Julia Haas, Taking the Lead on Motivation, Predictive Processing and Reinforcement Learning
Taking the Lead on Motivation
Proponents of Predictive Processing (PP) describe it as a grand unifying theory of the mind (Hohwy 2014, Clark 2015). However, the relationship between PP and its closest rival, reinforcement learning (RL), is controversial. Unificationists about PP sometimes argue that active inference can account for core features of RL (Friston et al. 2009). Anti-unificationists reject this and defend explanatory pluralism as the most promising avenue for scientific progress (Colombo and Wright 2016). I argue for an intermediate position: even if RL is a special case of Bayes-optimal inference, it remains better suited to explaining motivation – and failures of motivation – than its more abstract counterpart (Dayan and Abbott 2001, Trappenberg 2002, Woodward 2014, Klein 2016).