The Next Frontier: Welcoming AI Pioneer Jürgen Schmidhuber to Sakana AI

Sakana AI says Jürgen Schmidhuber is joining as chief scientific advisor, while keeping his current roles, and will also work with its Recursive Self-Improvement lab. The company is pitching the move as part of a bigger shift: AI moving out of screens and into the physical world.

Schmidhuber is framed by Sakana AI as a giant in the field, with work on World Models, meta-learning, the Gödel Machine, and deep learning techniques from 1991. He, in turn, cast the move as a chance to connect Japan’s history in neural networks and robotics with what he called physical AI.

That Japan angle runs through the whole announcement. Sakana AI says it wants to reverse the flow of AI talent leaving the country and turn Tokyo into a place that attracts top researchers. Schmidhuber will travel to Tokyo regularly, the company says, to work with its team and support the wider Japanese AI ecosystem.

The technical bet is on Agent-Native World Models, systems that can simulate the physical consequences of actions before anything happens in the real world. Sakana AI says those models could help with manufacturing, supply chains, and robotic deployment by cutting down on costly trial and error.

The RSI Lab sits at the center of that plan. Sakana AI says its goal is a compounding loop of scientific discovery, with systems that research, discover, rewrite their own code, and interact with the physical world. The company also ties the work to earlier projects such as Darwin Gödel Machine and The AI Scientist.

It’s an ambitious pitch, and also a very Sakana AI pitch: Japan as the place where AI gets its body, not just its mouth.

The industry keeps pretending the next breakthrough will be another chat window with a nicer font. This story says the real race is for systems that can reason about physics, logistics, and robots without making a mess in the real world first. That is the harder problem, which is exactly why it matters.