AlphaGo Co‑Architect Thore Graepel Launches AI Reasoning Startup Metis

Former DeepMind researcher and AlphaGo co‑architect Thore Graepel has announced the launch of Metis Reasoning, a London‑based startup that aims to bring structured search and probabilistic planning to the next generation of artificial intelligence systems.

Why Metis Reasoning Matters

Metis focuses on symbolic reasoning techniques that complement the statistical learning methods dominating today’s AI landscape, promising more transparent and controllable decision‑making in complex environments.

From AlphaGo to Probabilistic Planning

Graepel’s work on AlphaGo demonstrated the power of combining deep neural networks with Monte‑Carlo tree search. Metis builds on that legacy by extending the search component into a general‑purpose reasoning engine capable of handling uncertainty across domains such as robotics, logistics and finance.

Funding and Early Backers

The startup is raising a “tens of millions” round, with participation from venture firms that have previously backed DeepMind alumni and AI‑focused enterprises.

• Seed capital from AI‑focused VC firms
• Strategic commitments from robotics and autonomous‑vehicle companies
• Advisory board featuring former DeepMind and Google researchers

Potential Applications

Metis’s technology could improve task planning for autonomous drones, enable more reliable supply‑chain optimization, and support decision‑support tools that require rigorous probabilistic reasoning.

Our goal is to give AI systems a "thinking" layer that can reason about possibilities, not just patterns," Graepel said in a recent interview.

The launch marks a notable shift toward hybrid AI approaches that blend learning‑based perception with formal reasoning, a trend many analysts see as essential for scaling AI safely and responsibly.

For more details, see The Robotics Media coverage of Metis Reasoning launch.