Bayesian Fundamentalism or Enlightenment? On the explanatory status and theoretical contributions of Bayesian models of cognition

Abstract

The prominence of Bayesian modeling of cognition has increased recently largely because of mathematical advances in specifying and deriving predictions from complex probabilistic models. Much of this research aims to demonstrate that cognitive behavior can be explained from rational principles alone, without recourse to psychological or neurological processes and representations. We note commonalities between this rational approach and other movements in psychology - namely, Behaviorism and evolutionary psychology - that set aside mechanistic explanations or make use of optimality assumptions. Through these comparisons, we identify a number of challenges that limit the rational program's potential contribution to psychological theory. Specifically, rational Bayesian models are significantly unconstrained, both because they are uninformed by a wide range of process-level data and because their assumptions about the environment are generally not grounded in empirical measurement. The psychological implications of most Bayesian models are also unclear. Bayesian inference itself is conceptually trivial, but strong assumptions are often embedded in the hypothesis sets and the approximation algorithms used to derive model predictions, without a clear delineation between psychological commitments and implementational details. Comparing multiple Bayesian models of the same task is rare, as is the realization that many Bayesian models recapitulate existing (mechanistic level) theories. Despite the expressive power of current Bayesian models, we argue they must be developed in conjunction with mechanistic considerations to offer substantive explanations of cognition. We lay out several means for such an integration, which take into account the representations on which Bayesian inference operates, as well as the algorithms and heuristics that carry it out. We argue this unification will better facilitate lasting contributions to psychological theory, avoiding the pitfalls that have plagued previous theoretical movements.

Why some surprises are more surprising than others: Surprise as a metacognitive sense of explanatory difficulty

“…However, these proposals assume (i) that probabilities can be meaningfully assigned to information items in the scenario, (ii) to any inferences made about the scenario, and (iii) that these can then be meaningfully related to expected probabilities for the outcome. These questions may be the most significant stumbling block for the development of any probabilistic account (cf., more general observations made by Jones & Love, 2011). Here we consider briefly what these questions mean for the two best Bayesian accounts in the literature.…”

“…Similarly, because probability accounts tend not to be performance theories (Jones & Love, 2011), they are silent on the findings about the specific time-course of processing surprise judgments or, indeed, the productivity of explanations. So, it is hard to escape the conclusion that the best probability accounts envisaged for these results provide limited coverage of the results found.…”

“…Theoretically, at the computational level, a theory of surprise needs to specify the "goal of the computation" and "what needs to be computed?" (see Jones & Love, 2011;Keane, Ledgeway & Duff, 1994). Following on from the above proposals, we see the cognitive goal of surprise as being "to explain the discrepancy that arises when a surprising outcome occurs".…”

“…Very surprising events are those that are difficult to explain, while less surprising events are those which are easier to explain. Importantly, this high-level, computational account of surprise is supplemented by algorithmic proposals that generate the key empirical predictions tested in this paper (see Jones & Love, 2011;Keane et al, 1994).…”

“…10 Perhaps, the real fundamental theoretical problem for probabilistic accounts is the need for their computational-level proposals to be supplemented by additional algorithmiclevel propositions. Following Jones and Love's (2011) suggestions, such a Bayesian account would have to be supplemented with a mechanism, an algorithmic-level description that accounts for the impacts of various empirical factors on the components of the Bayesian model. We believe that such a complete account would end up having to subsume many of the precepts of the current theory (e.g., about how the search in memory is directed, how relevant concepts are retrieved, and how these are built into explanations for outcomes).…”