The NSF still writes the biggest check
The National Science Foundation accounted for 34.9% of external funds to US CS departments in 2021 — the largest single source, and a shrinking one.
Universities are training the AI PhDs. Industry is taking them.
Chapter 5 of the AI Index 2023 asks who is learning AI and who is left to teach it. The postsecondary numbers come from the CRA Taulbee Survey and run through 2021; the K–12 numbers come from Code.org and from a UNESCO survey of 193 member states. The shape of it: a record share of computer science doctorates now specialize in AI, two-thirds of them go straight to industry, and the faculty hiring line has been shrinking for a decade.
19.1% of new CS PhD students in North America specializing in AI in 2021 (14.9% in 2020)
65.4% of new AI PhDs who took industry jobs in 2021 (28.2% went to academia)
68.6% of new North American CS PhD graduates who were international in 2021 (45.8% in 2010)
710new CS, CE, and information faculty hires in North America in 2021 (733 in 2012)
181,040AP computer science exams taken in the US in 2021, up 1.0% year over year
11countries that had endorsed and implemented a K–12 AI curriculum as of 2021
5.1 — The undergraduate boom stops dead at the doctorate
New CS bachelor’s graduates in North America have almost quadrupled since 2012. Master’s degrees doubled and then flattened. PhDs never moved at all — there were fewer of them in 2021 than in 2012.
In 2021 North American universities produced 33,059 new CS bachelor’s graduates, nearly four times the 2012 figure. Master’s graduates roughly doubled over the same stretch to 15,068, but the growth stopped in 2018: the count has drifted down from 15,532 that year. At the doctoral level there was no boom to speak of. North America graduated 1,893 new CS PhDs in 2021 — fewer than the 1,997 of 2020, and fewer than the 1,929 of 2012.
Graduate school runs on international students
- 16.3% of new CS bachelor’s graduates in North America were international students in 2021, a rise of 9.5 percentage points since 2012.
- At master’s level the majority are international: 65.2% in 2021. That share has actually been sliding since 2016, and the total number of master’s graduates has been flat since 2018.
- At PhD level the international share is still climbing — 45.8% in 2010, 68.6% in 2021.
- Then it inverts. Only 13.2% of new CS, CE, and information tenure-track faculty hires in North America in 2021 were international.
AI is quietly taking over the doctorate
19.1% of new CS PhD students in North America specialized in AI in 2021 — up 4.2 percentage points in a single year from 14.9% in 2020, and up 8.6 points since 2012. In 2010 the figure was 10.2%. Close to one in five computer science doctorates is now an AI doctorate, inside a cohort whose total size has not grown in a decade.
The people who would teach AI are not being hired
Total CS faculty headcount keeps creeping up. New hiring has been shrinking for a decade, and its tenure-track core has fallen by nearly a quarter since 2019.
North America counted 8,149 CS, CE, and information faculty in 2021 — 2.2% more than the year before and 32.8% more than in 2011. In the United States alone the number was 6,789, up 2.0% in a year and 39.0% over the decade. Hiring tells the opposite story. There were 710 new faculty hires in North America in 2021 against 733 in 2012, and tenure-track hires peaked at 422 in 2019 before dropping to 324 in 2021.
Where the new faculty come from
- 40% of new CS, CE, and information faculty hires in 2021 came straight out of a PhD, 34% from another academic position, and 15% from a postdoc.
- Just 11% came from industry — down from 13% in both 2018 and 2019, at exactly the moment industry was absorbing 65.4% of new AI PhDs.
- 89.3% of new faculty positions were filled in 2021, up from 82.7% in 2011. Departments are not failing to fill posts; they are opening fewer of them.
- When a post did stay open, 53% of the time the reason was that the offer had been turned down. 22% of the time hiring was still in progress, and 14% of the time no candidate met the department’s hiring goals.
- 36.3% of faculty losses in 2021 were people moving to an academic position elsewhere. 15.2% left for nonacademic positions — almost exactly the 15.9% of a decade earlier.
Pay is rising, slowly, before inflation
The median nine-month salary of a full professor of computer science in the United States reached $176,010 in 2021 — 3.2% above 2020 and 12.8% above 2015. Associate professors took a median of $127,470 and assistant professors $114,070. The chapter notes that none of these figures are adjusted for inflation, which in 2021 was doing most of the work in the other direction.
Who actually pays for computer science research in America
The CRA tracks the external funding sources of US CS departments. Federal money still dominates, but its composition is shifting — and the gap between private and public departments keeps widening.
Defense is the second-largest funder
Defense agencies outside DARPA — the Army Research Office, the Office of Naval Research, and the Air Force Research Laboratory — supplied 20.3% of external funds in 2021.
Industry money is growing
Industrial sources supplied 12.1% of external funds to US CS departments in 2021, ahead of DARPA (8.8%) and the NIH (6.8%).
Private departments now outspend public ones
In 2021 the median US private CS department spent $9.71 million of external research money. The median public department spent $5.69 million.
