More top papers now list two or more of the people shaping AI: 39% in 2020–2024, up from 16%

How often do the people shaping AI write their most-cited papers together? We find that 39% of top papers from 2020–2024 list two or more of them, up from 16% in 1975–1979.

Key findings

• The share with five or more of them is 6% in 2020–2024, compared with 0% in 1975–1979.
• Gemini: A Family of Highly Capable Multimodal Models has the largest team in our count: 25 people.
• George van den Driessche has the most distinct co-authors across their own top works: 44.

All counts include only the people we track, so full teams can be larger. We combine up to 5 of each person's most-cited works: 9,376 unique papers from 2,211 people with trusted author records. We count each paper once and include all of its authors among the people we track, even when it falls outside a co-author's own top works.

Shared authorship over time

The recent estimate comes from 457 of 1,167 papers. The early estimate comes from 26 of 165.

The share with two or more of them rose from 16% to 39%, comparing 1975–1979 with 2020–2024. The share with five or more was 0% and 6%, respectively.

Source: People of Superintelligence analysis of OpenAlex (CC0), as of 17 October 2026.

Data table

Publication period	Papers	2+ of our people	Share	5+ of our people	Share
1975–1979	165	26	16%	0	0%
1980–1984	221	44	20%	0	0%
1985–1989	366	72	20%	0	0%
1990–1994	491	112	23%	2	0%
1995–1999	771	191	25%	6	1%
2000–2004	1,203	321	27%	5	0%
2005–2009	1,308	385	29%	13	1%
2010–2014	1,280	426	33%	18	1%
2015–2019	2,110	757	36%	46	2%
2020–2024	1,167	457	39%	68	6%

These shares cover each person's own top works. Citation age and our coverage of different generations affect which papers enter the selection.

The papers that bring people together

Bars compare 6 papers by the number of authors we track. Gemini: A Family of Highly Capable Multimodal Models brings together 25 of the people we track.

Source: People of Superintelligence analysis of OpenAlex (CC0), as of 17 October 2026.

Data table

Paper	Year	Venue	Authors we track	Authors with this in their own top works
Gemini: A Family of Highly Capable Multimodal Models	2,023	arXiv	25	24
TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems	2,016	arXiv	18	18
Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models	2,022	arXiv	16	12
Highly accurate protein structure prediction with AlphaFold	2,021	Nature	16	16
Mastering the game of Go with deep neural networks and tree search	2,016	Nature	14	14
Mastering the game of Go without human knowledge	2,017	Nature	14	13

24 of the authors of Gemini: A Family of Highly Capable Multimodal Models list it among their own top works.

Each OpenAlex work ID counts once. Separate records for versions of the same research remain separate papers here.

Partnerships that recur

We find 13 shared papers for David Silver and Demis Hassabis, the most for any pair. These papers can enter the selection through any author's top works.

Shared papers for the leading pair

Pair	Paper	Year
David Silver and Demis Hassabis	Human-level control through deep reinforcement learning	2,015
David Silver and Demis Hassabis	Mastering the game of Go with deep neural networks and tree search	2,016
David Silver and Demis Hassabis	Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm	2,017
David Silver and Demis Hassabis	Mastering the game of Go without human knowledge	2,017
David Silver and Demis Hassabis	A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play	2,018
David Silver and Demis Hassabis	Unsupervised Predictive Memory in a Goal-Directed Agent	2,018
David Silver and Demis Hassabis	Grandmaster level in StarCraft II using multi-agent reinforcement learning	2,019
David Silver and Demis Hassabis	Improved protein structure prediction using potentials from deep learning	2,020
David Silver and Demis Hassabis	Applying and improving AlphaFold at CASP14	2,021
David Silver and Demis Hassabis	Highly accurate protein structure prediction with AlphaFold	2,021
David Silver and Demis Hassabis	Discovering faster matrix multiplication algorithms with reinforcement learning	2,022
David Silver and Demis Hassabis	Olympiad-level formal mathematical reasoning with reinforcement learning	2,025
David Silver and Demis Hassabis	Mastering Atari, Go, chess and shogi by planning with a learned model	Unknown

Who works with the most other people?

We count distinct co-authors across each person's own top works. Repeated collaborations count once. Of the 7 people in the leading 6 ranks, including ties, 5 get at least half of their co-authors from a single paper.

For George van den Driessche, Gemini: A Family of Highly Capable Multimodal Models supplies 24 of 44 co-authors (55%).

We also remove papers with ten or more of our people and count the remaining co-authors. This uses the same top-work lists, without adding replacement papers.

Bars compare 6 people. Each co-author counts once per person. Malcolm Reynolds has the most co-authors after removing papers with ten or more of our people: 23. The table includes leaders before and after removing large papers.

Source: People of Superintelligence analysis of OpenAlex (CC0), as of 17 October 2026.

Data table

Person	Co-authors, all top works	Co-authors, papers with fewer than 10 of our people	Top works left	Most co-authors on one paper
George van den Driessche	44	13	2	24
Geoffrey Irving	42	3	2	24
Paul R. Barham	41	3	2	24
Clemens Meyer	40	7	2	24
Demis Hassabis	39	0	0	15
Malcolm Reynolds	36	23	4	24
Oriol Vinyals	36	4	3	17
Andrew W. Senior	25	20	4	15
Caglar Gulcehre	20	20	5	8
Katie Millican	31	20	4	24
Saffron Huang	19	19	5	8
Mehdi Mirza	17	17	5	8