Artificial General Intelligence: A New Perspective, with Application to Scientific Discovery

Abstract

The dream of building machines that have human-level intelligence has inspired scientists for decades. Remarkable advances have been made recently; however, we are still far from achieving this goal. In this paper, I propose an alternative perspective on how these machines might be built focusing on the scientific discovery process which represents one of our highest abilities that requires a high level of reasoning and remarkable problem-solving ability. By trying to replicate the procedures followed by many scientists, the basic idea of the proposed approach is to use a set of principles to solve problems and discover new knowledge. These principles are extracted from different historical examples of scientific discoveries. Building machines that fully incorporate these principles in an automated way might open the doors for many advancements.

Deep Learning Applied to Scientific Discovery: A Hot Interface with Philosophy of Science

“…Should future neural nets (in the niche considered) include not so much a 'theory hub' than a 'principle hub'? Interestingly, this possibility has actually been proposed by computer scientists (Khalili 2020).…”

“…From here, the next program might involve the formal sciences and aim at formalizing and coding this gradual process of abstraction. For results in computer science that could indicate first steps in this direction, see e.g., besides the mentioned articles, Battaglia et al (2018) and Khalili (2020).…”