Language and speech leads the latest AI doctorate cohort

How has the center of AI moved across doctorate cohorts? We estimate that Language and speech has the largest share among doctorates from 2020–2026: 29%. Language and speech gained the most share from the 1960s to 2020–2026.

Key findings

• Language and speech: 29% of the 2020–2026 cohort: 19 of 66 people.
• Language and speech gained the most share: 16.5 percentage points. Knowledge, logic and reasoning had the largest fall: 12.4 percentage points.
• Recorded early company starts rose from 8% for the 2000s to 21% for the 2010s.

Which areas gained share?

We compare 1,584 people shaping AI that we track. Cohorts use the recorded year of the earliest completed doctorate. Each decade needs at least 50 people with a dated doctorate and an area. Smaller cohorts contain another 24 people.

Current tags describe whole careers, so an area's share can reflect work done after the doctorate. Coverage of well-known people can vary across cohorts.

Lines compare career-area shares across 7 doctorate cohorts. Language and speech gained the most share: 16.5 percentage points. Knowledge, logic and reasoning had the largest fall: 12.4 percentage points. The table gives every area's count and share in every cohort.

Source: People of Superintelligence cited biographies and career records, as of 18 October 2026; OpenAlex (CC0). Areas from current profile tags.

Data table

Doctorate cohort	Area	People in area	Cohort size	Share
1960s	Computer vision	15	65	23.1%
1960s	Language and speech	8	65	12.3%
1960s	Robotics and autonomous systems	3	65	4.6%
1960s	Reinforcement learning, agents and planning	0	65	0%
1960s	Safety, alignment and evaluation	0	65	0%
1960s	Policy, ethics and society	0	65	0%
1960s	Theory, statistics and optimization	10	65	15.4%
1960s	Knowledge, logic and reasoning	11	65	16.9%
1960s	Science, health and biology	0	65	0%
1960s	Quantum computing	2	65	3.1%
1960s	Systems, hardware and security	6	65	9.2%
1960s	Human-computer interaction	1	65	1.5%
1960s	Cognitive science and neuroscience	7	65	10.8%
1960s	Business, products and public voice	0	65	0%
1960s	General machine learning	2	65	3.1%
1970s	Computer vision	22	140	15.7%
1970s	Language and speech	21	140	15%
1970s	Robotics and autonomous systems	3	140	2.1%
1970s	Reinforcement learning, agents and planning	6	140	4.3%
1970s	Safety, alignment and evaluation	2	140	1.4%
1970s	Policy, ethics and society	5	140	3.6%
1970s	Theory, statistics and optimization	20	140	14.3%
1970s	Knowledge, logic and reasoning	19	140	13.6%
1970s	Science, health and biology	6	140	4.3%
1970s	Quantum computing	5	140	3.6%
1970s	Systems, hardware and security	18	140	12.9%
1970s	Human-computer interaction	2	140	1.4%
1970s	Cognitive science and neuroscience	10	140	7.1%
1970s	Business, products and public voice	0	140	0%
1970s	General machine learning	1	140	0.7%
1980s	Computer vision	52	242	21.5%
1980s	Language and speech	36	242	14.9%
1980s	Robotics and autonomous systems	16	242	6.6%
1980s	Reinforcement learning, agents and planning	17	242	7%
1980s	Safety, alignment and evaluation	2	242	0.8%
1980s	Policy, ethics and society	3	242	1.2%
1980s	Theory, statistics and optimization	31	242	12.8%
1980s	Knowledge, logic and reasoning	26	242	10.7%
1980s	Science, health and biology	10	242	4.1%
1980s	Quantum computing	16	242	6.6%
1980s	Systems, hardware and security	22	242	9.1%
1980s	Human-computer interaction	4	242	1.7%
1980s	Cognitive science and neuroscience	5	242	2.1%
