Has the time arrived to rebuild the foundations of cognitive science? If Mitchell et al. had been the first to challenge the basic notion of associationism, then perhaps the field could survive this seeming coup d'état by simply pointing to the mounting evidence in favor of associationism – the classic research on the changes at the synaptic level that are responsible for classical conditioning (e.g., Kandel Reference Kandel, Kandel, Schwartz and Jessell2000; LeDoux Reference LeDoux2000; Thompson Reference Thompson2005); and investigations on the implicit/automatic processes mediating the learning of sequences, motor skills, attitudes (Stadler & Frensch Reference Stadler and Frensch1998), and the execution of various kinds of “input-output” processes (Hallett Reference Hallett2007; Pessiglione et al. Reference Pessiglione, Petrovic, Daunizeau, Palminteri, Dolan and Frith2008). Afterwards, we could all continue to promulgate that the brain learns as it does because “cells that fire together wire together” – a concept that, like the reflex, exemplifies our switchboard intuition regarding how intelligent behavior is implemented neurally. But Mitchell et al.'s is not the only challenge, and it may not be the most ground shaking.
Gallistel and Gibbon (Reference Gallistel and Gibbon2001) challenged associationism by demonstrating that it could not account for basic conditioning phenomena (in rats and pigeons) such as time-scale invariance or the observation that “neither the delay of reinforcement nor the ratio of reinforced to unreinforced presentation of the conditioned stimulus affects rates of acquisition and extinction” (p. 146). If cells that fire together do wire together, then intermittent conditioning should show less resistance to extinction than fixed schedules of conditioning; but this is not the case in classical and operant conditioning (Gallistel & Gibbon Reference Gallistel and Gibbon2001; Skinner Reference Skinner1953). Hence, the target article should be seen as a red flag that can no longer be ignored – a wake-up call to begin to carefully take stock of what is actually known regarding the basis of human cognition. Moreover, if the time has come to rebuild the foundations of cognitive science, perhaps it is also worthwhile to clean up our terms and reexamine the value of the notion of “mental representation,” an ambiguous term that has come to mean all kinds of things to all kinds of researchers.
Unlike consciousness, which has been regarded as the main “unsolved anomaly within the domain of the [scientific] approach” (Shallice Reference Shallice1972, p. 383), the neural mechanisms proposed to underlie conditioning reflect our intuitive understanding of how the mind/brain should work: Fear conditioning is mediated by the amygdala (LeDoux Reference LeDoux2000) in a manner that is consonant with our switchboard intuition regarding nervous function. However, despite our intuitions, once Tolman (Reference Tolman1948) demonstrated that there is reason to doubt that learning is due to simple stimulus-response (SR) models (e.g., by showing that rats could solve mazes without relying on any external or prioprioceptive cues; see also Lashley Reference Lashley and Jeffress1951; Terrace Reference Terrace2005), doubt should have fallen over all SR explanations of behavior. But it did not. Complex behaviors continued to be explained by complex mechanisms (e.g., cognitive maps, reasoning), and simple behaviors continued to be explained by simple mechanisms (e.g., SR strength).
When explaining how a pigeon pecks a button for food, one appeals to the principles of operant conditioning, but when explaining how wasps and pigeons are able to find their home in the absence of external cues, one invokes the term “cognitive map” (Gallistel Reference Gallistel1990), which depends on neural machinery that, for some reason, is believed to not be at play during button-pecking. It is seldom appreciated that, if all an animal possesses happens to be a sophisticated faculty of navigation, then this faculty will be used even for button pressing. The hands of evolution have been seen as behaving economically, using simple mechanisms for simple behaviors but reserving complex mechanisms for complex behaviors, which is a wrong way to think about evolution (cf. de Waal Reference de Waal2002). The fallacy has persisted even though, as noted by Mitchell et al., there has always been more evidence for the existence of high-level representational mechanisms than for simple mechanisms (e.g., SR strength).
It remains an empirical question whether, in humans, there are no instances in which associative learning is instantiated by the kinds of synaptic-level changes that have been identified in animals (see Phelps & LeDoux Reference Phelps and LeDoux2005). If so, what is the neural foundation of human cognition? Perhaps the time has come to examine whether conditioning is mediated by higher-level processes such as “interregional synchrony” or “neural coherence” (cf., Buzsáki Reference Buzsáki2006; Fries Reference Fries2005), processes believed to serve an important role in communication and in “binding” representations (Hummel & Gerloff Reference Hummel and Gerloff2005).
Are mental representations those tokens used by the propositional system proposed by Mitchell et al.? If so, Gallistel (Reference Gallistel, Smelser and Baltes2001) defined a mental representation as “a system of symbols isomorphic to some aspect of the environment, used to make behavior-generating decisions that anticipate events and relations in that environment … [and, cognitive] psychology is the study of mental representations” (p. 9691). However, the nature of the isomorphism to the world remains unclear with respect to many “representational” processes, such as non-intentional states (e.g., moods or the experience of holding one's breath; Gray Reference Gray2004). More concretely, in light of Gallistel's definition, it remains unclear to what the pungent flavor of hydrogen peroxide is isomorphic. This chemical differs molecularly from water only by the addition of a single oxygen atom, but few would perceive it as “water with a little too much oxygen.” Instead, the toxic chemical is perceived (or “represented”) as something that should be violently expelled from the body. This may lead one to hypothesize that, as with subjective urges (Morsella Reference Morsella2005) and percepts (Sperry Reference Sperry and Isaacson1964), the representation of H2O2 is isomorphic with respect to how one should respond to the stimulus, but this is not in line with the traditional view (based on the cognitive map) of what a mental representation is (Hommel et al. Reference Hommel, Müsseler, Aschersleben and Prinz2001). Hence, a more precise term is needed for the tokens that furnish the contents of the propositional reasoning system proposed by Mitchell et al. Cognitive science may be far from developing its periodic table, but it can still be rigorous about delineating what is known and what is not yet known.
