What Are Association Formation Models Good For?
De Houwer, Jan, Learning & Behavior
Dickinson (2009) agrees with two of the core claims that were made in my target article (De Houwer, 2009): Associative learning effects (often) depend on (1) the formation of propositions and (2) the operation of nonautomatic processes. Whereas I derived these claims from propositional models of learning, Dickinson (2009) argues that they are also compatible with association formation models (AFMs). He also defends the position that only AFMs can explain "nonrational responses to associative experiences." In this comment, I question whether there is much value in continuing to cling to AFMs.
In the first part of his comment, Dickinson (2009) discusses earlier work (Dickinson, 1980) in which he put forward the hypothesis that many aspects of associative learning are driven by propositional knowledge. I fully subscribe to this hypothesis and acknowledge that Dickinson was one of the first to propose it in such an explicit manner. I do dispute, however, the claim that association formation models (AFMs) can account for the acquisition of propositional knowledge. A proposition can be seen as a qualified association between representations of events. It specifies not only that the events are related, but also how they are related. Whereas all AFMs incorporate assumptions about how organisms learn that events are related, I do not know any AFM that specifies how organisms acquire information about the way in which events are related. For instance, if there is a contingency between the occurrence of a disease and the presence of a substance in the blood, how would an AFM decide that the disease is an effect of the substance in the blood, or that the disease is the cause of the substance in the blood (see Waldmann & Holyoak, 1992)? Until it is clarified how AFMs learn about the way in which events are related, one should not take seriously the claim that they can account for the acquisition of propositions. Propositional models, on the other hand, postulate that organisms form and evaluate propositions on the basis of other propositions that they entertain. It is precisely this core assumption that has led to the identification of many new determinants of associative learning, such as causal knowledge, instructions, and deductive reasoning (see De Houwer, 2009, for a review). Little attention was given to these determinants until the arrival of propositional models, because AFMs have little to say about them.
In the second part of his comment, Dickinson (2009) correctly points out that certain important AFMs incorporate the assumption that association formation is nonautomatic, in the sense of being dependent on attention and cognitive resources. Proponents of propositional models go a step further still by postulating that the process underlying associative learning (i.e., the formation and evaluation of propositions) is also nonautomatic with regard to the need for awareness and goals (see De Houwer, 2009). No current AFM shares the latter assumption. It is true, however, that all evidence for the nonautomatic nature of associative learning can easily be accommodated by AFMs simply by adding assumptions about the way in which the association formation process is nonautomatic. Although adding these assumptions would make the models more compatible with the existing evidence, it would also decrease their appeal. One of the reasons many researchers believe in the existence of association formation processes is precisely that these processes might account for seemingly automatic, irrational types of learning. If association formation processes are as nonautomatic as propositional reasoning, what is the use of assuming their existence? Also, there is no a priori reason to assume that association formation is a nonautomatic process. There are, however, good reasons to assume that propositional reasoning is a largely nonautomatic process. From this perspective, one could even argue that AFMs that emphasize the role of attention and cognitive resources are actually mathematical formalizations of propositional reasoning. …