GNOSTRON: a framework for human-like machine understanding
Yan M. Yufik · 2018
Reports in the recent literature abundantly indicate that for many people, in the general public and scientific community alike, advances in machine intelligence present a disturbing prospect. Realizing that machines are devoid of understanding is likely to exacerbate the discomfort: the current generation of AI systems can accumulate knowledge but remains clueless about its meaning and incapable of explaining how it is used in making decisions. Machines can recognize familiar conditions and select appropriate responses but are incapable of responding adequately to novel situations. Humans encountering unfamiliar situations seek to understand them and optimize responses based on such understanding. All species can learn but understanding-based learning is a unique and definitive feature of human intelligence. The gnostron proposal strives to simulate brain mechanisms underlying human understanding. This paper summarizes some of the key ideas, aiming at facilitating further inquiries and accelerating progression from knowledge-based to understanding-based AI.