Empirical Analysis of Destiny Dominance in Multi-Agent Systems with Adversarial RL Agent

Bridging Reinforcement Learning and the Free Will vs. Determinism Debate

Abstract

This paper investigates the tension between deterministic attractors (“destiny”) and autonomous agents exhibiting “free will” in multi-agent reinforcement learning systems. We present a theoretical framework establishing conditions under which attractor dynamics dominate adversarial policies, supported by extensive empirical validation. Our results demonstrate a critical threshold attractor strength (k* ≈ 2.5) beyond which all agents inevitably converge to the attractor regardless of their escape strategies. This work bridges philosophical debates on destiny and free will with computational multi-agent systems, offering insights for robotics, neuroscience, and complex systems where autonomy operates within deterministic constraints.