Moderate Predictive Processing

Abstract

Recent developments in the predictive processing literature have led to the emergence of two opposing positions regarding the representational commitments of the framework (Hohwy 2013; Clark 2015; Gładziejewski 2016; Orlandi 2015). Proponents of the conservative approach to predictive processing claim that the explanatory power of the framework comes from postulating a rich nesting of genuine representational structures which can account for many cognitive functions (Gładziejewski 2016). Supporters of the more radical interpretation of predictive processing, on the other hand, postulate that not all elements of the computational architecture should be interpreted as full-blown representations, stressing the framework’s connection to ecological and embodied approaches to cognition instead (Clark 2015, Orlandi 2015). Surprisingly, despite defending opposing positions, both camps seem to adopt William Ramsey's representational ‘job description challenge’ (Ramsey 2007) as a standard for genuine ascriptions of representational function. The aim of this paper is to evaluate competing approaches and show that both sides of the debate must overcome additional challenges with regard to determining the representational commitments of the predictive processing framework. Following the discussion of the opposing views and the way they employ the representational job description in their arguments, I raise a worry that Ramsey’s criterion may be ill suited for establishing a strong distinction between the opposing positions. In light of this I propose to frame the debate between the two camps as a disagreement over the contents of generative models, rather than the functional roles fulfilled by the structures under investigation. Finally, presented with the problem of content determination and the lack of framework constrains strong enough to help resolve the disagreement, I call for moderation in making claims about predictive processing’s representational status.