A computational model of attentional learning in a cognitive agent

Abstract

Abstract Biologically inspired cognitive architectures should faithfully model the high-level modules and processes of cognitive neuroscience. Also, they are expected to contribute to the BICA “challenge of creating a real-life computational equivalent of the human mind”. One important component of the mind is attention and attentional learning. In this paper, we describe conceptual and computational model of attention and attentional learning for intelligent software agents in the context of the broad-based biologically inspired cognitive architecture, LIDA. In LIDA attention is defined as the process of bringing content to consciousness. Implementing Global Workspace Theory, the mechanism of consciousness consists of a continuing sequence of broadcasts of the most salient current contents to all of cognition. We argue that the term attention describes the selection of conscious contents and should be distinguished from mechanism of consciousness itself. Attentional learning, the learning of to what to attend, has been relatively little studied by memory researchers. Here we describe a mechanism for attentional learning using the LIDA architecture. A basic implementation of such an attentional learning mechanism in a LIDA-based agent is presented. The agent performs a psychological attention experiment and produces results comparable to human subjects. The agent’s contribution in determining internal parameters for the LIDA architecture is also described. Our model of attentional learning distinguishes different aspects of selectionist and instructionalist learning. Attentional learning has not received its deserved attention in cognitive architecture research. This work represents a first step toward implementing the full range of cognitive faculties associated with attention and attentional learning in the LIDA cognitive architecture.