Timnit Gebru: Her Google Exit and Critique of Large AI

Timnit Gebru holds a PhD in computer vision from Stanford and founded the Distributed Artificial Intelligence Research Institute (DAIR). DAIR's stated focus is the effects of AI on marginalized groups. She left Google in December 2020 after a dispute over a paper on the risks of large language models (LLMs). Google says she resigned, and she says she was fired.

What Happened to Timnit Gebru at Google

Timnit Gebru left Google in December 2020 after a dispute over a research paper. Google says she resigned and that it accepted the resignation. Gebru says she was fired. Her best-known paper is the Stochastic Parrots paper; she also co-authored the Gender Shades study and founded DAIR.

The paper's title asks whether language models can be too big. For the wider debate, see our guide to who is fighting AI. Before Google, she had worked at Apple and Microsoft.

Her Path to Google

Gebru was born in Addis Ababa, Ethiopia, in 1983. She left Ethiopia at age 15 during the Eritrean-Ethiopian War. She received political asylum in the United States after an initial visa denial.

She was accepted at Stanford University in 2001 and earned a BS and MS in electrical engineering there. She worked at Apple from 2004 to 2013, starting as an intern. Her Apple work included audio circuitry and signal processing for the first iPad.

She completed a PhD in computer vision at Stanford in 2017, with Fei-Fei Li as her doctoral advisor. In summer 2017 she joined Microsoft as a postdoctoral researcher in its FATE (Fairness, Accountability, Transparency and Ethics) lab. In 2018 she joined Google as technical co-lead of its Ethical Artificial Intelligence team, which she co-led with Margaret Mitchell.

She also co-founded Black in AI with Rediet Abebe. The group aims to increase the presence, visibility and well-being of Black professionals in AI. For the wider question of bias in AI systems, see our explainer on whether AI is racist.

Two Accounts of the Exit

Google's account is that management asked Gebru to withdraw the Stochastic Parrots paper or remove the Google authors' names from it. Google says her email asking for the reviewers' identities and feedback, and referring to "a last date," was read as an immediate resignation. Google then accepted that resignation.

Gebru's account differs. She says she did not formally offer to resign. She says she threatened to resign after an appropriate period if her conditions were not met. Google told her it could not meet her demands and accepted her resignation immediately. She says she was fired.

Her internal email, which Platformer published, says her manager's manager told her "it has been decided" that the paper had to be retracted by 27 November. She wrote that her detailed written response to the feedback was ignored. Platformer also reported that Google found her email "inconsistent with the expectations of a Google manager."

Jeff Dean, Google's head of AI, wrote an internal email that Platformer published in full on 3 December 2020. In it he said the paper "ignored too much relevant research." He also wrote that "we require two weeks for this sort of review." The cited sources don't record her reply on the review timing.

Nine members of the US Congress sent Google a letter asking it to clarify the circumstances of her departure. Her co-lead, Margaret Mitchell, publicly criticized Google and was later terminated. Wikipedia says that was allegedly for writing scripts that searched internal servers for evidence about Gebru's treatment.

Sundar Pichai opened a months-long internal investigation. Afterward, Dean announced changes to how employees leave the company and how papers on sensitive topics are reviewed. The cited sources don't publish the investigation's findings.

In November 2021, the Nathan Cummings Foundation, with Open MIC, filed a shareholder proposal calling for a racial equity audit at Alphabet. In December 2021, Reuters reported that California's Department of Fair Employment and Housing was investigating Google's treatment of Black women. Neither side's account of the exit has been adjudicated in public.

The Stochastic Parrots Paper

The paper's full title is "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" Its named authors are Emily M. Bender, Timnit Gebru, Angelina McMillan-Major and Margaret Mitchell. The Association for Computing Machinery (ACM) published it at pages 610 to 623. Of the four authors, Bender was the only one who was not a Google employee at the time.

Wikipedia's citation is dated March 2021, but its body text places the paper in 2020. This explainer uses only the March 2021 citation date. The paper argues that very large language models carry several risks.

Those risks include environmental footprint, financial cost and inscrutability. The paper also lists prejudice against groups, an inability to actually understand language, and use to spread disinformation. Google's main objection, covered in the critics section below, was that the paper ignored too much relevant research.

Earlier Work: Gender Shades

Gebru's earlier Gender Shades study, co-authored with Joy Buolamwini in 2018, found that one commercial facial-recognition system was 35% less likely to recognize Black women than White men. For the wider debate, see our guide to facial recognition ethics. See also our profile of Joy Buolamwini, her co-author on that study.

DAIR, the Institute She Founded

Gebru founded the Distributed Artificial Intelligence Research Institute (DAIR). She announced fundraising for it in June 2021, and it launched on 2 December 2021. DAIR describes itself as "an independent organization conducting community-rooted research." It also calls itself "a globally distributed group of academics, activists, and engineers."

