Training General Intelligence with Gameplay: General Intuition's Big Bet

Why gameplay data matters

Video games provide dense, interactive, multimodal environments where agents can repeatedly explore, fail, and adapt at scale. Gameplay captures decision-making under uncertainty, visual-motor coordination, social interaction, and long-horizon planning - all signals useful for building agents that generalize beyond narrow tasks. General Intuition's funding round validates investor belief that this simulated curriculum can speed progress toward more human-like intuition.

Potential business and technological impacts

If models trained on gameplay transfer effectively to physical or enterprise domains, industries like robotics, logistics, and autonomous systems could gain cheaper, safer training grounds than real-world trials. This would lower the barrier for companies to develop adaptive agents and could spawn marketplaces for curated simulation datasets and environments.

Limitations and risks

Sim-to-real transfer remains a core technical challenge: visual fidelity, dynamics mismatch, and reward shaping can produce brittle policies. There are also intellectual property and licensing questions around using commercial game assets at scale. Commercial success will depend on demonstrable transfer and sample efficiency compared with real-world alternatives.

What leaders should do now

Evaluate simulation-driven pilots for use cases with high-cost real-world training. Form partnerships with simulation or gaming firms to access diverse synthetic data. Insist on metrics that measure transfer effectiveness (e.g., zero-shot or few-shot performance in real settings) and build internal capabilities in domain randomization and fine-tuning to bridge the sim-to-real gap. These steps let organizations experiment safely while watching whether this approach delivers operational advantages.