Microsoft

Microsoft's single AI and machine learning patent tackles the problem of making NPC behavior adaptive without sacrificing creative control. The system restricts an AI model to choosing only from a predefined list of developer-authored actions, never inventing its own responses, while the selection itself adapts based on the NPC's memory of past interactions and the current state of the game world. Traditional approaches require developers to hand-script every possible NPC reaction in spreadsheets or dialogue editors, producing interactions that feel rigid and repetitive. By letting fixed choices compose over time in response to player actions, the method aims to create characters that evolve without stepping outside the boundaries developers set.

The Platforms patent addresses a gap in how games track help that happens inside a session rather than through external resources like forum posts or YouTube walkthroughs. When another player takes over and completes a challenge using the original player's saved game state, the system captures the updated state and ties any resulting achievements to an assistance record in the user database. This creates a persistent link between the accomplishment and the fact that someone else performed it, marking the achievement as assisted rather than independently earned.