Relevance Realization as a Solution to the Frame Problem: A Comparison of Four Cognitive Architectures

Abstract

One of the largest remaining challenges in the field of artificial intelligence is the creation of a so-called artificial general intelligence (AGI): a system that is competent across a large number of domains of action and that is able to efficiently learn new skills in novel domains. Here, I argue that the main reason we currently have not been able to build an AGI is because of the unsolved frame problem from the philosophy of artificial intelligence. I present the framework of relevance realization, an attempt to a solution to the frame problem (Vervaeke, 2012), discuss some of the features of the framework and suggest two additional ones. I argue that a complete theory of relevance realization would indeed form a solution to the frame problem and thereby also to the problem of general intelligence. I compare four existing cognitive architectures for artificial general intelligence (CLARION, LIDA, AKIRA and IKON FLUX), and evaluate each architecture on the basis of the features of the framework of relevance realization as suggested by Vervaeke and myself. I conclude that AKIRA and IKON FLUX score highest on this comparison and end with a suggestion of a rough sketch for a new cognitive architecture that incorporates the best of each of the four earlier discussed architectures. I then argue that such a design, since it best incorporates the framework of relevance realization, thereby best approximating a solution to the frame problem, will be most promising as a design of a cognitive architecture for artificial general intelligence.