We study civilised agentic AI as a computer science problem - generative models, multi-agent systems, incentive mechanisms, trustworthy evaluation, and human-agent interaction.
Each direction turns the lab's manifesto into concrete research on algorithms, systems, evaluation protocols, and computational institutions for agentic AI.
We study controllable image generation, visual editing, multimodal prompting, and creative agents that can plan, critique, and revise visual artifacts with users.
We investigate video generation, temporal consistency, scene dynamics, and agent-controlled media systems for narrative, education, and simulation.
We design and evaluate systems of multiple AI agents that coordinate, negotiate, delegate, compete, and form social or organizational structures.
We study computational markets where humans, agents, models, tools, and data services interact through incentives, reputation, pricing, allocation, and governance rules.
We develop benchmarks, auditing methods, reliability tests, provenance mechanisms, and safety protocols for agents that act across tools, users, and institutions.
We build interfaces and computational models that help agents reason about cultural context, plural values, user intent, and human judgement in real workflows.
