's head. This diminishes Penn et al.'s otherwise laudable attempt to get things back on a more appropriate evolutionary footing while, at the same time, it leaves no room for the truly novel aspects of human cognition that seem likely to account for the differences that exist between us and other species. Specifically, Penn et al.'s rejection of “non-representational, purely embodied” processes as having anything much to tell us about nonhuman cognition, and their conclusion that the comparative data “strongly suggests that nonhuman minds are … highly structured, information-processing devices in a way that stomachs and Watt governors are not” (sect. 11, para. 3) seems premature, not least because the comparative literature reviewed in their article is framed and interpreted within a computational metaphor that regards internal representational structure and information processing as axiomatic; such data must inevitably support this conclusion, but do not rule out other possible mechanisms. By being in thrall to a representational theory of mind based on the computer metaphor, Penn et al. are obliged to draw a representational line in the sand that animals are unable to cross in order to account satisfactorily for the differences between ourselves and other animals. The suggestion here is that, if Penn et al. step back from this computational model and survey the problem more broadly, they may recognize the appeal of an embodied, embedded approach, where the ability of humans to outstrip other species may be a consequence of how we exploit the elaborate structures we construct in the world, rather than the exploitation of more elaborate structures inside our heads.

First and foremost, we need to recognize that all cognition is, by definition, “purely embodied,” for how can it be otherwise? Indeed, even the use of the term “embodied cognition” is rather misleading, for it suggests that there is an alternative – a disembodied cognition – that does not, in fact, exist (with the exception of a computer interface, perhaps). The fact of the matter is that all animals possess bodies, and all animals did so before they possessed anything remotely recognizable as a brain: this is the substance of Brooks's (1991; 1999) criticism of classical approaches. As he correctly points out, most of evolutionary history has been spent perfecting the perception and action mechanisms that enable survival in a dynamic world. It is these mechanisms, rather than “high-level” forms of cognition, like planning, logical inference, and formal reasoning, that are most informative to an evolutionary cognitive science because they constrain the forms that these high-level processes can take. Linked to this is the idea that, unless we take perception-action mechanisms seriously, exploring both their scope and limits, the “cognitive processes” that we see may be illusory, reflecting only our own frame of reference and not that of the animal itself (Brooks Reference Brooks1991; Reference Brooks1999; Pfeifer & Bongard Reference Pfeifer and Bongard2007). The realization that an organism's understanding of the world will be shaped by, and grounded in, the means by which it perceives and acts in the world, is at least as old as von Uexküll's (1934/1957) expression of the Umwelt, and it seems both necessary and vital to the comparative project. Equally, van Gelder's (1995) analogy of the Watt governor is as much about broadening the definition of a “cognitive system” to include the body and environment, as well as the brain – and the dynamic coupling that exists between these elements – as it is about contesting the notion of cognition as symbolic computation. Of course, cognitive systems are not literally like Watt governors, but neither are they literally like computers. Unlike these other analogies, however, the brain-as-computer has been taken both very literally and very seriously, and it underpins the particular view of cognition that current comparative studies, and Penn et al., endorse, where an animal's brain is placed at a remove from its body and the world in which it lives. This, in turn, implies that brain processing is completely insulated from the world, raising all the difficulties of the “symbol grounding” problem.

Giving the body and the environment their due as integral parts of biological cognitive systems has a further corollary, in that we should not expect evolved organisms to store or process information in costly ways, when they can use the structure of the environment, and their ability to act in it, to bear some of that cognitive load (Clark Reference Clark1989; Reference Clark1997). The Klipsch horns built by mole crickets to give their mating calls a boost and the watery vortices that blue fin tuna create and exploit to increase muscular propulsion while swimming are both superb examples of how organisms exploit the structure of the environment in adaptive fashion (Clark Reference Clark1997). Why should cognitive systems be any different? If we think of cognitive systems as distributed across brain, body, and world (Clark Reference Clark1997), it gives rise to a theory of cognition that is both fully grounded, as an evolutionary account requires, and able to account for the functional differences between ourselves and other animals in terms of the degree to which our minds are extended beyond the strictures of “skin and skull.” As Clark (Reference Clark1989) has also suggested, this insight raises the prospect that many classic, symbolic van Neumann architectures may have mistakenly modeled “in-the-head” computation “on computation that, in humans, consists of both an in-the-head component and (to begin with) an in-the-world component” (p. 135). Our current ability, for example, to solve logical syllogisms by constructing Venn diagrams in our heads may only be possible because, initially, we were able to construct or observe such diagrams in a concrete, external physical form (Clark Reference Clark1989, p. 133). From a purely internal perspective, then, the cognitive processes of humans and other animals may well be quite similar. The difference, paradoxically, may lie in our ability to create and exploit external structures in ways that allow us to augment, enhance, and support these rather mundane internal processes.