Modeling attention in associative learning: two processes or one?

Certain studies of associative learning show that attention is more substantial to cues that have a history of being predictive of an outcome than to cues that are irrelevant. At the same time, other studies show that attention is more substantial to cues whose outcomes are uncertain than to cues whose outcomes are predictable. This has led to the suggestion of there being two kinds of attention in associative learning: one based upon a mechanism that allocates attention to a cue on the basis of its predictiveness, the other based upon a mechanism that allocates attention to a cue on the basis of its prediction error (e.g., Le Pelley, Quarterly Journal of Experimental Psychology, 57B, 193-243, 2004). As an alternative, it has been demonstrated that the effects of both predictiveness and uncertainty can be accounted for with only one kind of attention: one that emphasizes the role of prediction (Esber & Haselgrove, Proceedings of the Royal Society B, 278, 2553-2561, 2011). Here, we consider the alternative: whether the effects of predictiveness and uncertainty can be reconciled with a model of learning that emphasizes the role of prediction error (Pearce, Kaye, & Hall, 1982). Simulations of this model reveal that, in many cases, it too is able to account for the influence of predictiveness and uncertainty in associative learning.