Automatic Generation of Board Game Manuals

Learning state-action features for general game playing

“…In this regard, the focus in this thesis is on distilling trained policies into various forms of decision trees, such that they can identify small sets of key features and be used as explanations of relevant tactics to human players. This could, for example, produce useful explanations or visualisations for automaticallygenerated tutorials or instruction manuals for arbitrary games (Green et al, 2018;Stephenson et al, 2022). Another type of insight that is briefly considered is an analysis of a game's tactical or strategic depth (Lantz et al, 2017), which can be a useful indicator of a game's quality (Browne, 2009(Browne, , 2012Apeldoorn and Volz, 2017).…”

“…In contrast, in this chapter one of the aims is to extract general, simple tactics for general use throughout an entire game (or substantial portions thereof ). These could, for example, be included in automatically-generated instruction manuals for humans (Stephenson et al, 2022). Coppens et al (2019Coppens et al ( , 2021; Deproost (2021) distil trained policies into various forms of decision trees and rules, which can lead to local (state-specific) as well as global (game-wide) explanations of policies.…”

“…The new types of decision trees proposed in Chapter 8, which operate on state-action pairs as inputs, are a first step towards extending previous work on explaining RL-based policies (Coppens et al, 2019(Coppens et al, , 2021Deproost, 2021) to domains with large action spaces. Explainability in games (Silva et al, 2021;Pálsson and Björnsson, 2022;Stephenson et al, 2022), search Kaisers, 2020, 2021), RL (Liu et al, 2019;Hilton et al, 2020;Lin et al, 2021), and AI more gen-IMPACT PARAGRAPH 275 erally (Molnar, 2020;Zhang et al, 2021), are increasingly recognised as crucial fields of research. The successful results in terms of producing stronger agents through feature selection are furthermore of interest to game AI and RL research in general.…”