72% of AI researchers with top works connect into a single 2,592-person co-author group
Data as of
- Researchers in the largest connected group
- 2,592 of 3,593 (72%)
- Share of connected researchers in that group
- 92%
- Median co-author steps in the main group
- 5
- Most researchers reached within 2 steps (Yoshua Bengio)
- 570 (22%)
How closely connected are the people shaping AI through their most-cited papers, who bridges the network across multi-step paths, and which organizations share the most co-authored work? In our analysis of paper teams, we examined how often top works bring multiple tracked authors together on a single paper. Here we join those collaborations into a site-wide co-author network across 3,593 researchers and 14,977 unique works (4,650 co-authored papers forming 10,775 co-author pairs). We estimate that 2,592 of the 3,593 researchers (72%, and 92% of the 2,822 who have at least one tracked co-author) belong to a single connected group, where any two people are separated by a median of 5 co-author steps. Across that 2,592-person group, Yoshua Bengio leads on both shortest-path bridging (6.7% of all shortest paths) and 2-step reach (570 people, or 22% of the group), and Allen Institute for AI (Ai2) and University of Washington share the most co-authored papers by publication-year employers (84).
Key findings
- 2,592 of the 3,593 people with trusted top-work lists (72%) connect into one co-author group (92% of the 2,822 people who share at least one paper with another person we track). Another 230 connected people (6%) sit in 89 separate groups of 2 to 6 people, and 771 (21%) have no tracked co-author on any selected paper.
- Across all 3,357,936 pairs of researchers inside the 2,592-person group, the shortest co-author path has a median of 5 steps (middle half 4 to 6 steps, mean 5.4 steps, maximum 17 steps), and 78% of all pairs (2,604,312) connect within 6 steps.
- Across multi-step paths inside the 2,592-person group, Yoshua Bengio sits on the largest share of shortest paths between researcher pairs (6.7%), ahead of Bernhard Schölkopf (5.4%) and Geoffrey Hinton (3.6%). Within 2 co-author steps, Yoshua Bengio also reaches the most researchers: 570 people (22% of the main group: 67 at 1 step and 503 at 2 steps), followed by Oriol Vinyals (482) and Allan Dafoe (463).
- By publication-year employers across 4,461 dated co-authored papers where at least two co-authors have a known organization, Allen Institute for AI (Ai2) and University of Washington lead all organization pairs with 84 shared papers, followed by Google and Stanford University (72) and Carnegie Mellon University and Stanford University; and Google and University of Oxford (56 each). At Allen Institute for AI (Ai2) and University of Washington, 83 of the 84 papers (99%) have an author at both organizations in the publication year; counting only papers without an author at both organizations in the publication year, Carnegie Mellon University and Massachusetts Institute of Technology lead with 31 papers, ahead of Google and Microsoft (28).
One connected group holds 72% of researchers with top works
Across the 3,593 people with trusted OpenAlex top-work lists and 14,977 unique works, 4,650 papers have two or more tracked authors and form 10,775 distinct co-author pairs. In total, 2,822 researchers (79% of 3,593) share at least one paper with another tracked person: 2,629 (73%) have a shared paper inside their own top-5 list, and 193 (5%) join the network through co-authors’ top-5 lists.
Of those 2,822 connected people, 2,592 (92%, or 72% of all 3,593 researchers) belong to a single connected component. The remaining 230 connected people (6% of 3,593) sit in 89 separate components of 2 to 6 people: 3 groups of 6, 3 groups of 5, 5 groups of 4, 21 groups of 3 and 57 groups of 2. The 3 largest separate groups (6 people each) are Antonio Lieto, David Vernon, José Santos-Victor, Lorenzo Natale, Manuel Lopes and Pierre-Yves Oudeyer; Benjamin J. Kuipers, Bernard Vauquois, John McCarthy, Joseph Weizenbaum, Kenneth Ford and Patrick J. Hayes; Gerard Milburn, Howard M. Wiseman, Jonathan P. Dowling, Kae Nemoto, Raymond Laflamme and Yasunobu Nakamura. The other 771 researchers (21%) have no tracked co-author on any selected paper.
