New AI Finds and Steers Life-Like Patterns All on Its Own

Picture a digital pond where simple rules create swirling, moving shapes that look alive. For years, scientists could only set the pond in motion and watch what happened. Now, researchers have built an AI agent called CARL that jumps into the simulation and interacts with it. Using a technique called autotelic reinforcement learning (AI that sets its own goals and learns through trial and error), CARL experiments with tiny interventions, nudging patterns this way and that to see what works.

In the digital world of Lenia, a well-known playground for life-like patterns, CARL did three things that stood out. It discovered stable 'solitons' — self-sustaining wave-like shapes that keep their form — much more reliably than older computer methods. It learned to control the direction those shapes moved, not just create them. And perhaps most exciting, human testers could control CARL in real time: they gave simple directional commands like 'go left' or 'turn up,' and the AI translated those into precise low-level tweaks to guide a soliton through a maze.

The AI was trained across many different scenarios, so it generalized to new conditions it had never seen before, without needing extra training. This 'zero-shot' ability is a big deal because it means the skills carry over to unfamiliar situations. The team describes this as a step toward 'artificial experimentalists' — AIs that can independently explore and control complex systems, or work alongside humans as smart lab partners.

The implications go beyond digital play. Complex systems appear everywhere: cells organizing into tissues, chemical reactions forming patterns, even climate and traffic flows. A tool that can actively experiment, adapt, and steer such systems could help scientists discover how they work and guide them toward desired outcomes. For now, CARL works in a virtual world, but the approach points to a future where AI acts not just as a simulator, but as a hands-on explorer and controller of the messy, emergent phenomena all around us.