OLIVAW: Mastering Othello without Human Knowledge, nor a Fortune

Abstract

We introduce OLIVAW, an AI Othello player adopting the design principles of the famous AlphaGo series. The main motivation behind OLIVAW was to attain exceptional competence in a non-trivial board game, but at a tiny fraction of the cost of its illustrious predecessors. In this paper we show how OLI-VAW successfully met this challenge.

AlphaZero-Inspired General Board Game Learning and Playing

“…Recently in 2021, Norelli and Panconesi [10] came up with a very interesting paper that is close to our work: They pursue as well the goal to set up an AlphaZero-inspired algorithm at much lower cost than the original AlphaZero [3]. The agent in [10] is trained solely by self-play, is able to play 8x8 Othello and defeat the strong Othello program Edax [11] up to level 10.…”

“…Recently in 2021, Norelli and Panconesi [10] came up with a very interesting paper that is close to our work: They pursue as well the goal to set up an AlphaZero-inspired algorithm at much lower cost than the original AlphaZero [3]. The agent in [10] is trained solely by self-play, is able to play 8x8 Othello and defeat the strong Othello program Edax [11] up to level 10. Although much less computationally demanding than the original AlphaZero [3], their training time took roughly one month on Colaboratory, a free Google cloud computing service offering GPUs and TPUs.…”

“…See Sec. V (Discussion) for further comparison between Norelli and Panconesi [10] and our work. N-tuple networks, which are an important building block of our approach, have shown to work well in many games, e.g., in ConnectFour [14], [15], Othello [16], EinStein würfelt nicht [17], 2048 [18], SZ-Tetris [19] etc.…”

“…Apart from Norelli and Panconesi [10], there are only few works on Othello game learning that actually benchmark against Edax: Liskowski et al [12] presented in 2018 an agent obtained by training a convolutional neural network (CNN) with the help of a database of expert moves. Their agent could defeat Edax up to and including level 2.…”

“…In 2021, Norelli and Panconesi [10] obtained with their system OLIVAW the best Othello results up-to-date: It defeats Edax up to and including Edax level 10. This is a truly impressive result, but it also took considerable computational resources to achieve it: Although much cheaper than Deep-Mind's original AlphaZero, they needed an informal crowd computing project with 19 people for game generation and then about 30 days to train a single agent on Google Colaboratory using GPU and TPU hardware.…”