General Intuition is making an optimistic bet on a new path for AI progress: using video games as training data. While today’s leading models are remarkably capable with language, they often lack a deep understanding of how objects move, collide, and change over time.
Games could help fill that gap. Unlike static text scraped from the internet, video games provide dynamic, interactive worlds where actions have consequences. That makes them a promising source of data for teaching AI systems about physics, space, timing, and cause-and-effect relationships.
A richer training ground for general intelligence
The company’s view is that better world understanding will be essential for more general AI. By learning from game environments, future systems may become better at reasoning about real-world tasks, robotics, simulations, and complex decision-making.
- Games provide structured, visual, and temporal data.
- They allow AI to observe interactions rather than just read descriptions.
- This approach could complement, rather than replace, language-model training.
While still an emerging idea, it highlights an exciting direction for AI research: moving beyond text alone and toward systems that can understand the world more like people do — through experience, motion, and interaction.