1980s	Business, products and public voice	0	242	0%
1980s	General machine learning	2	242	0.8%
1990s	Computer vision	90	386	23.3%
1990s	Language and speech	63	386	16.3%
1990s	Robotics and autonomous systems	28	386	7.3%
1990s	Reinforcement learning, agents and planning	20	386	5.2%
1990s	Safety, alignment and evaluation	5	386	1.3%
1990s	Policy, ethics and society	16	386	4.1%
1990s	Theory, statistics and optimization	55	386	14.2%
1990s	Knowledge, logic and reasoning	30	386	7.8%
1990s	Science, health and biology	14	386	3.6%
1990s	Quantum computing	12	386	3.1%
1990s	Systems, hardware and security	24	386	6.2%
1990s	Human-computer interaction	8	386	2.1%
1990s	Cognitive science and neuroscience	12	386	3.1%
1990s	Business, products and public voice	3	386	0.8%
1990s	General machine learning	6	386	1.6%
2000s	Computer vision	123	407	30.2%
2000s	Language and speech	48	407	11.8%
2000s	Robotics and autonomous systems	22	407	5.4%
2000s	Reinforcement learning, agents and planning	25	407	6.1%
2000s	Safety, alignment and evaluation	15	407	3.7%
2000s	Policy, ethics and society	14	407	3.4%
2000s	Theory, statistics and optimization	51	407	12.5%
2000s	Knowledge, logic and reasoning	13	407	3.2%
2000s	Science, health and biology	24	407	5.9%
2000s	Quantum computing	10	407	2.5%
2000s	Systems, hardware and security	33	407	8.1%
2000s	Human-computer interaction	10	407	2.5%
2000s	Cognitive science and neuroscience	9	407	2.2%
2000s	Business, products and public voice	5	407	1.2%
2000s	General machine learning	5	407	1.2%
2010s	Computer vision	70	278	25.2%
2010s	Language and speech	61	278	21.9%
2010s	Robotics and autonomous systems	21	278	7.6%
2010s	Reinforcement learning, agents and planning	23	278	8.3%
2010s	Safety, alignment and evaluation	10	278	3.6%
2010s	Policy, ethics and society	10	278	3.6%
2010s	Theory, statistics and optimization	23	278	8.3%
2010s	Knowledge, logic and reasoning	7	278	2.5%
2010s	Science, health and biology	26	278	9.4%
2010s	Quantum computing	5	278	1.8%
2010s	Systems, hardware and security	14	278	5%
2010s	Human-computer interaction	3	278	1.1%
2010s	Cognitive science and neuroscience	0	278	0%
2010s	Business, products and public voice	0	278	0%
2010s	General machine learning	5	278	1.8%
2020–2026	Computer vision	10	66	15.2%
2020–2026	Language and speech	19	66	28.8%
2020–2026	Robotics and autonomous systems	4	66	6.1%
2020–2026	Reinforcement learning, agents and planning	2	66	3%
2020–2026	Safety, alignment and evaluation	9	66	13.6%
2020–2026	Policy, ethics and society	5	66	7.6%
2020–2026	Theory, statistics and optimization	4	66	6.1%
2020–2026	Knowledge, logic and reasoning	3	66	4.5%
2020–2026	Science, health and biology	3	66	4.5%
2020–2026	Quantum computing	0	66	0%
2020–2026	Systems, hardware and security	4	66	6.1%
2020–2026	Human-computer interaction	2	66	3%
2020–2026	Cognitive science and neuroscience	0	66	0%
2020–2026	Business, products and public voice	1	66	1.5%
2020–2026	General machine learning	0	66	0%

Language and speech rose from 12.3% in the 1960s to 28.8% in 2020–2026.

What leads the latest cohort?

The 2020–2026 cohort has 66 people. This incomplete decade's shares can change as more people graduate. Computer vision has the next largest share: 15%, or 10 people.

Language and speech: 29% of the 2020–2026 cohort. Bars show all 12 areas represented among 66 people.

Source: People of Superintelligence cited biographies and career records, as of 18 October 2026; OpenAlex (CC0). Areas from current profile tags.

