Abstract: The “problem of skilled action” refers to the task of explaining how it is that skilled action displays robust intelligence despite being largely governed by automatic control processes that are often characterized as brute, reflex-like, and paradigmatically unintelligent. Recently, some have argued that the automatic control involved in skill is sensitive in key respects to the semantic content of an agent’s intentions and other propositional attitude states, and is thereby genuinely intelligent, or “proto-intelligent”, contra what others have assumed (e.g., Fridland 2014, 2016).

In this talk, I argue that, though automatic control is clearly sensitive to the semantic content of cognitive states, it is nonetheless informationally encapsulated to a significant degree, so that this sensitivity is not the best place to look for its intelligence. I offer an alternative account of the intelligence of automatic control that is not wholly derived from its sensitivity to propositional states, and draw out implications of this view for our understanding of skilled action, as well as other nearby challenges in action theory.