714 attendees, down two years running
The 2023 workshop drew 714 people — nearly eight times the 89 who attended in 2010, but lower than each of the two preceding years.
Chapter 8 is the one place in the report where the subject is people rather than models: who studies computer science, who teaches it, who shows up at NeurIPS, who sits the AP exam. The report’s own framing is that AI developers do not look like AI users, and that the gap can carry bias into the systems. Read across the 2022 data, ethnicity is moving and gender is not.
The Computing Research Association’s annual Taulbee Survey tracks new CS graduates and faculty across the United States and Canada. Four different measures, taken a decade apart, all land in the same narrow band.
Women were 22.2% of new CS bachelor’s graduates in 2022, the top of a decade-long climb; another 0.1% identified as nonbinary or other. Master’s programs are the least lopsided of the three degrees at 26.3% women — but that is up from 24.6% in 2011, which is 1.7 percentage points in eleven years. At PhD level the share slipped to 22.1% in 2022, and the report is explicit that the long-term trend is unchanged. In other words, the pipeline does not narrow for women as it goes up; it was already narrow at the bottom.
As of 2022 faculty in CS, computer engineering, and information fields were 75.6% male, 24.3% female, and 0.1% nonbinary — up since 2011, but by a small amount. New hires are the exception worth watching: 28.0% of new faculty hires in 2022 were women, against 71.7% men, a share the report notes sits well above the proportion of new female PhDs. Whatever is happening in hiring is running ahead of what the doctoral pipeline is producing.
For the second year running the CRA asked departments how many students received disability accommodations. The answers were low and flat: 4.1% of bachelor’s students, 1.5% of master’s students, and 1.1% of PhD students in 2022, against 4.1%, 0.8%, and 1.0% the year before. The chapter says only that the proportion has remained consistent year over year — it does not claim the figure measures how many students have a disability.
This is the part of the chapter that actually moved. Since 2011 the Asian share of CS bachelor’s graduates has risen 19.8 percentage points and the Hispanic share 5.2 points; in raw numbers, Hispanic graduates grew 4.7 times over the decade and Black or African American graduates 2.5 times.
Among resident CS bachelor’s graduates in 2022, white students were 44.6% and Asian students 35.7% — the two groups are now nine points apart, where a decade ago the gap was far wider. Hispanic students of any race were 11.0%, multiracial students 4.4%, and Black or African American students 4.1%. The head counts behind those shares: 10,970 white, 8,795 Asian, 2,708 Hispanic, 1,072 multiracial, 1,004 Black or African American, 33 American Indian or Alaska Native, and 28 Native Hawaiian or Pacific Islander.
Every ethnicity figure above covers domestic or resident students and faculty only. The CRA reports how many nonresident aliens there are at each level, but not their ethnicity, so a large share of the people in these classrooms is simply outside the chart. The survey is also partial: of the 297 PhD-granting departments targeted, 182 responded, a 61% response rate. And the AI Index warns against per capita comparisons between this North American data and the European figures that follow, because the European numbers come from national statistical offices and cover far more ground.
New to this year’s chapter is data from Informatics Europe covering informatics, CS, computer engineering, and IT graduates. The finding is unanimous and it is one of the chapter’s three headline points.
At bachelor’s level men dominate in the majority of surveyed nations, and the gap has narrowed only slightly since 2011. France (14.8%), the United Kingdom (17.8%), and Germany (21.5%) report some of the lowest female shares on the list — three of the largest computing economies in Europe sitting at or near the bottom. Bulgaria has the highest share of female graduates at 35.2%.
Master’s degrees show the same shape with a wider spread. Estonia (42.0%), Romania (41.9%), and Bulgaria (40.4%) come closest to parity; Belgium (13.7%), Italy (14.1%), and Switzerland (15.8%) are furthest from it. The ordering is worth sitting with: the three countries closest to parity all sit on the eastern side of the continent, the three furthest from it in the west, and the two ends are about threefold apart.
PhD graduates are predominantly male in every surveyed country without exception. The United Kingdom, Germany, and Switzerland have narrowed their gaps over the last decade, with women taking a growing share; Finland and Spain have seen theirs widen slightly. Read those movements carefully, though — where the cohorts are small, a single year swings the percentage. Bulgaria produced 24 PhDs in total in 2022, Estonia 26, and Latvia 12. One label note: a year here means the year an academic year ended, so 2022 covers the 2021/2022 intake.
Women in Machine Learning, founded in 2006, runs an annual technical workshop at NeurIPS. It is the chapter’s only window onto the conference side of the field — four measures, each with the caveat that goes with it.
The 2023 workshop drew 714 people — nearly eight times the 89 who attended in 2010, but lower than each of the two preceding years.
As a share of total NeurIPS attendance, the 2023 WiML workshop accounted for 4.4% of attendees.
Of participants who answered, 56.4% live in North America, 21.8% in Europe, 11.4% in Asia, and 8.9% in Africa.
Among participants who answered the survey, 84.2% identified as female, 10.0% as male, and 3.2% as nonbinary.
Code.org’s data on AP computer science is the chapter’s view of the school-age end of the pipeline. Both exams — Computer Science A and Computer Science Principles — are counted together.
In 2022, male students took 68.9% of AP CS exams, female students 30.5%, and students identifying as neither 0.7%. The female share was 16.8% in 2007, so it has nearly doubled in fifteen years — the fastest-moving gender number anywhere in this chapter, and still the minority position by better than two to one.
The states with the highest female share of AP CS test-takers in 2022 were Mississippi (41%), Alabama (37%), and Washington, D.C. (37%). California, Texas, and Washington — the states usually named for CS and AI activity — each land at about 31%, respectable but well behind. At the other end sit Kansas (16%), North Dakota (18%), and Iowa (18%). Whatever produces a balanced exam cohort, it is not proximity to the industry.
White students remain the largest group of AP CS test-takers at 38.2%, followed by Asian students at 27.8% and Hispanic/Latino/Latina students at 17.6%; Black or African American students were 6.7% and students of two or more races 4.6%. Native American/Alaskan students were 0.6% and Native Hawaiian/Pacific Islander students 0.15%. In exam counts that is 77,070 white, 56,098 Asian, 35,528 Hispanic/Latino/Latina, 13,577 Black/African American, 9,293 two or more races, 1,285 Native American/Alaskan, and 295 Native Hawaiian/Pacific Islander. Participation from Asian, Hispanic/Latino/Latina, and Black/African American students has risen year over year — the same direction as the postsecondary numbers, which is the chapter’s third headline finding.
The three chapter highlights, and the admission the overview closes on.
US and Canadian bachelor’s, master’s, and PhD CS students continue to grow more ethnically diverse. Since 2011 the proportion of Asian CS bachelor’s graduates has increased by 19.8 percentage points, and the proportion of Hispanic CS bachelor’s graduates by 5.2 percentage points.
Every surveyed European country reported more male than female graduates in bachelor’s, master’s, and PhD programs for informatics, CS, CE, and IT. While the gender gaps have narrowed in most countries over the last decade, the rate of this narrowing has been slow.
The proportion of AP CS exams taken by female students rose from 16.8% in 2007 to 30.5% in 2022.
Demographic data regarding AI trends, particularly in areas such as sexual orientation, remains scarce. The AI Index urges other stakeholders in the AI domain to intensify their endeavors to track diversity trends associated with AI.
Chapter 8 (sections 8.1–8.3) with every figure, footnote, and citation is free from Stanford HAI. Or head back to the report highlights and nine-chapter overview.
Open the AI Index 2024 →