Build intelligent systems that can learn, reason, code, use tools, and act across practical settings while remaining guided by human agency, judgement, values, and oversight.
AI Future
Foundations for next-generation intelligence.
AI Future explores what it would take to build systems that can reason, code, remember, use tools, stay grounded, and operate reliably in open-ended settings.
Many baseline capabilities are now widely accessible, but agency, critical thinking, judgement, taste, trustworthy long-horizon action, and the deeper foundations of intelligence remain open.
Agent architectures, judgement, coordination, evaluation, reliability, physics of intelligence, low-energy AI, and full embodiment.
Core Directions
Agency, judgement, and new foundations for intelligence.
Agent ecosystems
Beyond a single model, the long-range goal is to understand how multiple agents, tools, and memory systems can work together while preserving human agency and judgement.
Critical thinking and memory
Structured reasoning, memory, retrieval, evaluation, and system-level capability remain central because capability without critical thinking and judgement is not enough.
Alignment and safety
Truth, values, oversight, and practical governance are treated as integral parts of the system design problem.
Physics of intelligence
Work on first-principles accounts of intelligence as a physical, non-equilibrium process involving energy, memory, prediction, control, abstraction, and structure transfer.
Low-energy and operator-first AI
Long-term alternatives to scale-heavy AI, including operator-first architectures, sparse reuse, symmetry-aware design, and systems judged by adaptation cost as well as capability.
Embodiment and quantum systems
Embodied multisensory intelligence, world-coupled cognition, and hybrid quantum-classical AI systems for circuit invention, verification, partitioning, and honest advantage assessment.
- AI: The tipping point, VASEA, 2025
- The 2024 Nobel Prize in Physics, public lecture, 2024
- Deep learning and reasoning: Recent advances, VIASM Summer School, 2023
- Machine reasoning tutorials and invited talks across 2020-2024