Player Expectations of Strategy Game AI

The behaviour of AI in modern strategy games is universally recognised as flawed. To compensate for this and successfully challenge humans, it must often be given significant advantages, such as luck bonuses, access to extra in-game resources or knowledge of the entire game state. Players often proclaim their dislike of these flaws, discussing nonsensical moves AI opponents have made, or the fact that the AI ‘cheats’ — out of necessity, as creating competent strategy game AI on consumer hardware is incredibly difficult, even with state-of-the-art techniques. Therefore, this thesis asks: what frustrates players about the opponents — human and AI — that they play against? By asking this, we can establish the most efficient ways to improve player experience when facing AI opponents. To answer, we explore the computer science that drives AI, the psychology that drives players, and the nature of game interactivity as a whole. Flaws in a range of popular strategy games were investigated, forming a grounded theory on how AI play typically annoys strategy game players. We find that players expect their opponents to conform to a set of expectations. Two scenarios were crafted for an existing strategy game. A mix of qualitative and quantitative methods were used to evaluate how players’ experience of one of those expectations — tension — changes under different, controlled conditions. We find that tension can be observed, and is connected to both player uncertainty and perceptions of power. In addition, analysis of player experiences allowed extraction of practical, concrete methods with which game developers can directly influence player experiences of tension in-game. A further experiment clarifies that investment is also connected to tension, but that it is more effective to phrase it as need when questioning players about their investment in a given objective. It also demonstrates that too little information given to players can remove the connection between perceived powers and tension. Finally, we connect our findings to the current literature on player experience in games, and highlight where further work needs to be done.