5.2 — Whether you can take computer science depends on your state
27 states required every high school to offer a computer science course in 2022. The share of public high schools actually teaching it runs from 98% in Maryland to 21% in Minnesota.
Code.org’s 2022 data shows 27 states requiring that all high schools offer a computer science course. The requirement does not map neatly onto what gets taught. Maryland leads on the share of public high schools teaching CS at 98%, followed by South Carolina (93%) and Arkansas (92%). At the other end sit Minnesota at 21%, Louisiana at 32%, and Arizona and Montana at 36% each. California — the largest state, and the one taking the most AP CS exams in absolute terms — is at 40%.
AP exams: ninefold growth, then a pause
American students took 181,040 AP computer science exams in 2021, a 1.0% increase on the previous year after several years of large gains. The AI Index attributes the leveling off to the pandemic. Measured against 2007, the number of AP CS exams has grown more than ninefold. The count covers both AP Computer Science A and AP Computer Science Principles, the latter first offered in 2017.
Volume and density are two different maps
California recorded the most AP CS exams in 2021 with 31,189, ahead of Texas (17,307), Florida (14,864), New York (13,304), and New Jersey (9,391). Normalize by population and the ranking rearranges itself: Maryland leads with 124.1 exams per 100,000 inhabitants, then New Jersey (101.3), Connecticut (89.7), California (79.7), and Massachusetts (78.0). At the bottom of the raw count, Alaska took 100 exams for the whole year and Montana 42.
What a national AI curriculum actually contains
UNESCO surveyed representatives of 193 member states plus over 10,000 private- and third-sector actors on the state of government-endorsed K–12 AI curricula. Five questions from what came back.
How many countries have an AI curriculum in place?
Eleven had endorsed and implemented a K–12 AI curriculum as of 2021: Armenia, Austria, Belgium, China, India, Kuwait, Portugal, Qatar, Serbia, South Korea, and the United Arab Emirates. Five more were developing one — Bulgaria, Germany, Jordan, Saudi Arabia, and Serbia, which appears on both lists because it is developing further curricula on top of what it has already implemented.
Which school levels do those curricula cover?
Coverage thins out at the youngest ages. China, Portugal, Qatar, and the United Arab Emirates run endorsed and implemented AI curricula across primary, middle, and high school. Armenia, India, and Serbia cover middle and high school. Austria, Belgium, and South Korea reach high school only. Kuwait is the one country in the implemented group that covers primary and middle school but not high school. Among the countries still developing, Bulgaria, Germany, and Saudi Arabia are each building for all three levels at once.
What do students actually spend the time on?
Algorithms and programming take the largest slice of time in the K–12 AI curricula UNESCO profiled, at 18%. AI technologies follow at 14%, then data literacy and the application of AI to other domains at 12% each, contextual problem-solving at 11%, developing AI technologies at 9%, ethics of AI at 7%, social implications of AI at 5%, and AI techniques at 2%. A further 10% of the time is unspecified.
So is this ethics education or engineering education?
Overwhelmingly engineering. Ethics of AI (7%) and social implications of AI (5%) together take 12% of curriculum time — less than algorithms and programming take on their own (18%). The three categories UNESCO groups under understanding, using, and developing AI — AI technologies (14%), developing AI technologies (9%), and AI techniques (2%) — account for 25% between them.
What does one of these curricula look like up close?
The UNESCO report documents the Austrian Data Science and Artificial Intelligence curriculum in detail. It opens with digital basics — using an operating system to store and print files, designing presentations, spreadsheets and word processing — then moves to types of digital media, social issues in digital media, and safe use. High school students take on programming languages, algorithms, and simulations, and learn the basic principles of data literacy: collecting data, structuring a spreadsheet, running analyses and visualizations, and applying criteria to judge whether a data source or a piece of digital content is credible. Students are expected to know about careers in ICT and AI, to create digital media, and to understand the cloud and how computers are networked. It closes on civics: understanding the ethical dilemmas attached to these technologies, becoming active participants in social discourse about them, and using technology to make public statements.
The chapter in five lines
The headline findings of Chapter 5 · Education, in the report’s own terms.
The proportion of new computer science PhD graduates from U.S. universities who specialized in AI jumped to 19.1% in 2021, from 14.9% in 2020 and 10.2% in 2010.
In 2011, roughly the same proportion of new AI PhD graduates took jobs in industry (40.9%) as opposed to academia (41.6%). In 2021, 65.4% of AI PhDs took jobs in industry, more than double the 28.2% who took jobs in academia.
In the last decade, the total number of new North American computer science, computer engineering, and information faculty hires has decreased: There were 710 total hires in 2021 compared to 733 in 2012.
In 2021, the median expenditure for private universities was $9.7 million, compared to $5.7 million for public universities.
In 2021, a total of 181,040 AP computer science exams were taken by American students, a 1.0% increase from the previous year. Since 2007, the number of AP computer science exams has increased ninefold.
Read Chapter 5 in full
Chapter 5 (sections 5.1 and 5.2) with every figure, state map, and citation is free from Stanford HAI. Or head back to the report highlights and the eight-chapter overview.
Open the AI Index 2023 →