Coherent aggregated volition: A method for deriving goal system content for advanced, beneficial AGIs

Abstract

This paper discusses the concept of Coherent Aggregated Volition (CAV) as an approach to defining the goal system content for advanced, beneficial artificial general intelligences (AGIs). CAV attempts to capture the essence of Eliezer Yudkowsky’s Coherent Extrapolated Volition (CEV) proposal, which aims to derive goal system content from the collective desires, values, and beliefs of a population. Unlike CEV, CAV emphasizes logical consistency, compactness of representation, and the inclusion of actual, rather than extrapolated, volition. The paper proposes a multi-extremal optimization algorithm for finding a CAV solution and suggests that CAV could be prototyped in a simplified form using existing AI tools. It also highlights the importance of investigating the stability of CAV solutions under self-modification and environmental variations as a crucial next step. – AI-generated abstract.