Making Strategy Opponents Less Frustrating with Daniel Gomme
What do players want from an AI opponent? This question, while valuable to ask, is too broad to be investigated systematically as there are too many different kinds of games. It has therefore been narrowed down to a particular genre that presents major challenges for AI: strategy games. Determining what players want is also difficult because players don’t say what they want unless asked directly; it’s much easier to identify what aspects of strategy game opponents frustrate players, because players routinely complain about such flaws in public forums. Studying such responses will reveal common flaws found in current commercial game AI opponents, which can then be addressed.
Forum comments expressing opinions on AI have been sampled from communities around two modern and popular strategy games: Endless Space 2 and Sid Meier’s Civilization VI. These have been assessed and analysed to construct a grounded theory presented in this talk. The results show that a sense of tension seems to be key to having satisfying opponents in strategy games, as well as the degree of influence players have over each other. How this seems to work and what can contribute to it is discussed.
These findings can be tested by modifying AI opponents around these desired properties, and review human players’ opinions about the experience afterwards. Taking the results further, it’s reasonable to ask if they could be applied to human players too. If so, there are prospects for guiding game designs towards more enjoyable human player interaction.
This talk briefly covers the methodology, before discussing in detail the results, and the possible implications they have for current and future AI design.