SAI, a Sensible Artificial Intelligence that plays Go

TL;DR

This paper introduces SAI, a Go AI that models winrates across different komi values using a sigmoid function, enhancing self-play training and score estimation on 7x7 Go.

Contribution

It presents a novel multiple-komi approach with sigmoid modeling, improving reinforcement learning efficiency and score estimation in Go AI.

Findings

Achieved strong playing agents on 7x7 Go.

Successfully modeled winrate as a function of komi with a sigmoid.

Enabled score difference estimation and game decisiveness evaluation.

Abstract

We propose a multiple-komi modification of the AlphaGo Zero/Leela Zero paradigm. The winrate as a function of the komi is modeled with a two-parameters sigmoid function, so that the neural network must predict just one more variable to assess the winrate for all komi values. A second novel feature is that training is based on self-play games that occasionally branch -- with changed komi -- when the position is uneven. With this setting, reinforcement learning is showed to work on 7x7 Go, obtaining very strong playing agents. As a useful byproduct, the sigmoid parameters given by the network allow to estimate the score difference on the board, and to evaluate how much the game is decided.