Pushmeet Kohli
Last reviewed
Jun 7, 2026
Sources
8 citations
Review status
Source-backed
Revision
v1 · 1,361 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 7, 2026
Sources
8 citations
Review status
Source-backed
Revision
v1 · 1,361 words
Add missing citations, update stale details, or suggest a clearer explanation.
Pushmeet Kohli is an Indian-British computer scientist who serves as Vice President of Research at Google DeepMind, where he founded and leads the AI for Science and Strategic Initiatives unit. The group applies machine learning to scientific problems in biology, materials, mathematics, fusion, software and cybersecurity, and is credited with a string of high-profile systems including AlphaFold, GNoME, FunSearch and AlphaProof. [1][2] Kohli concurrently holds the title of Chief Scientist of Google Cloud, and was named one of the 100 most influential people in artificial intelligence by Time magazine in 2023. [2][3]
Kohli's research spans computer vision, discrete optimization, program synthesis, AI safety and computational biology. According to Google Scholar his publications have been cited more than 180,000 times, placing him among the most heavily cited researchers in the field. [4]
Kohli grew up in Dehradun, a city in the northern Indian state of Uttarakhand. [1][5] He studied computer science and engineering at the National Institute of Technology, Warangal, graduating with a Bachelor of Technology (BTech) degree. [1]
He then moved to the United Kingdom for doctoral study at Oxford Brookes University, where he was supervised by the computer-vision researcher Philip Torr. Kohli completed his PhD in 2007 with a thesis titled "Minimizing dynamic and higher order energy functions using graph cuts," which dealt with efficient inference in probabilistic models for vision. [1] The thesis won the British Machine Vision Association's Sullivan Doctoral Thesis Award and was a runner-up for the British Computer Society's Distinguished Dissertation Award. [1] After his doctorate he held a postdoctoral position at the Psychometric Centre of the University of Cambridge. [1]
Following his postdoctoral work, Kohli joined Microsoft Research, where he spent roughly a decade across the company's labs in Redmond, Cambridge and Bangalore. [1][2] During this period he served as a technical advisor to Rick Rashid, Microsoft's Chief Research Officer, and rose to become Director of Research for the Cognition group. [1][2] His Microsoft-era work concentrated on computer vision and machine learning, including widely cited contributions to real-time 3D reconstruction and scene understanding. The KinectFusion system for real-time dense surface mapping (2011) and a method for indoor segmentation and support inference from RGB-D images (2012) remain among his most cited papers outside of his later biology work. [4]
Kohli joined DeepMind in 2017. [6] He founded and now heads the unit variously described as the AI for Science and Strategic Initiatives unit, charged with directing the lab's machine-learning capabilities at problems in the natural sciences and other strategically important domains. [2] As Vice President of Research he has overseen a portfolio of systems that have appeared in journals such as Nature and Science, and he has become one of DeepMind's most prominent public voices on the use of AI in scientific discovery. By 2025 and 2026 he was also identified as Chief Scientist of Google Cloud, a role he held while presenting Google's "Gemini for Science" tools. [2][3]
The table below summarizes the main stages of his career.
| Period | Role | Organization |
|---|---|---|
| to 2007 | PhD researcher, computer vision | Oxford Brookes University |
| after 2007 | Postdoctoral fellow, Psychometric Centre | University of Cambridge |
| late 2000s to 2017 | Researcher rising to Director of Research, Cognition group | Microsoft Research |
| 2017 to present | VP of Research; founder and head, AI for Science and Strategic Initiatives | Google DeepMind |
| 2025 to present | Chief Scientist | Google Cloud |
Kohli is best known for leading and supervising a large group of "Alpha"-branded scientific AI systems at DeepMind, although the systems are the product of large teams and he is rarely the sole or first author. His unit's flagship achievement is AlphaFold, the deep-learning system for predicting the three-dimensional structures of proteins. The 2021 Nature paper "Highly accurate protein structure prediction with AlphaFold" has accumulated more than 50,000 citations, one of the most cited papers in modern AI. [4] The AlphaFold effort spawned the DeepMind spin-off Isomorphic Labs, which applies the technology to drug discovery, and its principal architects Demis Hassabis and John Jumper shared the 2024 Nobel Prize in Chemistry for protein-structure prediction. [2][4] A 2024 successor, AlphaFold 3, extended structure prediction to interactions among proteins, nucleic acids and other biomolecules. [4]
Beyond biology, Kohli's group has produced systems across several scientific fields:
| System | Year | Domain |
|---|---|---|
| Magnetic plasma control (with EPFL) | 2022 | Nuclear fusion / tokamak control |
| AlphaTensor | 2022 | Discovering matrix-multiplication algorithms |
| GNoME | 2023 | Discovery of new stable materials |
| FunSearch | 2023 | Mathematical discovery via program search |
| SynthID | 2023 | Watermarking AI-generated content |
| AlphaProof and AlphaGeometry 2 | 2024 | Formal mathematical reasoning |
| AlphaProteo | 2024 | Designing novel protein binders |
| AlphaGenome | 2025 | Predicting effects of genome variants |
| AlphaEarth | 2025 | Earth observation and mapping |
In 2022 DeepMind and the Swiss Plasma Center at EPFL reported in Nature that a reinforcement learning controller had learned to shape and stabilize superheated plasma inside the TCV tokamak, one of the most challenging real-world control tasks to which the method had been applied. [7] The work was widely cited as an example of AI accelerating nuclear fusion research.
In mathematics, Kohli's teams built AlphaProof, a system that trains itself to write formal proofs in the Lean proof language by coupling a language model with the AlphaZero reinforcement-learning algorithm. Working together with AlphaGeometry 2 at the 2024 International Mathematical Olympiad, the systems solved four of six problems and scored 28 of a possible 42 points, reaching the standard of a silver medalist for the first time. The solutions were graded by the mathematicians Timothy Gowers and Joseph Myers. Kohli described the result as "great progress in the field of machine learning and AI," noting that no prior system had solved problems "at this success rate with this level of generality." [8]
His unit has also pioneered AI agents built on large language models, including AlphaEvolve, a general-purpose evolutionary coding agent, and an "AI Co-Scientist" intended to generate and test scientific hypotheses. Earlier programming-related systems from the group include AlphaCode for competitive programming and work on AlphaMissense for classifying disease-causing genetic mutations. [2]
Kohli's citation record reflects the breadth of his output. As of 2026 his Google Scholar profile reports the following indicators. [4]
| Metric | Value |
|---|---|
| Total citations | over 182,000 |
| h-index | 134 |
| i10-index | 343 |
He was included on the inaugural Time 100 AI list in 2023, which recognized leading figures shaping artificial intelligence. [3] In computer vision he has received the Koenderink Prize, awarded for work that has stood the test of time, in addition to the Sullivan Doctoral Thesis Award noted above, and he has collected best-paper awards at conferences including ECCV and CVPR. [1] He has also served as an ACM Distinguished Speaker. [1] Through his role he is frequently invited to speak on AI and scientific discovery, including appearances at venues such as Sequoia Capital's podcast and SXSW London, and he has written on the prospect of AI systems that "begin to do science" rather than merely assist it. [2][3]