Stanford professor Percy Liang and Simile AI CEO Joon Sung Park — the duo behind Stanford’s “Smallville” virtual-town experiment — have published a blog post arguing that AI is now entering the “simulation era.” They see the field moving through three stages: prediction, reasoning, and simulation. And simulation, they argue, is where the path to superintelligence actually runs.
Liang’s argument: AI’s most impressive achievements so far have come in settings where the environment and the reward function are crystal clear — think Go, an IMO math competition, or writing an entire app from scratch. In those sandboxes, reinforcement learning algorithms can safely try different moves and observe what happens. But real-world problems that center on people are a different story. How does remote work change organizational culture? How do you redesign education for millions of students...