The evolution of non-player characters (NPCs) in video games goes back decades: the 1980 arcade game Pac-Man was a milestone because its ghosts each reacted individually to the player’s movements. Since then NPCs have taken on many roles, from populating game worlds to participating in story-driven dialogues.
For a long time those interactions followed prewritten scripts. The arrival of generative artificial intelligence changed that model by enabling more free-form, natural-language conversations: players no longer need to choose from preset options and can instead engage in open dialogue with NPCs.
Promise and debate within the industry
Industry participants largely agree that generative technology can significantly broaden what games can do. There is far less agreement, however, about whether that expansion is overall beneficial or poses new dangers.
Replica Studios’ Matrix-themed demo: promise meets practical problems
Shortly after ChatGPT’s 2022 debut, Australian developer Replica Studios released a game demo based on the Matrix franchise that initially looked promising. The ChatGPT-driven NPCs could hold unconstrained conversations with players. When a player used a microphone to tell the virtual city’s inhabitants that they actually lived in a simulation, the characters reportedly panicked.
According to accounts, one NPC began to question whether they were real, while another spoke about needing to break out of the simulation to return to his wife. Those reactions disrupted progress in the demo’s underlying storyline.
Financial strain: massive dialogue volume and daily costs
Eoin McCarthy, Replica Studios’ chief technology officer, told The New York Times that at peak usage players generated more than 100,000 lines of dialogue with the NPCs. Handling that volume of interaction cost the company roughly $1,000 per day.
Those technical and financial burdens ultimately contributed to Replica Studios’ bankruptcy: the small company could not sustain the operating costs of the game.
Lessons and questions going forward
Replica’s experience highlights that while generative AI can create richer and more flexible in-game interactions, integrating it raises substantial design, narrative and cost challenges. Developers need to consider how to constrain NPC behaviour, how to manage the data and expense such systems produce, and how these systems align with story-focused game objectives.
The technology is likely to become more widespread, but the industry has not yet reached consensus on how to balance the creative potential of generative AI with its financial and narrative risks.