By current employer sector, the share inside the 2,592-person connected group is highest among people with a current role at companies at 84% (901 of 1,071), compared with 70% among people with a current role at universities (1,559 of 2,218) and 70% among people with a current role at institutes and labs (473 of 678). By role category, lab leaders have the highest share inside the main connected group at 91% (90 of 99), and public voices have the lowest share at 45% (38 of 84).
Connected group membership by current employer sector and role category
| Slice | Researchers | In main connected group | Main-group share | In smaller groups (2–6) | No tracked co-author |
|---|---|---|---|---|---|
| All researchers with top works | 3,593 | 2,592 | 72% | 230 (89 groups) | 771 |
| Any current role at: Companies | 1,071 | 901 | 84% | 27 | 143 |
| Any current role at: Universities | 2,218 | 1,559 | 70% | 160 | 499 |
| Any current role at: Institutes and labs | 678 | 473 | 70% | 50 | 155 |
| Role category: Lab leaders | 99 | 90 | 91% | 1 | 8 |
| Role category: Research | 3,336 | 2,420 | 73% | 220 | 696 |
| Role category: Builders | 74 | 44 | 59% | 6 | 24 |
| Role category: Public voices | 84 | 38 | 45% | 3 | 43 |
Horizontal bars show the distribution of shortest co-author path lengths across 3,357,936 pairs of researchers in the 2,592-person connected group, peaking at 5 steps (25.4% of pairs) with a median of 5 steps.
Source: People of Superintelligence analysis of OpenAlex (CC0) most-cited works as of 21 October 2026 and career records as of 22 October 2026.
Data table
| Shortest path length | Researcher pairs | Share of pairs | Cumulative share |
|---|---|---|---|
| 1 step (direct co-authors) | 10,619 | 0.3% | 0.3% |
| 2 steps | 70,482 | 2.1% | 2.4% |
| 3 steps | 303,570 | 9.0% | 11.5% |
| 4 steps | 678,261 | 20.2% | 31.7% |
| 5 steps | 851,865 | 25.4% | 57.0% |
| 6 steps | 689,515 | 20.5% | 77.6% |
| 7 steps | 416,409 | 12.4% | 90.0% |
| 8 steps | 201,630 | 6.0% | 96.0% |
| 9 steps | 84,779 | 2.5% | 98.5% |
| 10 steps | 33,040 | 1.0% | 99.5% |
| 11 steps | 12,100 | 0.4% | 99.8% |
| 12 steps | 4,072 | 0.1% | 100.0% |
| 13 steps | 1,214 | 0.0% | 100.0% |
| 14 steps | 315 | 0.0% | 100.0% |
| 15 steps | 55 | 0.0% | 100.0% |
| 16 steps | 9 | 0.0% | 100.0% |
| 17 steps | 1 | 0.0% | 100.0% |
Inside the 2,592-person connected group, 3,357,936 unordered pairs of researchers have a finite co-author path. Direct co-authors (one step) make up 10,619 pairs (0.3%), and researchers who share a co-author (two steps) make up 70,482 pairs (2.1%). Path counts peak at 5 steps (851,865 pairs, or 25.4%), with the middle half of all pairs spanning 4 to 6 steps (median 5 steps, mean 5.36 steps) and 78% (2,604,312 pairs) connecting within 6 steps. Pairs separated by 10 to 17 steps make up 50,806 pairs (1.5%).
Who bridges the co-author network
Within 2 co-author steps—a researcher’s direct co-authors across all selected papers plus their co-authors—Yoshua Bengio reaches 570 people, or 22% of the 2,592-person connected group (67 direct co-authors at 1 step and 503 additional researchers at 2 steps), followed by Oriol Vinyals with 482 (19%: 90 at 1 step and 392 at 2 steps) and Allan Dafoe with 463 (18%: 60 at 1 step and 403 at 2 steps). Yoshua Bengio also has the shortest mean path to everyone else in the 2,592-person group, at 3.49 steps.
