Theory
The hardest problems in AI safety may not be solvable with experiments alone — they require the kind of foundational thinking in mathematics and philosophy that gives the field something solid to build on. Streams in this track work on agent foundations, formal models of trust and agency, mechanistic interpretability theory, and AI welfare. We're looking for researchers with deep mathematical maturity who want to tackle the problems that will still matter when AI systems are far more capable than they are today.
- Initial application: No track-specific questions.
- Stream applications & follow-up: Apply to individual streams; follow-up includes interviews or additional assessments depending on the stream.
This track works on problems where the goal is durable conceptual progress rather than experimental results on today's models. The bet is that some of the hardest alignment questions, such as around agency, optimization, trust, and the structure of cognition, will not be settled by scaling current empirical techniques, and that mathematical and philosophical foundations will matter when AI systems are much more capable than they are now. Projects here cover agent foundations, formal models of trust and agency, mechanistic interpretability theory, and AI welfare. Methods are largely paper, pen, and proof, though some work intersects with empirical interpretability or formal verification.
We are looking for researchers with serious mathematical maturity and a willingness to sit with problems where the right formalization is itself part of the work. Essential traits are research independence (theory questions are open-ended and require self-direction), fluency with formal reasoning (proofs, probability, type theory, dynamical systems, or analogous), and the ability to write clearly about abstract ideas. Strong candidates have come from mathematics, theoretical computer science, theoretical physics, formal philosophy, and economic theory, but background is less load-bearing than demonstrated ability to do hard formal work, ideally with written output we can read.
Fellows are matched to mentors based on fit and produce concrete artifacts by program end: papers, technical reports, conceptual write-ups, or formal results. Target audiences include the agent foundations and alignment theory communities, alignment-relevant teams at frontier labs, and academic venues for formal work. Theory outputs typically have longer time horizons than empirical ones, and we expect many fellows to continue refining results past the program.
In this stream we will explore extensions and implications of our discovery that neural networks pretrained on next-token prediction represent belief-state geometry in their activations. We will build on this fundamental theory of neural network representations in order to discover what AI systems are thinking, and understand their emergent behaviors.
My MATS fellows will do philosophical thinking about multi-agent intelligence and how agents change their values. This will likely involve trying to explore and synthesize ideas from game theory, signaling theory, reinforcement learning, and other related domains.