UC Santa Cruz

Existing forms of player-assistance, such as game walkthroughs and strategy guides, provide broad, general advice that disengage players from gameplay when seeking help at a particular moment. Furthermore, the assistance is static, unable to adapt to a player's current situation and evolving goals. I introduce Live Coaches as a genre of integrated AI systems that provide high-context and real-time player-assistance. Live Coaches support players in game completion and beyond, through integrating two key design dimensions: liveness, the ability to maintain real-time contextual awareness, and coachness, the ability to leverage deep game knowledge. I analyze prior systems exhibiting Live Coach characteristics and present two newly created systems: ChozoMaps and Rotom. Furthermore, I propose a design framework that enables designers to develop future Live Coach systems within the genre. An ablation study of 10 participants examined ChozoMaps' individual and combined impacts of liveness and coachness on player experiences. Findings illustrate that liveness without coachness leads to context-aware, yet inaccurate guidance, while coachness without liveness provides relevant guidance only after players manually describe their situation. When both dimensions were present, participants received specific help grounded in what the system knew about their current gameplay. These results highlight that preferences for assistance vary widely, illustrating the need for systems that adapt to individual thresholds of wanted assistance within games and beyond.