AGIs as collectives

Abstract

The essay examines the implications of developing artificial general intelligence (AGI) as a collective of multiple agents, rather than as a single entity. It compares and contrasts this “collective AGI” approach with other models of multi-agent systems, such as a single AGI with multiple modules or a system of interconnected AIs. It argues that collective AGIs may possess advantages in interpretability and flexibility compared to single AGIs, but they also present challenges in fine-tuning and managing agency. The essay highlights the need for further research into the safety and efficacy of both single and collective AGI approaches, particularly regarding the potential trade-offs between individual agent intelligence and the ability of the collective to cooperate at larger scales. – AI-generated abstract.