Data table

Area	People	Share	Share in the 1960s	Change (percentage points)
Language and speech	19	28.8%	12.3%	+16.5
Computer vision	10	15.2%	23.1%	-7.9
Safety, alignment and evaluation	9	13.6%	0%	+13.6
Policy, ethics and society	5	7.6%	0%	+7.6
Robotics and autonomous systems	4	6.1%	4.6%	+1.4
Systems, hardware and security	4	6.1%	9.2%	-3.2
Theory, statistics and optimization	4	6.1%	15.4%	-9.3
Knowledge, logic and reasoning	3	4.5%	16.9%	-12.4
Science, health and biology	3	4.5%	0%	+4.5
Reinforcement learning, agents and planning	2	3%	0%	+3
Human-computer interaction	2	3%	1.5%	+1.5
Business, products and public voice	1	1.5%	0%	+1.5
Cognitive science and neuroscience	0	0%	10.8%	-10.8
General machine learning	0	0%	3.1%	-3.1
Quantum computing	0	0%	3.1%	-3.1

Do later cohorts enter companies sooner?

We find a company role within 5 years after the doctorate year for 58 of 278 people from the 2010s cohort (21%).

Recorded early company starts rose from 8% for the 2000s to 21% for the 2010s. Bars give the share and count for each of 6 complete doctorate decades. Area counts and shares are in the table.

Source: People of Superintelligence cited biographies and career records, as of 18 October 2026; OpenAlex (CC0). Areas from current profile tags. Company roles come from cited biographies. Corporate labs count with their parent company.

Data table

Doctorate cohort	Area	People	Company start within 5 years	Share
1960s	All areas	65	0	0%
1970s	All areas	140	7	5%
1970s	Computer vision	22	1	4.5%
1970s	Language and speech	21	1	4.8%
1970s	Theory, statistics and optimization	20	0	0%
1980s	All areas	242	13	5%
1980s	Computer vision	52	2	3.8%
1980s	Language and speech	36	1	2.8%
1980s	Theory, statistics and optimization	31	0	0%
1980s	Knowledge, logic and reasoning	26	4	15.4%
1980s	Systems, hardware and security	22	2	9.1%
1990s	All areas	386	29	8%
1990s	Computer vision	90	8	8.9%
1990s	Language and speech	63	10	15.9%
1990s	Robotics and autonomous systems	28	1	3.6%
1990s	Reinforcement learning, agents and planning	20	1	5%
1990s	Theory, statistics and optimization	55	2	3.6%
1990s	Knowledge, logic and reasoning	30	1	3.3%
1990s	Systems, hardware and security	24	3	12.5%
2000s	All areas	407	34	8%
2000s	Computer vision	123	17	13.8%
2000s	Language and speech	48	7	14.6%
2000s	Robotics and autonomous systems	22	0	0%
2000s	Reinforcement learning, agents and planning	25	0	0%
2000s	Theory, statistics and optimization	51	1	2%
2000s	Science, health and biology	24	0	0%
2000s	Systems, hardware and security	33	1	3%
2010s	All areas	278	58	21%
2010s	Computer vision	70	11	15.7%
2010s	Language and speech	61	24	39.3%
2010s	Robotics and autonomous systems	21	3	14.3%
2010s	Reinforcement learning, agents and planning	23	6	26.1%
2010s	Theory, statistics and optimization	23	6	26.1%
2010s	Science, health and biology	26	3	11.5%

Areas with at least 20 people in both cohorts; cells show shares and people:

We count the first company role starting after the doctorate year. Roles include founding, visits, internships and jobs held alongside academic work.

Year-only dates limit the count to roles starting in a later year. Missing job histories can lower these shares. Coverage and the mix of career areas can affect comparisons.

Examples: Arindam Mitra: doctorate 2019, Microsoft Research role 2024. Hyung Won Chung: doctorate 2019, OpenAI role 2023.