Evolutionary Game Theory (EGT) solutions become analytically intractable as game complexity increases, thereby limiting its applicability to complex, multi-agent systems. In this paper, we present a new computational framework that redefines evolutionary games as stateless multi-agent reinforcement learning (MARL) problems using independent Proximal Policy Optimization (PPO) agents. The proposed framework can effectively approximate evolutionarily stable strategies (ESS) without requiring closed-form replicator dynamics and extends naturally to multi-player, multi-strategy games. Based on a three-player e-commerce game case study, it is shown that the method aligns with two analytically derived equilibria, E₄(1,0,1) and E₃(0,0,1), with learned strategy probabilities greater than 0.999, thereby demonstrating near-perfect agreement. The thorough sensitivity analysis identifies important parametric thresholds that trigger changes in strategic behavior, explaining why the approach is effective in revealing equilibrium boundaries. Its key contributions are as follows: (1) scaling of MARL-based solutions for ESS approximation in standard evolutionary games; (2) empirical validation of theoretical equilibria through decentralized learning; and (3) a reproducible computational framework that complements traditional analytical EGT. The limitations include the use of a one-shot, stateless game model and validation restricted to two equilibrium states. This paper highlights the potential of MARL as a practical and generalizable tool for studying complex strategic interactions in contexts where traditional solution methods are impractical.