Geoffrey Irving
Geoffrey Irving’s work connects machine learning with AI alignment. He studies mathematics and computer science at Caltech and Stanford, then works in computational physics and geometry, earns credits on four films, and helps solve Pentago. At Google Brain, he applies machine learning to theorem proving and coauthors TensorFlow; later, he coauthors research on debate and human feedback and leads alignment teams at OpenAI and DeepMind. He joins the UK AI Security Institute as Research Director, becomes its Chief Scientist, and describes an alignment project that brings governments, industry, philanthropists, venture capital, and researchers together. As of October 2026, he is a cofounder and Chief Scientist of Resolution, whose alignment work uses theory and empirical research.
- Now at
- Resolution
- Works
- 48
- Citations
- 25,989
- h-index
- 29
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
2003–2007
Studying computer science
He earns a BSc in mathematics and computer science from Caltech in 2003, then a PhD in computer science from Stanford in 2007.
-
2007–2014
Computational physics on screen
After graduate school, he works in computational physics and geometry at Otherlab, D. E. Shaw Research, Pixar, and Weta Digital, and earns screen credits on four films. He says this work continued until around 2013. In 2014, he coauthors a paper that solves Pentago for a first-player win.
-
2015–2017
Teaching machines to search proofs
He joins Google Brain in 2015 to apply machine learning to theorem proving. In 2016, he coauthors TensorFlow, a large-scale machine learning system; in 2017, he coauthors Deep Network Guided Proof Search.
-
2018–2019
Debate and human feedback
In 2018, he coauthors AI safety via debate, which trains agents to argue while a human judges their claims. A 2019 paper tests fine-tuning language models from human preferences.
-
By December 2023
Leading alignment teams
He leads the Reflection Team at OpenAI and the Scalable Alignment Team at DeepMind. He leaves DeepMind in December 2023 to join the UK AI Security Institute as a Research Director.
-
December 2023–July 2025
Taking alignment into government
He joins the UK AI Security Institute as a Research Director in December 2023 and officially starts in April 2024. By July 2025, he is its Chief Scientist and describes the Alignment Project as an effort to bring governments, industry, philanthropists, venture capital, and researchers together to close gaps in alignment research.
-
May 2026
Planning a nonprofit
He announces plans to leave AISI, move back to the Bay Area, and start a nonprofit alignment research group.
-
As of October 2026
Starting Resolution
He is a cofounder and Chief Scientist of Resolution, a nonprofit alignment research organization. Its program pursues higher confidence through theory and empirical research.
Sources and credits
Photo: aisi.gov.uk; Geoffrey Irving | AISI Team; creator not stated. License: Not stated by source. Source: aisi.gov.uk. Cropped to a square and resized. Ask us to remove or replace this photo.
- Geoffrey Irving
- Situational Awareness in Government, with UK AISI Chief Scientist Geoffrey Irving
- Pentago is a First Player Win: Strongly Solving a Game Using Parallel In-Core Retrograde Analysis
- TensorFlow: A system for large-scale machine learning
- Deep Network Guided Proof Search
- AI safety via debate
- Fine-Tuning Language Models from Human Preferences
- Reflections on AISI
- AI Security Institute launches international coalition to safeguard AI development
- UK’s AISI launches coalition to protect against advanced AI
- Resolution: Scale and Automation for Higher Confidence in Alignment