Foundations of human problem solving

Abstract

Despite great strides in both machine learning and neuroscience, we do not know how the human brain solves problems in the general sense. We approach this question by drawing on the framework of engineering control theory. We demonstrate a computational neural model with only localist learning laws that is able to find solutions to arbitrary problems. Using a combination of computational neural modeling, human fMRI, and representational similarity analysis, we show here that the roles of a number of brain regions can be reinterpreted as interacting mechanisms of a control theoretic system. The results suggest a new set of functional perspectives on the orbitofrontal cortex, hippocampus, basal ganglia, anterior temporal lobe, lateral prefrontal cortex, and visual cortex, as well as a new path toward artificial general intelligence.

The whole prefrontal cortex is premotor cortex

“…We are proposing that the canonical function of OFC is optimization of goal-selection policies. This proposal is consistent with several recent models that portray OFC as performing goal-selection and representing the value of an agent's current state with respect to its distance to achieving a goal-fulfilling state [63,71,86,87]. Further support for a goal-selection characterization of OFC is already found in extant OFC lesion studies and their behavioural consequences.…”

Combining Imagination and Heuristics to Learn Strategies that Generalize

“…This ability appears to be meditated, at least in part, by the hierarchical organization of prefrontal cortico-basal ganglia-thalamic networks [15,16,17] that are organized in such a way that anterior regions, reflecting more abstract task representations, have a stronger influence on more posterior regions, reflecting more concrete task goals, than vice versa [18]. The prefrontal networks are also thought to have hierarchically organized error signals that facilitate learning complex and conditional rules [19,20], and more rostral areas (higher in the hierarchy) appear to be engaged during both rumination [21,22] and imaginative play [23,24,25], suggesting a common underlying computation [26]. Thanks to the phasic dopaminergic signals from midbrain regions [27,28,29] these pathways are also capable of learning action values, where rewarded (or punished) actions become more (or less) likely to be executed in the future.…”