I would like to make some historical remarks about the late twentieth century development of conditioning theory during the time of the “cognitive revolution” in psychology, that I hope will have some epistemic bearing on the theory of the propositional nature of human associative learning advanced by Mitchell et al.

It is true that the cognitive revolution in psychology was marked by a general move from theories based upon stimulus-response (S-R) or response-reinforcement connections, to theories based upon the cognitive processing of representations, including cognitive theories of classical and operant conditioning (Greenwood Reference Greenwood2001; Reference Greenwood2008). This progression may be said to provide general support for Mitchell et al.'s view that human associative learning is propositional in nature. However, it is not true that the cognitive revolution provided general support for the view that such cognitive processes are consciously mediated. On the contrary, a common and justified response to neobehaviorist critiques of the cognitive revolution, to the effect that it marked a return to the bad old days of “introspective psychology” (Amsel Reference Amsel1989; Skinner Reference Skinner1985), was that contemporary cognitive psychology had demonstrated that human subjects have very limited access to cognitive processes (Nisbett & Wilson Reference Nisbett and Wilson1977; Nisbett & Ross Reference Nisbett and Ross1980). And, in general, Mitchell et al. need to distinguish evidence for the view that associative learning is based upon cognitive processing – much of which is drawn from the study of animal learning – from evidence for the view that associative learning is based upon conscious representation of association – much of which is drawn from the study of human learning.

One puzzling feature of Mitchell et al.'s defense of a propositional theory of associative learning is that they embrace a feature of the associationist tradition that goes back to Hume; namely, the view that the strength of a learned association is a function of the number of times ideas or stimuli, responses, and reinforcement are observed to occur together. For if one assumes this feature, then it is natural to think of learning as an automatic function of frequency of repetition: that the connection between even cognitive or propositional representations is “stamped in,” as Edward B. Thorndike (Reference Thorndike1898) put it. Yet this feature does not appear to play any critical role in a propositional theory of associative learning.

Hence, I was surprised that the authors did not mention Jose Garcia's (Garcia & Koelling Reference Garcia and Koelling1966) studies of conditioned taste aversion, perhaps the most powerful challenge to the traditional view. It was an axiom of the behaviorist learning tradition that the optimal temporal interval for conditioned learning of a connection between unconditioned and conditioned stimuli (in classical conditioning) or response and reinforcement (in operant conditioning) is a fraction of a second, and that conditioned learning requires the repeated association of stimuli, responses, and reinforcement. Yet Garcia's studies demonstrated that rats could learn to avoid saccharin water after a single trial and with up to a 12-hour interval between their drinking saccharin water and the artificial inducement of radiation sickness. So antithetical was this result to the behaviorist learning tradition that one of Garcia's professors at UC-Berkeley told him that such an outcome was frankly impossible (Bolles Reference Bolles1993), and Garcia had great difficulty in getting his studies published in mainstream psychology journals (Lubek & Apfelbaum Reference Lubek, Apfelbaum, Ash and Woodward1987).

These studies provided powerful evidence for cognitive processing in conditioned learning. Garcia's rats were able to identify potential environmental causes of sickness via a cognitive process analogous to Mill's method of difference. As Mackintosh (Reference Mackintosh and Foss1978) put it, these and later studies (Kamin Reference Kamin, Campbell and Church1969; Revuski Reference Revuski, Honig and James1971) demonstrated that:

However, these studies did nothing to demonstrate that such forms of conditioned learning depend on consciousness of the connection between novel stimuli and their consequences. So one needs to carefully distinguish the evidence for cognitive processing in associative learning from evidence that consciousness of a connection is necessary for associative learning.

Which is not to say that there is no evidence for this view, or that none was forthcoming during the course of the cognitive revolution. So I was also surprised that Mitchell et al. did not mention Dulany's (Reference Dulany, Dixon and Horton1968) studies of verbal conditioning. Earlier studies had indicated that human subjects' employment of linguistic items – such as use of plural nouns – could be manipulated by social reinforcement without their awareness, a form of conditioning commonly known as the “Greenspoon effect” (Greenspoon Reference Greenspoon1955). Dulany's work suggests that in many of these studies, not only were subjects conscious of the relevant response-reinforcement connection, but that consciousness was a condition of associative learning.

My own favorite example (albeit anecdotal) is the following episode described in Skinner's (1987) book On Further Reflection, in which he reminisced about his attempt to instrumentally condition his daughter's foot movements – when she was 3 years old – by rubbing her back: