TOMO KIHARA / 木原 共

Game of Possible Lives is an arcade game where a locally running AI narrates the life of a statistically plausible person. Players guide this person through major crossroads in life, such as education, work, and relationships, and follow the consequences of each choice until the very end.

The work joins a long lineage of games that have tried to compress a human life into a set of rules, from Moksha Patam in ancient India to Milton Bradley's Checkered Game of Life (1860). Each encoded its era's assumptions about how a life should be lived, forming a playable archive of normative beliefs. Game of Possible Lives asks what happens when the role of narrator for these games passes to an AI.

In this game, Gemma 3, an open-weight language model, runs locally inside the cabinet, generating each life in real time. Each life begins as a fictional person, where age, gender, income, and residence are assembled from a country's national census data into a combination that could plausibly exist in that country today. Because such models are trained on vast amounts of human text, the worlds they produce mirror the present, its patterns and its biases.

When a life ends, players return once to an earlier decision and follow the path not taken, seeing how alternate choices lead to different outcomes.

By running the model locally and preserving it as it exists today, the cabinet becomes an experiential archive of the present, a distorted mirror of the forces shaping life now, one a future audience can return to and play. The version shown at Mori Art Museum's Roppongi Crossing 2025 draws on Japan's latest census data.

→More info on the game

_{*The images in this game are generated by an image generation model developed by pixel artists themselves. For its training, the model utilizes consented artwork data from a community of creators who are compensated for their contributions.}