Stacked horizontal bars for 10 researchers reaching the most people within 2 co-author steps inside the 2,592-person connected group, split into direct co-authors at 1 step and additional researchers at 2 steps.
Source: People of Superintelligence analysis of OpenAlex (CC0) most-cited works as of 21 October 2026 and career records as of 22 October 2026. The table combines the leading ranks by 2-step reach and shortest-path bridging.
Data table
| Person | Reached within 2 steps | Share of main group | At 1 step (direct co-authors) | At 2 steps | Share of shortest paths | Mean steps in main group |
|---|---|---|---|---|---|---|
| Yoshua Bengio | 570 | 22% | 67 | 503 | 6.7% | 3.49 |
| Oriol Vinyals | 482 | 19% | 90 | 392 | 2.5% | 3.59 |
| Allan Dafoe | 463 | 18% | 60 | 403 | 1.6% | 3.70 |
| David Silver | 460 | 18% | 106 | 354 | 2.3% | 3.67 |
| Demis Hassabis | 448 | 17% | 101 | 347 | 2.1% | 3.66 |
| Geoffrey Irving | 445 | 17% | 68 | 377 | 0.9% | 3.72 |
| Christopher D. Manning | 426 | 16% | 58 | 368 | 2.5% | 3.77 |
| Geoffrey Hinton | 424 | 16% | 40 | 384 | 3.6% | 3.66 |
| Pushmeet Kohli | 406 | 16% | 64 | 342 | 3.5% | 3.69 |
| Koray Kavukcuoglu | 402 | 16% | 78 | 324 | 1.1% | 3.73 |
| Bernhard Schölkopf | 314 | 12% | 48 | 266 | 5.4% | 3.78 |
| Jure Leskovec | 295 | 11% | 27 | 268 | 2.4% | 3.81 |
| Sebastian Thrun | 171 | 7% | 28 | 143 | 2.7% | 4.07 |
Across all 3,355,345 pairs of other researchers in the 2,592-person connected group, shortest-path betweenness centrality measures the share of shortest paths that pass through each person. Yoshua Bengio leads at 6.7% of all shortest paths, followed by Bernhard Schölkopf at 5.4%, Geoffrey Hinton at 3.6%, and Pushmeet Kohli at 3.5%. The list below ranks the top 8 shortest-path bridges in the connected group.
Which organizations share the most co-authored papers
On the 4,647 dated co-authored papers through 2026, at least two tracked co-authors have a known publication-year employer on 4,461 papers (96%). Of those 4,461 papers, 3,640 (82%) link two or more parent organizations across their co-authors, and 821 (18%) have all placed co-authors inside a single parent organization.
A paper can fall in more than one group: across the 3,640 cross-organization co-authored papers, 2,545 (70%) connect two universities, 1,547 (43%) involve a research institute, nonprofit, standalone lab, or government body, 1,557 (43%) connect a university and a company, and 310 (9%) connect two companies. Across all 3,640 cross-organization papers, 1,496 (41%) link every pair of organizations on the paper without an author at both organizations in the publication year, and 2,144 (59%) include at least one pair with an author at both organizations in the publication year.
Stacked horizontal bars for 10 pairs of parent organizations sharing the most co-authored papers by publication-year employers, split between papers without an author at both organizations and papers with an author at both organizations in the publication year.
Source: People of Superintelligence analysis of OpenAlex (CC0) most-cited works as of 21 October 2026 and career records as of 22 October 2026. A paper counts once for each organization pair among its co-authors in the publication year.
