Causal reasoning

Vaesen correctly points out that “causal understanding involves more than just noticing (e.g., through trial and error) the covariance between a cause … and an effect” (sect. 4, para. 1). But then Vaesen goes on to claim, incorrectly in our view, that a cognizer must “infer a mechanism” in order to possess true causal understanding. To be sure, there are those who have advanced such a view (e.g., Ahn et al. Reference Ahn, Kalish, Medin and Gelman1995). However, the notion that prior knowledge of a mechanism is required for causal understanding offers no insight into how causal learning can get started: that is, how can a reasoner infer a causal mechanism from noncausal observations (Cheng Reference Cheng and Medin1993; Reference Cheng1997)? More recent theoretical work based on variants of causal Bayes nets has established that a cognizer can recognize a relation as specifically causal without necessarily understanding anything about unobservable causal mechanisms (for reviews see Gopnik & Schulz 2007; Holyoak & Cheng Reference Holyoak and Cheng2011).

Because Vaesen overlooks the distinction between causal reasoning and the representation of unobservable causal mechanisms, he misconstrues the results of Povinelli's rake experiments as evidence that chimpanzees learn through “associative learning” (sect. 4, para. 3). We have argued that the chimpanzees in these experiments were perfectly capable of first-order causal understanding (Penn & Povinelli Reference Penn and Povinelli2007a). It is the ability to reason about higher-order causal relationships that eludes them (Penn et al. Reference Penn, Holyoak and Povinelli2008; Povinelli Reference Povinelli, Reaux, Theall and Giambrone2000).

Function representations

Vaesen is probably correct that chimpanzees do not form “functional representations” (sect. 5) of tools in the same manner as humans. But it seems implausible to us that chimpanzees do not form functional representations at all. They certainly perceive stick-like objects as able to “function” in a certain manner for achieving certain goals, and these representations generalize over a fairly wide variety of shapes, colors, and textures. In our view, chimpanzees are perfectly able to form functional representations of stick-like objects in terms of surface features of the objects – they just fail to represent “functions” in terms of the underlying causal mechanisms involved (Povinelli Reference Povinelli, Reaux, Theall and Giambrone2000).

Explaining the discontinuity

In summarizing his findings from the first half of the paper, Vaesen (sect. 11) argues that “no individual cognitive trait” can be singled out as the key trait differentiating humans from other animals, and then claims that his argument is an antidote to “single-trait explanations of ‘humaniqueness’” (sect. 11, para. 3). This is our major point of contention with Vaesen.

To be sure, we know of no researcher who claims that there is one and only one trait that distinguishes human and nonhuman cognition. There are, indeed, a large number of cognitive traits that appear to be distinctively human – ranging from mental state attribution and language to causal reasoning and contingent cooperation. But Vaesen does not consider or rebut the possibility that there might be a deep dependency between many or even all of these disparate traits both at a cognitive/computational level of explanation and at an evolutionary/biological level of explanation.

It is possible, of course, that each of our uniquely human cognitive traits evolved independently of each other, and that each is embodied in a separate and independent “module” in the human brain. There are certainly researchers who defend such a “massively modular” explanation for human cognition (Carruthers Reference Carruthers and Stainton2005; Tetzlaff & Carruthers Reference Tetzlaff and Carruthers2008). But to our eyes, it seems wildly implausible that one and only one species was lucky enough to have evolved separate and independent mechanisms for each of these uniquely human traits (in a few million years to boot), whereas no other species evolved any of them. It seems much more likely (not to mention parsimonious) that there are deeper dependencies among these disparate traits such that a species that evolved a few linchpin traits would be in a more propitious state, from an evolutionary point of view, to acquire the others.

We have argued that the ability to represent and reason about the relation among relations – that is, “high-order relational reasoning” – is a plausible candidate for one of these linchpin traits (Penn et al. Reference Penn, Holyoak and Povinelli2008). It certainly seems noteworthy that many of the cognitive traits Vaesen identifies as instrumental in the evolution of human tool use – causal reasoning, functional representations, foresight, teaching, mental state attribution, contingent reciprocity, goal sharing – appear to depend upon a common set of higher-order relational competences.

Numerous researchers, for example, have demonstrated a strong empirical relationship between higher-order relational reasoning and theory-of-mind competence (e.g., Andrews et al. Reference Andrews, Halford, Bunch, Bowden and Jones2003; Zelazo et al. Reference Zelazo, Jacques, Burack and Frye2002). And almost all theoretical models of mental state attribution presume higher-order relational reasoning as an underlying mechanism (e.g., see the theories proposed in Carruthers & Smith Reference Carruthers and Smith1996). With respect to causal reasoning, most contemporary researchers agree that the ability to reason about a network of causal relations in a systematic and allocentric fashion is the bedrock of human causal cognition (e.g., Lagnado et al. Reference Lagnado, Waldmann, Hagmayer, Sloman, Gopnik and Schulz2005; Tenenbaum et al. Reference Tenenbaum, Griffiths and Kemp2006). Higher-order relations are also central to language (e.g., Gomez & Gerken Reference Gomez and Gerken2000; Hauser et al. Reference Hauser, Chomsky and Fitch2002; Pinker & Jackendoff Reference Pinker and Jackendoff2005).

The cognitive traits Vaesen subsumes under the heading of “executive control” are a motley set. There is good evidence that some of these – e.g., inhibition, autocuing, and self-monitoring – are necessary components of the ability to reason about higher-order relations (Andrews et al. Reference Andrews, Halford, Bunch, Bowden and Jones2003; Cho et al. Reference Cho, Moody, Fernandino, Mumford, Poldrack, Cannon, Knowlton and Holyoak2010; Halford et al. Reference Halford, Wilson and Phillips1998; Robin & Holyoak Reference Robin, Holyoak and Gazzaniga1995). Others – for example, foresight, hierarchical planning, and inferential coherence – are plausibly the result of being able to reason about higher-order relations.

Much work remains to be done to disentangle the necessary and sufficient components of higher-order relational reasoning in humans, and to understand how such a unique computational mechanism evolved in the brain of one particular species. However, there is already strong evidence, from a wide variety of domains and researchers, that this ability lies at the heart of “what makes us so smart” (Gentner Reference Gentner, Gentner and Goldin-Meadow2003). Our principle difference with Vaesen is that he neither considers nor rebuts this possibility.