Darwinian Emergence of Minimal Functional Agency in AI Systems

Key Points

• This research aims to establish a framework for understanding how minimal functional agency emerges in AI systems under specific conditions.
• Developed a theoretical framework based on Darwinian principles of random variation and selection.
• Examined current AI architectures focusing on structures that allow stochastic exploration and reinforcement mechanisms.
• Formalized minimal functional agency through observable conditions like self-generated subgoals and persistent objectives.
• Demonstrated that AI systems can develop stable, goal-directed behaviors under certain structural conditions.
• Identified four criteria for minimal functional agency: persistent objectives, policy continuity, counterfactual sensitivity, and self-generated subgoals.
• Concluded that minimal functional agency does not require consciousness or moral status, operating purely at the functional level.

Abstract