AI generates playful, human-like games
An NYU research team including Associate Professor Julian Togelius developed an AI model that learns to generate human-like goals by studying how people create games. Participants invented nearly 100 games in virtual rooms with various objects, following principles of physical plausibility and recombination. The AI then created its own games based on these principles. When rated by humans, the AI-generated games were indistinguishable from human-created ones, advancing our understanding of goal representation and potentially improving AI alignment with human intentions.
Researchers
