Father of Reinforcement Learning Richard Sutton Launches New AI Startup

Richard Sutton, widely considered the father of modern reinforcement learning, has launched a new AI startup, Oak Lab, alongside researcher Khurram Javed. 

Sutton and Javed announced the same in a post on X. The latter revealed that Oak Lab has developed a “fairly complete roadmap” centred on its OaK architecture, which aims to create “animal-like intelligence that learns purely from its own experience”. 

The company plans to build a prototype over the next few years that it says would resemble “a baby learning in its first year” more closely than today’s AI systems.

Rather than immediately pursuing large-scale foundation models, Oak Lab said its research strategy will first demonstrate the limitations of current AI methods in simple environments, then develop domain-independent algorithms to address those shortcomings. 

The company said it will only build larger systems after making sufficient progress on core learning algorithms.

The startup’s broader research agenda focuses on continual learning, reinforcement learning and AI agents that adapt in real time without relying on static datasets or replaying stored experiences. 

According to Oak Lab, its long-term goal is to develop AI systems that can continuously learn and plan while operating efficiently in dynamic environments.

Sutton and Javed previously worked together at Keen Technologies, John Carmack’s AI research company, where they were part of a small team exploring reinforcement learning, robotics and real-time decision-making systems. 

Javed joined Keen as a research scientist in October 2024 after completing his PhD at the University of Alberta under Sutton’s supervision.

Sutton, who won the Turing Award for reinforcement learning in 2024, has been an outspoken critic of scaling current deep learning techniques. 

In recent months, he has argued that generative AI models excel at imitation but lack the ability to autonomously evaluate their own outputs and discover new knowledge, positioning continual learning as a more promising foundation for future AI.