Comparison of Four AI Algorithms in Connect Four

Abstract:
It is vital to use machine learning methods to complete game competitions beyond the human level. This paper mainly focuses on the performance of four commonly used algorithms in board game artificial intelligence development, including greedy, α-β pruning, principal variation search, and monte carlo tree search. Specifically, this paper compares algorithms' effectiveness by letting them play connect four against each other 12 times between two methods, alternating the first mover and measuring the time of wins and the total moving steps. Based on wins and total steps of each method, the paper analyzes which algorithm gives the best performance and explains why that algorithm gives the best result. As the result, monte carlo has the highest winning rate against all of the other algorithms with relatively smaller moving steps, and it can win the opponents as fastest compare to other algorithms, which shows that monte carlo gives the best performance in connect four against other algorithms.