Timnit Gebru
Last reviewed
Sources
20 citations
Review status
Source-backed
Revision
v2 · 2,401 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
20 citations
Review status
Source-backed
Revision
v2 · 2,401 words
Add missing citations, update stale details, or suggest a clearer explanation.
Timnit Gebru is an Ethiopian-born computer scientist and a leading researcher in AI ethics, best known for co-authoring the 2018 "Gender Shades" study on bias in facial recognition, co-leading Google's Ethical AI team, and her contested December 2020 departure from Google during a dispute over the "On the Dangers of Stochastic Parrots" paper. She is the founder and executive director of the Distributed Artificial Intelligence Research Institute (DAIR), which she launched on December 2, 2021, and a co-founder of the advocacy group Black in AI. Her departure from Google drew an open letter signed by roughly 2,700 Google employees and more than 4,300 academics and members of civil society [1][2][3]. Accounts of how that exit happened differ: Gebru has said she was fired, while Google has described the outcome as accepting a resignation [2][3]. Across her work she is widely regarded as one of the most prominent critics of AI bias, large language model hype, and overstated claims about artificial general intelligence [4][5].
Gebru was born around 1982 or 1983 in Addis Ababa, Ethiopia. Both of her parents were from Eritrea. Her father, an electrical engineer who held a doctorate, died when she was about five years old, and she was raised mainly by her mother, an economist [6]. At around the age of 15, during the Eritrean-Ethiopian War, she left Ethiopia after members of her family were deported to Eritrea and conscripted. She has said she was initially denied a United States visa and spent a period in Ireland before receiving political asylum in the United States [6].
Gebru studied at Stanford University, where she earned a bachelor's degree and a master's degree in electrical engineering and then a doctorate in computer vision in 2017. Her doctoral advisor was Fei-Fei Li [6]. Her dissertation work, which she described as visual computational sociology, combined deep learning with publicly available Google Street View imagery to estimate demographic and socioeconomic characteristics of neighborhoods from features such as the makes and models of parked cars [6][7]. Earlier in her career she worked as an engineer at Apple, where she contributed to signal-processing work on hardware including early iPad circuitry [6].
In 2017 Gebru joined Microsoft Research as a postdoctoral researcher in its Fairness, Accountability, Transparency, and Ethics in AI group [6]. During this period she and Joy Buolamwini published "Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification," presented at the Conference on Fairness, Accountability, and Transparency in 2018. The study audited three commercial facial-analysis products, from IBM, Microsoft, and the Chinese firm Face++, and reported that error rates for classifying gender were never worse than about 0.8 percent for lighter-skinned men but reached 20.8 percent to 34.7 percent for darker-skinned women, with the largest single gap, about 34.4 percentage points, found in IBM's system [8][9]. For the darkest-skinned women in the sample, two systems misclassified gender at rates of 46.5 percent and 46.8 percent [9]. The authors used the Fitzpatrick scale, a dermatological classification of skin tone, to organize their analysis, and they assembled a new benchmark of 1,270 images, the Pilot Parliaments Benchmark, balanced across skin type and gender [8][9]. The paper is frequently cited in later debates over the deployment of facial recognition, with more than 3,000 citations, and several companies announced changes or limits to their facial-analysis offerings in the years that followed [8][9].
Gebru also led work on documentation standards for machine learning. In "Datasheets for Datasets," first posted in 2018 and later published in Communications of the ACM in 2021, she and her co-authors proposed that every dataset be accompanied by a standardized record of its motivation, composition, collection process, and recommended uses, by analogy with the datasheets that describe electronic components [10][11]. The published version organized this record into 57 questions across seven categories, covering motivation, composition, collection process, preprocessing and labeling, uses, distribution, and maintenance [11]. The proposal became one reference point for broader efforts to improve transparency and accountability in how datasets and models are reported.
Gebru co-founded Black in AI in 2016 with the computer scientist Rediet Abebe. The effort first gathered as a workshop affiliated with the Conference on Neural Information Processing Systems, held in December 2017, and grew into a membership organization that supports Black researchers and practitioners in artificial intelligence and advocates for broader participation in the field [12]. Gebru has said the idea took shape partly after she attended a major machine learning conference and observed how few Black researchers were present among thousands of attendees [6][12].
Gebru joined Google in 2018 and became a co-lead of its Ethical AI research team alongside Margaret Mitchell [6][2]. In late 2020 a conflict developed over a paper she had co-authored, titled "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" Google managers asked her either to withdraw the paper or to remove the names of Google-affiliated authors from it [1][2]. Reporting indicates that Gebru asked for details about who had reviewed the paper internally and on what grounds, and that she said she would discuss a departure date if the company would not provide that information [2][3].
The two sides have described what happened next differently. Gebru has said she was fired. Google has said it accepted what it treated as a resignation [2][3]. Jeff Dean, then the head of Google AI, wrote to staff that the paper did not meet the company's bar for publication, stating that it did not take account of recent research on mitigating the risks it described and that it had not gone through the full internal review process before submission [1]. Critics of that account, including some researchers and journalists, questioned the stated reasons and noted that the paper carried a long list of references [1]. Separately from the publication dispute, Gebru had circulated an internal email criticizing the company's diversity and inclusion efforts, which several news outlets reported as a factor in the events [3][13].
