TRustworthy Artificial intelligenCE
@ University of Cambridge
We build trustworthy AI for high-stakes settings like healthcare, education, and public safety, studying when and how it can be trusted, integrated into decisions, and evaluated in the real world.
University of Cambridge · Centre for Human-Inspired AI · King's College
Research
Societal Impacts
AI is evaluated on what it does for the person using it. We study what it does to everyone else: the colleague they brief, the patient they advise, the student they teach. We build experimental paradigms and measurement instruments for these second-order effects, and test where they do harm, where they reverse sign, and how they vary across domains and cultures. The studies run on Externalis; Secondhand AI is the public campaign built on what they find.
Human AI Orchestration
Deciding when an AI system should act, defer, escalate, or stay silent means knowing what the person on the other side is actually good at. We develop methods for inferring, representing, and protecting individual expertise, machine-readable profiles of skill and judgment that travel across contexts and resist exploitation, then build orchestration algorithms that route work by those profiles, optimising not just for the task at hand but for what happens downstream. Deployed and tested on Tailor.
Deployment in Organisations
Findings rarely survive contact with an institution. This is our field arm: randomised deployments and large-scale studies inside schools, health systems, firms, and government, examining what happens to expertise, hierarchy, and institutional capability when AI is embedded in real work. We study how organisations keep the ability to develop the people they depend on, and maintain a public repository of deployments and their outcomes.
People
Principal Investigator
Umang Bhatt 
Assistant Professor of Trustworthy AI · Fellow of King's College
PhD Candidates
Staff
Visiting and Affiliated Students
- Manoj Kahdan RWTH Aachen
- Hannah Liu Imperial
Collaborators
- Katie Collins MIT
- Mateja Jamnik Cambridge
- Francesco Quinzan Oxford
- Ilia Sucholutsky Purdue
- Adrian Weller Cambridge
Publications
- 2026
- 2026
Revisiting Rogers’ Paradox in the context of human–AI interaction
Philosophical Transactions of the Royal Society A
- 2025
- 2024
Building machines that learn and think with people
Nature Human Behaviour
News
- April 2025
Umang taught a three-week course on Responsible AI at the African Institute for Mathematical Sciences in Cape Town. Three AIMS students, Hassan Suliman, Dagmawi Misker, and Ghaida Ibrahim, have since joined the lab as affiliated researchers. Read more.
- March 2025
Agentic AI has a hidden cost: humans recruited to sense the physical world on an agent's behalf, potentially without consenting to it. Umang writes on this in Noema Magazine.
- February 2025
Umang and Ariella will represent the University of Cambridge at the India AI Impact Summit 2026.
Partners
Industry
Academic
Foundations
Join
Work with us
We take students through CHIA's MPhil and PhD programmes in Human-Inspired AI. Name Dr. Umang Bhatt in your application. If you are already at Cambridge, or asking about postdoctoral roles, write to us directly.