Theo, the AI Physicist
Building an AI Physicist for scientific discovery
AI to explore the laws of nature
The field of physics contains some of our deepest, most foundational questions. Exploring them requires increasingly complex reasoning across domains, which is becoming unfeasible under traditional research approaches. By building an AI physicist that can reason rigorously about complex physical systems, we aim to expand how scientific inquiry is performed.
Such a system would increase both the scale and depth of scientific exploration, advancing our understanding of the universe and ultimately influencing fields built on physical principles, such as energy, materials, technology, and medicine. If our moonshot is achieved, it will mark a transformational step in humanity’s understanding of the universe.
AI physicist architecture
Theo is being built to evolve into a fully autonomous research engine capable of formulating its own questions, proposing new hypotheses, solving complex problems, testing results, and continuously improving itself over time.
Modular by design
At its core, a central Evaluator analyzes intermediate findings and provides feedback to the research modules, including a Deep Literature Search that spans our corpus of scientific knowledge. When a potentially novel direction is identified, it enters a stage of hypothesis-generation and problem solving, using Theo’s Specialized AI Models and Modular Tools to gain clarity on the problem. The resulting predictions are then tested and analyzed by the system, including Simulations and Experimentation capabilities planned for future development.
The output: Dynamic Research Objects
Theo then produces scientific outputs as a Dynamic Research Object (DRO) - a digital-first alternative to the traditional “paper” format that is dynamic, traceable, reproducible, and fully open - viewed through the Theo Notebook UI. Finally, the scientific outputs are then opened up for feedback provided by the system, by Theo Collaborators, and by Reinforcement Learning from Human Feedback (RLHF) capabilities planned for future development.
This interconnected architecture allows the system to capture the creativity and intuition necessary for generating new hypotheses in fundamental science while also maintaining the strict rigor required to validate and develop new theories. The result is an AI system that serves as an infrastructure for open collaboration that researchers can trust and build upon.
How to partner
The AI Physicist is being developed as long-term research infrastructure. Its success depends on technical progress, but also on how it is integrated into the broader scientific ecosystem.
We work with partners who want to shape how new research capabilities are built and stewarded over time. Rather than a single model of participation, we support several forms of partnership depending on institutional goals and expertise.
Foundations of the AI Physicist
Tool integration
Symbolic reasoning, simulations, and literature analysis are combined in a modular way, giving the system multiple paths for exploring problems and testing ideas.
Specialization & orchestration
Individual models are specialized to become experts in an array of physics sub-domains. They are then orchestrated in a network so their outputs reinforce one another, and they are given access to tools that help them remain reliable across the full workflow.
Data curation
Physics corpora are carefully built from open repositories and search, with structured labeling, metadata, and citation-graph analysis to ensure quality and transparency.
As development continues, we aim to refine these capabilities in close dialogue with the scientific community. If you are interested in contributing, we invite you to join our AI Physicist Slack space.
Design principles
Developing AI for fundamental physics requires more than technical performance; it demands systems that researchers can trust and extend upon. Our design principles reflect this commitment: to build an AI Physicist that is rigorous, transparent, and aligned with the scientific method. Not a black box, but an evolving framework shaped by the needs of science itself.
Rigor & transparency
Results should be interpretable and reproducible.
Modularity
Flexible components designed for multi-domain reasoning.
Specificity
Tailored to the structure and demands of science.
Scalability
Built to integrate new logic and domains as science evolves.
Design Principles
Developing AI for fundamental physics requires more than technical performance; it demands systems that researchers can trust, scrutinize, and extend. Our design principles reflect this commitment: to build an AI Physicist that is rigorous, transparent, and aligned with the scientific method — not a black box, but an evolving framework shaped by the needs of science itself.
Rigour & Transparency
Results should be interpretable and reproducible.
Modularity
Flexible components designed for multi-domain reasoning.
Specificity
Tailored to the structure and demands of science.
Scalability
Built to integrate new logic and domains as science evolves.
Join the AI Physicist community
A curated Slack group for researchers and thinkers working at the intersection of AI and science. Join the conversation, share insights, and get early updates as we build the AI Physicist.
Request an invite by filling out the form. We review each submission to keep the group focused and aligned with its purpose.
Join the AI Physicist Community
A curated Slack group for researchers, engineers, and thinkers working at the intersection of AI and science. Join the conversation, share insights, and get early updates as we build the AI Physicist.
Request an invite by filling out the form. We review each submission to keep the group focused, collaborative, and aligned with its purpose.
Values
Curiosity
We foster a relentless curiosity that drives us to explore the unknown and push the boundaries of understanding.
Ambition
We dare to dream big and aim for ambitious goals, challenging ourselves to achieve more than we ever thought possible.
Risk Taking
We embrace risks, knowing that innovation often requires stepping into the unknown and being unafraid of failure.
Collaboration
We believe in the power of collaboration, working together across disciplines and borders to achieve our shared vision.
Results-driven
We are committed to making a positive impact for all of humanity, driving our work towards tangible results that benefit people worldwide.