Optimizing AI-Driven Chess Bots: Strategies for Balancing Performance, Accuracy, and Computational Efficiency

This paper presents a novel AI-driven chess engine that integrates a lightweight deep learning architecture with an uncertainty-aware Monte Carlo Tree Search (MCTS) framework. Unlike traditional engines that rely on brute-force search, our model utilizes reinforcement learning with a transformer-based neural network to optimize decision-making under computational constraints. We trained our system using high-quality chess datasets from Lichess and implemented Proximal Policy Optimization (PPO) for stable learning. Experimental results demonstrate that our model achieves a 79.5% win rate, a move accuracy of 92.3%, and a 37.7% reduction in inference time compared to AlphaZero, making it well-suited for real-time and resource-constrained applications. Furthermore, our findings suggest that similar AI optimization techniques can be applied to other domains requiring strategic decision-making, such as robotic control and financial modeling. Future work will explore adaptive neural networks and energy-efficient computing to further enhance performance and accessibility. The framework achieves a 79.5% win rate against Stockfish 16, with a 37.7% reduction in inference time compared to AlphaZero.