Research and writing on Generative Collective Intelligence — a physics-grounded account of how groups of humans and AI agents form, revise, and converge on shared knowledge.
A mathematical framework for collective reasoning that treats GCI as a distinct wave of AI — additive to generative models, not a replacement. The group's reasoning is modelled as a field over a graph of agents, with convergence properties that can be measured and tuned.
CrowdSmart is the operational system where the theory is tested. It runs Bayesian belief networks, Fisher-Rao exploration, and spectral ranking on live organizational deliberations — turning the intelligence already inside an organization into something legible and measurable.
Co-founded with John Seely Brown and Kim Polese to bring the same collective-reasoning machinery to civic and social settings — places where the hardest problems are not technical but epistemic, and where ordinary disagreement is itself information.
The first national-PBS broadcast on AI, recorded four months after the KEE System launched. I come in around the fourteen-minute mark to demonstrate KEE — the Knowledge Engineering Environment I co-created — applied to a nuclear-reactor diagnostic. The rest of the panel was extraordinary to be in a room with.
Boston Global Forum — panel discussion on where generative AI fits into decision-making at scale.
An ongoing program that honors strong AI research work coming out of Vietnam.
An early public articulation of the physics-to-cognition bridge.
On compute efficiency, energy constraints, and where collective reasoning changes the energy ledger.
Co-creator of the Knowledge Engineering Environment — the frames-based AI platform built at IntelliCorp, demonstrated widely throughout the 1980s.
Dissertation on parametric instabilities and ferromagnetic resonance in polycrystalline films. The mathematics of coupled spins becomes, fifty years later, the mathematics for synthesizing knowledge from multiple sources.
Led the AI research group. First extended encounter with symbolic systems, expert systems, and the engineering of machine reasoning — before leaving, on Feigenbaum's invitation, for IntelliCorp.
Co-creator of KEE. First AI/expert-systems company to go public (1983). Recruited from Carnegie Mellon, MIT and Stanford; around 27% of staff held PhDs.
One of the earliest pure e-commerce platform companies. Went public in 1997. An Apple spinout working with Steve Case and AOL in the formative years of the web.
AI-enabled large-scale consumer brainstorming for LEGO, NBC, Procter & Gamble. The first real operational footprint for what would become collective-intelligence tooling.
Bayesian networks and transformer models — before “LLM” was a household acronym — applied to venture decision-making. Titan Fund I portfolio: 71% survival vs. the 38% baseline.
With John Seely Brown. Extends the collective-intelligence framework into civic and social benefit contexts where the hardest work is epistemic, not computational.
Teaching and research affiliation around AI, collective intelligence, and human-centered systems.
Dr. Thomas Kehler has worked in artificial intelligence for fifty years. He is Chief Scientist and Co-founder of CrowdSmart, Co-founder of Common Good AI with John Seely Brown, and holds an Industry Faculty appointment at UC Berkeley. He was previously CEO of IntelliCorp — the first AI/expert-systems company to go public — and CEO of Connect, one of the first e-commerce companies to go public.
His current research concerns Generative Collective Intelligence: a mathematical framework that treats collective human-and-AI reasoning as a field over a graph of agents, with convergence properties that can be measured, tuned, and engineered. The theoretical core is a direct descendant of his 1972 doctoral work on ferromagnetic resonance in polycrystalline films, and is being developed in collaboration with Scott Page (Michigan), Alex Pentland (MIT), Martin Reeves (BCG), and John Seely Brown.
He has authored over twenty publications in AI, natural-language processing, and physics, and holds four U.S. patents. He has served on the Information Technology Advisory Board of the National Research Council and on a range of corporate, academic, and non-profit boards.