Emergence of Symbolic Inference Based on Value-Driven Intuitive Inference Via Associative Memory

Humans use two types of inferences: intuitive and logical. However, they are studied separately. A site for logical inference has not been found in the brain, but modeling it as a distributed neural network form is desirable. In this study, we propose an inference model of an intuitive search process in continuous and distributed associative memory (AM), and it switches to a symbolic mode, in which each step of association converges to a stable state of self-recollection, realizing step-by-step logic. Switching is evoked by biasing the associative gain upon finding a valued state during the intuitive inference. In this study, we show the computational model of symbolic inference via AM, and we verify its practicality by solving a maze task. We show the emergence of a tree search-like behavior with pruning.