DAIR's stated focus is the effects of AI on marginalized groups, with particular attention to Africa and to African immigrants in the United States. An early DAIR project analyzed satellite imagery of South African townships. The homepage names three framings: cutting through AI hype, grounding research in community, and imagining alternative futures.

The site says its work centers "the people most impacted by technology" and is "rooted in community needs." That is DAIR's own wording. The cited sources don't include an outside assessment of its work.

Funding and Tax Filings

DAIR's support page says it "was founded with the support of grants from the MacArthur Foundation, Ford Foundation, Open Society Foundations." It also names the Rockefeller Foundation and a gift from the Kapor Center. The same page says DAIR "is supported by more than a dozen philanthropic foundations and nearly 100 individual donors."

The support page says DAIR is a 501(c)(3) organization. Its IRS registration lists the name Distributed Artificial Intelligence Research Institute, an employer identification number (EIN) of 99-2299327, and an address at 1440 Broadway Ste 200, Oakland, CA 94612. The IRS ruling date is 1 October 2024.

ProPublica's record for DAIR shows revenue of $7,130,636 and assets of $5,661,474 for tax period 2025-06. ProPublica lists the underlying Form 990 filing without extracted expense or officer pay figures, so those numbers aren't available there.

Two funding questions remain open. DAIR's support page doesn't say whether it accepts corporate or AI-industry money. Its headcount isn't published either, and the site describes DAIR only as "a globally distributed group."

Recognition and Current Role

Gebru received the 2019 VentureBeat AI Innovations Award for AI for Good, shared with Joy Buolamwini and Inioluwa Deborah Raji. She was on Fortune's World's 50 Greatest Leaders list and on Nature's ten people who shaped science in 2021. TIME named her to its 100 most influential people in 2022.

In 2023 she received the Carnegie Corporation Great Immigrants Award and was named to the BBC 100 Women. Wikipedia states she received the Right Livelihood Award in September 2026. The cited sources include no second source for that award.

Her current title and institutional affiliation beyond founding DAIR aren't confirmed in the sources cited here. Founder of DAIR is the safest description of her current role.

Why She Argues With the AI Doom Camp

Her published work documents specific harms, including the recognition gap found in Gender Shades and the risks listed in the Stochastic Parrots paper. For the doom camp's own case, see our AI doomers explainer.

Her disagreement concerns artificial general intelligence (AGI). She has criticized research into AGI as rooted in eugenics. She argues that attempting to build AGI is inherently unsafe.

With Émile P. Torres, Gebru coined the term TESCREAL for a bundle of ideologies: transhumanism, extropianism, singularitarianism, cosmism, rationalism, effective altruism and longtermism. She has compared proponents of that bundle to "the eugenicists of the 20th century." Those movements reject that characterization.

The TESCREAL argument appeared with Torres in the journal First Monday in 2024. DAIR's stated focus is on marginalized groups, on Africa, and on African immigrants in the United States.

What Her Critics Say

Google's main objection comes from Dean's internal email. Dean's email said the paper "ignored too much relevant research." He wrote that it discussed the environmental impact of large models but "disregarded subsequent research." He also wrote that it raised bias concerns without accounting for recent mitigation research.

Google's position throughout has been that she resigned rather than being fired. Dean's email stated, "We accept and respect her decision to resign from Google." Google's objection on the review timing rests on the two-week requirement quoted above.

A single-author preprint by Michael Lissack makes a related case. It was submitted to arXiv on 11 January 2021, and its title begins "The Slodderwetenschap (Sloppy Science) of Stochastic Parrots." It argues the paper fails to acknowledge that it is an advocacy piece rather than research.

Lissack also argues the paper does not state its underlying presuppositions. He says it lists potential harms while treating benefits as if they do not matter. The preprint is not a peer-reviewed rebuttal.

The TESCREAL term draws a separate objection. Critics say it lumps seven distinct movements together, and the comparison to 20th-century eugenicists is one those movements reject. The term is presented here as Gebru's argument, not as settled fact.

Nine members of Congress asked Google to clarify her departure. Sources disagree on how many people signed letters backing her, and the cited sources don't give a single total.

Both sides have published their accounts, and neither has been tested in public. To read the emails behind the dispute yourself, start with Platformer's published copy of Dean's email and Gebru's own email.

Sources

• Timnit Gebru, Wikipedia
• Platformer, internal emails on Gebru's departure from Google
• Distributed Artificial Intelligence Research Institute (DAIR), homepage
• DAIR support page
• ProPublica Nonprofit Explorer, DAIR record
• Michael Lissack, arXiv:2101.10098