CEO Thinks Video Games Trump Internet as Training Data for AGI

When it comes to achieving artificial general intelligence (AGI), large language models just don’t have what it takes. Models like ChatGPT and Claude are great at text, but they’re less skilled at understanding how things actually move through space and time — an essential skill for producing intelligence that generalizes. That gap, it turns out, might be filled by gaming data.

That’s the bet behind General Intuition, a new AI company founded by video game industry veteran and CEO, Mike Robbins. Robbins argues that video games offer a more structured and informative dataset than the internet, which is often noisy and unstructured. Games provide a simulated environment where characters and objects interact in predictable ways, allowing AI models to learn about the physical world.

“Video games are a much more efficient way to train AI models because they provide a controlled environment with clear goals and outcomes,” Robbins said. General Intuition aims to leverage this gaming data to develop AI models that can understand and interact with the physical world. The company is working on applying this approach to areas like robotics and autonomous vehicles, where understanding spatial relationships and object motion is critical.

While large language models have made significant progress in recent years, Robbins believes they are limited by their reliance on internet data. The use of gaming data to train AI models raises interesting questions about the future of AI development. If Robbins is correct, we may see a shift towards more specialized AI models that are trained on specific domains, rather than general-purpose models trained on broad datasets.

Why this matters: The success of General Intuition's approach could have significant implications for the development of artificial general intelligence. If video games prove to be a more effective training ground for AI, it could accelerate progress in areas like robotics, autonomous vehicles, and healthcare. Developers and businesses will be watching closely to see if this approach yields more capable and generalizable AI models.

For consumers, the potential benefits include more sophisticated virtual assistants, improved safety features in vehicles, and more efficient healthcare diagnostics. However, there are still open questions about the scalability and adaptability of this approach, and whether it can be applied to a wide range of domains. As the field continues to evolve, one thing is clear: the quest for AGI is taking a fascinating new turn.