GDC 2025 Session Viewer

This session covers how the logic of possibility theory can be used by game designers and developers to create utility functions for AI systems that rely on utility score evaluation for NPC decision-making. An explanation is provided for how possibilistic logic is applied through the use of contexts that designers can construct based on the information that's relevant to decisions made by NPCs. This session then continues to walk the viewer through a hypothetical example of an in-game decision made by an enemy NPC that can be abstracted to almost any game, as well as a practical example that demonstrates how possibilistic logic was applied to an AI system implementation for an unreleased colony management simulation game.

Takeaway

Offers a way of thinking about and designing utility functions for creating NPC behaviors that are more easily understandable from the players' perspective and in-game context, rather than NPC behaviors that players learn only through repeated encounters.

Intended Audience

Useful to anyone that uses utility systems in their AI architecture and wants to create unambiguous utility functions that are more easily understandable and learned through displayed NPC behaviors.

This talk would be raised at AI designers and developers who are more seasoned utility AI people. It assumes that the listener is at least aware of utility AI systems and have a rough idea of the fact that they work by score ranking optional actions by the application of some utility function.