Video games to train AI: General Intuition's $2.3B bet on real-world agents

The answer is simpler than you'd think. Inside video games, complicated things happen constantly without warning. An AI agent must make rapid decisions in environments that shift unpredictably, with no precise instructions on what to do. That's exactly how the real world works. General Intuition bet that millions of hours of gameplay contain decision patterns remarkably similar to what's needed to solve real-world problems.

Gameplay data is dense with situation-reaction pairs. An AI watches something happen on screen and learns how to respond. It learns when to rush, when to wait, when to improvise. That's a type of learning that text alone simply cannot deliver.

Start with a straightforward concept: record how millions of people play. Every movement, every click, every choice becomes data. General Intuition is building AI models trained on this massive ocean of real human actions. The stated goal is developing agents that cultivate something closer to human intuition instead of raw mathematical calculation.

The training itself demands enormous computing power. Those $320 million will scale the infrastructure needed to process millions of hours of video and movement data. It's not just collecting information, but transforming it into models that perform when the AI faces situations it's never encountered.

Here's where the strategic bet lives. If a neural network learns from video games how to navigate complex environments, it could later transfer that learning to real-world tasks: robotics, planning, dynamic problem solving.

Eventually you'll see robots and autonomous systems behaving less rigidly and more adaptively. Instead of executing programmed instructions step by step, they'll make situational decisions. This applies to drones, self-driving cars, logistics systems. If General Intuition succeeds, they won't be machines following deterministic algorithms anymore, but agents that learn to improvise.

The cost of this bet isn't small. $320 million is a substantial investment in research and training infrastructure. But the autonomous AI agent market could be worth far more if this approach actually works.