The conclusion Brewer (Reference Brewer, Weimer and Palermo1974) drew from his review of conditioning research using human participants came as a great shock. For decades it had been very widely accepted – and not just by the many behaviorists of those times – that the processes described by S-R (stimulus-response) or reinforcement theorists based on animal evidence also provided a basis for at least simple aspects of human behavior in a manner that was independent of belief or awareness. Brewer's conclusion was that there was no convincing evidence to support this assumption. His alternative account of what goes in a conditioning experiment is captured by the following quote: “The college sophomore does not leave his higher mental processes outside the door when he walks into the experimental room, but he uses them to try to understand what is going on and what he should do about it” (p. 2). Despite many subsequent attempts to show his conclusion to be wrong, Brewer (Reference Brewer, Weimer and Palermo1974) clearly was correct about the overwhelming influence of “higher mental processes” in determining a participant's behavior in the kind of conditioning experiment – mainly “conditioning” of autonomic responses or of small movements – that he reviewed.
The article by Mitchell et al. can be seen as a successor to Brewer (Reference Brewer, Weimer and Palermo1974), in which a similar argument is directed mainly at experiments from the past two decades that have used causal or predictive scenarios in experiments to test principles of associative learning. One similarity between past and present research is the overwhelming use of college students as participants, a population that has been selected on the basis of thoughtfulness and then encouraged to be curious. An odd aspect of too many causal judgment experiments is that, although the researchers want their participants' higher mental processes to operate in order to understand the sometimes complex instructions, interpretation of the results assumes the absence of any such influence following a participant's first response. In this respect, many points made by Mitchell et al are salutary, including the important one that associations are not expectancies. When applied to animal data, the absence from associative learning theories of explanations as to how links provide the basis for expectancies is rarely problematic; but when applied to human experiments, this is a major lacuna.
Mitchell et al.'s arguments are less compelling when one looks beyond the kind of study on which they focus. Odors provide an example of stimuli that people find hard to describe or identify, and therefore difficult to think about. Stevenson and I have suggested that this is the principal reason that conditioned changes in an odor's perceptual properties are independent of the poor explicit memory participants have for the stimulus contingencies they were given during training (Stevenson & Boakes Reference Stevenson, Boakes, Calvert, Spence and Stein2004). For a similar reason, changes in human flavor preferences produced, for example, by caffeine-based conditioning, also appear independent of belief or awareness (e.g., Chambers et al. Reference Chambers, Mobini and Yeomans2007). Moving beyond the laboratory, the strong aversions developed by cancer patients undergoing chemotherapy appear to develop quite independently of patients' valid beliefs about the cause of their distress; and, in a way, that shows very strong parallels with conditioned taste aversions in rats and many other animals (e.g., Bernstein Reference Bernstein, Braveman and Bronstein1985).
Particularly interesting, and far less widely known, examples come from the study of placebo effects, an area where a similar debate has continued as to the contributions of conscious expectancies and conditioning without awareness (Stewart-Williams & Podd Reference Stewart-Williams and Podd2004). Here it has turned out that, whereas some placebo effects are strongly determined by patients' beliefs about their treatment and are sensitive to the information that is provided, others depend on past treatment history in a way that is independent of belief and insensitive to information. Thus, in one major study, verbal suggestions accompanying medication had a large effect on reactions to pain and on the motor responses of Parkinson's patients, but no detectable effect on hormonal and cortisol levels, whereas the latter could, however, be altered by a placebo treatment following a conditioning procedure (Benedetti et al. Reference Benedetti, Pollo, Lopiano, Lanotte, Vighetti and Rainero2003).
What such examples suggest is that the conclusion to draw from Mitchell et al. is that the kind of research they review is not likely to reveal much about human learning. Preventing student participants from “reflection” so that their responses in, say, a causal judgment experiment are “intuitive” (Shanks Reference Shanks2007) turns out to be very difficult. Mitchell et al. refer to three examples that appear to have achieved this, including our own study in which we placed participants in a causal judgment experiment under strong time pressure, so giving them “little time to think” (Karazinov & Boakes Reference Karazinov and Boakes2007). Mitchell et al. attempt to explain away the result; but in doing so, they appear to accept that the non-rational response given by the average participant under these conditions must be based on a within-compound association and not on any kind of logical inference. Nevertheless, even though a few studies of this kind appear to have been successful in reducing the influence of logical inferential thinking, this is not enough to justify a confident return to this way of studying human associative learning.
