Comparing Traditional and Deep Learning Approaches in Developing Chess AI Engines

Artificial intelligence (AI) has a long and storied history with chess, dating back to the early days of AI research. Chess engines have served as a crucial benchmark to evaluate the progress of AI algorithms, as chess is a complex game with many possible moves and strategies. In this paper, we contribute to chess AI’s ongoing advancements by presenting two chess engines. The first engine leverages conventional AI methodologies, including Alpha-Beta Pruning, Quiescence Search, and sophisticated Move Ordering techniques, to construct a formidable and competitive chess engine possessing an efficacious evaluation function. In contrast, the second engine employs a simplified Alpha-Beta Pruning algorithm with a deep learning model, specifically an artificial neural network, which functions as a feature extraction mechanism for identifying optimal moves. We evaluate the performance of both engines against the renowned Stockfish engine, with the traditional engine attaining an ELO rating of 2200. At the same time, the deep learning model secures an ELO rating of 1900.