Manuel Baltieri
Agent foundations
What makes a system an agent?
Manuel develops mathematical theories for understanding agency: how agents are distinguished from their environments, what they model and how they pursue goals, and how these ideas can support safer AI.
His research at Araya and as an ARIA R&D Creator connects category theory, control, probability, dynamical systems, cognitive science, and artificial life.
Selected work
All publicationsMaths
Mathematical approaches to the study of agents
A review of mathematical approaches to studying agents across research fields.
Cognition
Disentangled representations for causal cognition
Relating causal representation learning to how natural and artificial agents acquire causal knowledge.
(A)Life
Hybrid Life: Integrating Biological, Artificial, and Cognitive Systems
A synthesis of biological, artificial, and cognitive perspectives on life and the formal languages that can connect them.
AI safety
Compositional Behavioral Semantics for State Abstraction in Reinforcement Learning
A unified framework for behavioural structures in reinforcement learning and their safe transfer between abstract and concrete systems.
Latest news
All news
Mathematical approaches to the study of agents
A unified review of mathematical approaches to individuality, normativity, and interactional asymmetry. Continue reading Mathematical approaches to the study of agents
World models of environment, agent and joint agent-environment systems
A formal distinction between environment, agent, and joint agent-environment world models. Continue reading World models of environment, agent and joint agent-environment systems
Current projects
All projectsWorld models
World models for evaluating and interpreting AI agents: their limits, their modular structure, their abstractions, and the channel they model. Continue reading World models
The internal model principle and good regulators
When a controller or agent can be said to model its environment, from the internal model principle to good regulators and possibilistic belief updating. Continue reading The internal model principle and good regulators