Data table
| Organization pair | Kinds | Shared papers | Without an author at both | With an author at both in year | Tracked authors at either org |
|---|---|---|---|---|---|
| Allen Institute for AI (Ai2) and University of Washington | Nonprofit + University | 84 | 1 (1%) | 83 (99%) | 64 |
| Google and Stanford University | Company + University | 72 | 27 (38%) | 45 (63%) | 82 |
| Carnegie Mellon University and Stanford University | University + University | 56 | 23 (41%) | 33 (59%) | 72 |
| Google and University of Oxford | Company + University | 56 | 6 (11%) | 50 (89%) | 78 |
| Google and University of California, Berkeley | Company + University | 53 | 27 (51%) | 26 (49%) | 93 |
| Carnegie Mellon University and University of Washington | University + University | 46 | 9 (20%) | 37 (80%) | 50 |
| Massachusetts Institute of Technology and Stanford University | University + University | 45 | 25 (56%) | 20 (44%) | 77 |
| Google and University College London | Company + University | 45 | 6 (13%) | 39 (87%) | 84 |
| Google and Microsoft | Company + Company | 44 | 28 (64%) | 16 (36%) | 79 |
| Carnegie Mellon University and Google | Company + University | 44 | 26 (59%) | 18 (41%) | 62 |
| Carnegie Mellon University and University of California, Berkeley | University + University | 39 | 20 (51%) | 19 (49%) | 52 |
| Microsoft and University of Washington | Company + University | 38 | 14 (37%) | 24 (63%) | 50 |
| Carnegie Mellon University and Massachusetts Institute of Technology | University + University | 37 | 31 (84%) | 6 (16%) | 50 |
| Google and University of Toronto | Company + University | 35 | 20 (57%) | 15 (43%) | 62 |
| Stanford University and University of California, Berkeley | University + University | 35 | 16 (46%) | 19 (54%) | 62 |
| Allen Institute for AI (Ai2) and Carnegie Mellon University | Nonprofit + University | 35 | 7 (20%) | 28 (80%) | 51 |
| Massachusetts Institute of Technology and University of California, Berkeley | University + University | 34 | 19 (56%) | 15 (44%) | 51 |
| Google and University of Washington | Company + University | 33 | 22 (67%) | 11 (33%) | 66 |
| CNRS and Inria | Government + Government | 33 | 1 (3%) | 32 (97%) | 26 |
| Google and Meta | Company + Company | 31 | 24 (77%) | 7 (23%) | 57 |
| University of California, Berkeley and University of Washington | University + University | 31 | 10 (32%) | 21 (68%) | 37 |
| Stanford University and University of Toronto | University + University | 25 | 23 (92%) | 2 (8%) | 31 |
Counting all co-authored papers where distinct tracked authors held roles at each organization in the publication year, Allen Institute for AI (Ai2) and University of Washington lead all 11,054 organization pairs with 84 shared papers across 64 tracked authors, followed by Google and Stanford University with 72 (82 authors), Carnegie Mellon University and Stanford University (72 authors); and Google and University of Oxford (78 authors), 56 papers each, and Google and University of California, Berkeley with 53 (93 authors).
At Allen Institute for AI (Ai2) and University of Washington, 83 of the 84 shared papers (99%) have an author at both organizations in the publication year, and 1 (1%) has no author at both. Counting only papers without an author at both organizations in the publication year, Carnegie Mellon University and Massachusetts Institute of Technology lead all organization pairs with 31 shared papers (84% of their 37 total), ahead of Google and Microsoft with 28 (64% of 44), followed by Google and Stanford University; and Google and University of California, Berkeley with 27 each. Among pairs of companies, Google and Microsoft lead with 44 shared papers (28 without an author at both and 16 with an author at both in the publication year), followed by Google and Meta (31).
Across the 4,461 dated co-authored papers with at least two placed co-authors, Google appears on the most papers of any single parent organization with 573 (504 cross-organization papers across 302 partner organizations and 69 internal-only papers; 315 papers include two or more Google co-authors), followed by Stanford University with 429 (397 cross-organization across 289 partners, 32 internal-only) and Massachusetts Institute of Technology with 397 (341 cross-organization across 292 partners, 56 internal-only). On papers where two or more tracked co-authors worked at the same parent organization in the publication year, Google leads with 315 papers, ahead of Microsoft with 166. The list below ranks the 8 organizations appearing on the most co-authored papers.
Career data comes from the biographies on this site and from OpenAlex author records, as of 22 October 2026. People whose OpenAlex record may be incomplete or mixed are left out of citation figures.