Overview of Machine Learning Bots Capable of Achieving Top-level or Superhuman Performance in 2d Competitive Games
Xiaoyu Li · Applied and Computational Engineering · 2025
With recent developments in machine learning, bots have conquered many games that were previously believed to be too difficult for computers. We evaluated 4 machine-learning bots that mastered 4 different games: AlphaGo for Go, Stockfish for chess, AlphaStar for StarCraft, and Juewu for Honor of Kings. We summarized and compared their fundamental algorithms and discovered two potential patterns: an increase in generalizability causes an increase in the amount of computation needed, and an increase in the complexity of the game environment causes a decline in the neural networks’ performance. To conclude, we discuss the implications of machine learning game bots on neural networks aimed at handling real-life scenarios and artificial general intelligence (AGI).