New Research Explores Autotelic AI and Self-Generating Goals

This research delves into the emerging field of autotelic AI, which challenges the traditional paradigm of AI systems operating with externally defined goals. Instead, autotelic agents are designed to autonomously discover and generate their own objectives. The paper explores the ramifications of this approach across various AI concepts, including intrinsic motivation, resource-driven learning, and the agent's embeddedness within its environment. A core challenge identified is not merely how an agent generates goals, but how it establishes and contextualizes the "self" to which these goals are attributed. The study suggests that an agent must both perceive its own boundaries to act effectively and transcend those boundaries for deeper understanding. The authors propose a unified framework, extending it with quantum formulations, philosophical interpretations, and a concrete implementation using Large Language Models.