The departure prompted an open letter, organized under the name Google Walkout For Real Change, that was signed by roughly 2,700 Google employees and more than 4,300 academics and members of civil society in support of Gebru [1][13]. In February 2021 Google dismissed Margaret Mitchell, the other co-lead of the Ethical AI team, after what the company said was an investigation into the movement of files; Mitchell had been openly critical of Gebru's treatment [3][13]. Google's chief executive, Sundar Pichai, said the company would review the circumstances, and the company later announced changes to how it handled research review and employee departures [3][13].
"On the Dangers of Stochastic Parrots" was published at the 2021 ACM Conference on Fairness, Accountability, and Transparency, where it appeared as pages 610 to 623 of the proceedings. Its listed authors are Emily Bender, Timnit Gebru, Angelina McMillan-Major, and "Shmargaret Shmitchell," a pseudonym widely reported to refer to Margaret Mitchell [14][15]. The paper sets out several categories of risk associated with very large language models. It points to the environmental and financial costs of training such systems, the difficulty of auditing models trained on web-scale text that can encode racist and other harmful language, the research effort directed toward ever larger models rather than toward genuine language understanding or smaller curated datasets, and the way fluent but ungrounded output can be used to mislead [1][14].
The paper defines a large language model as a "stochastic parrot," which the authors describe as a system for "stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning" [14]. The phrase "stochastic parrots" became a shorthand in subsequent discussion for the argument that such a model produces plausible text without understanding it [1][15]. The paper, and the circumstances of its publication, are often cited in coverage of debates over the direction of language model research that intensified after the public release of systems such as ChatGPT [15][5].
On December 2, 2021, one year after she left Google, Gebru announced the Distributed Artificial Intelligence Research Institute, or DAIR [4][5]. She described it as an independent, community-rooted institute meant to support AI research that is not shaped by the commercial incentives of large technology companies, with researchers based in different parts of the world. Initial backers reported at launch included the Ford Foundation, the John D. and Catherine T. MacArthur Foundation, the Kapor Center, and the Open Society Foundations [4][5]. The institute states that its work focuses on documenting and reducing the harms of AI, with attention to marginalized communities and to Africa and African diaspora populations [4][5]. In the launch announcement Gebru wrote that "when AI research, development and deployment is rooted in people and communities from the start, we can get in front of these harms and create a future that values equity and humanity" [4].
Gebru has argued that public discussion treats artificial intelligence as more capable and more inevitable than the evidence supports, and that this framing can obscure concrete harms to people affected by the systems. In materials accompanying the launch of DAIR she said that AI "needs to be brought back down to earth" and that it had been "elevated to a superhuman level that leads us to believe it is both inevitable and beyond our control" [4]. She has maintained that many harms in AI systems are preventable and that involving affected communities in how systems are built and deployed can change their outcomes [4][5]. She has also been a vocal critic of working conditions for the data workers who label and moderate content used to train AI systems [5].
Gebru is among the most outspoken skeptics of the pursuit of artificial general intelligence. In a February 2023 talk at Stanford she said, "To me, trying to build AGI is an inherently unsafe practice" [19]. In a 2024 paper in the journal First Monday, written with the philosopher Emile P. Torres, she introduced the acronym TESCREAL to describe what the authors call an interconnected bundle of ideologies, transhumanism, extropianism, singularitarianism, cosmism, rationalism, effective altruism, and longtermism, and argued that the framing of AGI as a near-term goal is rooted in this tradition rather than in technical evidence [20].
Gebru received a VentureBeat AI Innovation Award in the "AI for Good" category in 2019 for her work on algorithmic bias [6]. Nature named her one of its ten people who helped shape science in 2021, describing her as an AI ethics leader [16]. TIME included her on its list of the 100 most influential people of 2022 [17]. Fortune listed her among the world's greatest leaders, and the BBC named her to its 100 Women list in 2023 [6][18]. In 2023 the Carnegie Corporation of New York recognized her as a Great Immigrants honoree [6].
| Field | Detail |
|---|---|
| Name | Timnit Gebru |
| Born | Around 1982 or 1983, Addis Ababa, Ethiopia |
| Heritage | Eritrean; raised in Ethiopia; political asylum in the United States |
| Fields | Artificial intelligence, computer vision, AI ethics, algorithmic bias |
| Education | Stanford University (B.S. and M.S. electrical engineering; Ph.D. computer vision, 2017) |
| Doctoral advisor | Fei-Fei Li |
| Known for | "Gender Shades"; "Datasheets for Datasets"; "On the Dangers of Stochastic Parrots"; co-founding Black in AI; the TESCREAL critique |
| Former roles | Engineer at Apple; postdoctoral researcher at Microsoft Research; co-lead of Google Ethical AI (2018 to 2020) |
| Current role | Founder and executive director, Distributed Artificial Intelligence Research Institute (DAIR), founded December 2, 2021 |
| Recognition | Nature's 10 (2021); TIME 100 (2022); BBC 100 Women (2023); VentureBeat AI Innovation Award (2019) |