From Concrete to Abstract Rules: A Computational Sketch.

A multi-dimensional stimulus can elicit a range of responses depending on which dimension or combination of dimensions is considered. Such selection can be implicit, providing a fast and automatic selection, or explicit, providing a slower but contextualized selection. Both forms are important but do not derive from the same processes. Implicit selection results generally from a slow and progressive learning that leads to a simple response (concrete/first-order) while explicit selection derives from a deliberative process that allows to have more complex and structured response (abstract/second-order). The prefrontal cortex (PFC) is believed to provide the ability to contextualize concrete rules that leads to the acquisition of abstract rules even though the exact mechanisms are still largely unknown. The question we address in this paper is precisely about the acquisition, the representation and the selection of such abstract rules. Using two models from the literature (PBWM and HER), we explain that they both provide a partial but differentiated answer such that their unification offers a complete picture.