Last reviewed
May 31, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 · 1,739 words
Add missing citations, update stale details, or suggest a clearer explanation.
Margaret Mitchell is an American computer scientist who works on AI ethics, fairness in machine learning, and the documentation of AI systems. She is best known for proposing "model cards," a standard format for reporting how a trained model behaves, and for her time at Google, where she founded and co-led the company's Ethical Artificial Intelligence team alongside Timnit Gebru. Since late 2021 she has been a researcher and chief ethics scientist at Hugging Face [1][2][3].
This article is about the AI and machine learning researcher, who is sometimes called "Meg" Mitchell. She is a different person from the American novelist Margaret Mitchell, who wrote the 1936 novel "Gone with the Wind" and died in 1949. The two share a name but are unrelated [3].
| Field | Detail |
|---|---|
| Full name | Margaret Mitchell |
| Known as | Meg Mitchell |
| Born | 1983, United States |
| Fields | AI ethics, fairness in machine learning, natural language processing, computer vision |
| Education | Reed College (BA, 2005); University of Washington (MS, 2009); University of Aberdeen (PhD, 2013) |
| Doctoral thesis | "Generating Reference to Visible Objects" (2013) |
| Known for | Model cards; co-leading Google's Ethical AI team; "On the Dangers of Stochastic Parrots" |
| Employers | Microsoft Research; Google; Hugging Face |
| Current role | Researcher and chief ethics scientist, Hugging Face (2021 to present) |
Mitchell studied linguistics at Reed College in Portland, Oregon, completing a bachelor's degree in 2005 [3][4]. After working for about two years as a research assistant at the OGI School of Science and Engineering, she went on to the University of Washington, where she earned a master's degree in computational linguistics in 2009 [3][4].
She then enrolled in a doctoral program at the University of Aberdeen in Scotland. Her dissertation, titled "Generating Reference to Visible Objects," dealt with how a system can describe objects in a scene using language, and she received her PhD in 2013 [3][4]. Around this period she also held a postdoctoral position at the Human Language Technology Center of Excellence at Johns Hopkins University [3].
Mitchell's research sits at the meeting point of natural language processing and computer vision, with a later focus on fairness, transparency, and data practices in machine learning [1][3].
From 2013 to 2016 Mitchell was a researcher at Microsoft Research, where she worked on generating natural language descriptions of images, a task often called image captioning [3][5]. She was one of the research leads on Seeing AI, a Microsoft project that grew into a free app intended to help people who are blind or have low vision by narrating text, scenes, and objects captured through a phone camera [3][6]. Her group's wider line of vision-to-language work also fed into CaptionBot, a public Microsoft demonstration site that produced written descriptions of uploaded photographs [5][6].
In 2018 Mitchell and several colleagues circulated a preprint proposing model cards, and the resulting paper, "Model Cards for Model Reporting," appeared at the 2019 Conference on Fairness, Accountability, and Transparency [7]. The co-authors were Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, Ben Hutchinson, Elena Spitzer, Inioluwa Deborah Raji, and Timnit Gebru [7].
A model card is a short document that accompanies a trained model and records its intended uses, its limitations, and how it performs under different conditions. A central idea in the paper is that performance should be broken down by group, for example across demographic or phenotypic subgroups and across relevant operating conditions, rather than reported only as a single aggregate score [7]. The format was meant to make it easier to see where a model works well and where it does not, and it has since been adopted in various forms across the industry [1][7]. While at Google, Mitchell also helped lead the Google Cloud Model Cards effort that put the idea into a product [1][8].
Much of Mitchell's research addresses how machine learning systems can reflect or amplify social bias, and how that can be measured and reduced. She has published methods for removing unwanted associations tied to demographic groups from models, including approaches that use adversarial learning [1][3]. She frames documentation, evaluation, and careful handling of training data as practical tools for algorithmic fairness rather than abstract goals [1][9]. She also co-founded Widening NLP, an effort within the natural language processing community to broaden participation in the field [3].
Mitchell joined Google in November 2016 as a senior research scientist [3]. There she founded the company's Ethical Artificial Intelligence team and co-led it with Timnit Gebru, building a research group focused on fairness, accountability, and the social effects of AI [1][3].
In 2020 Mitchell, Gebru, and co-authors Emily M. Bender and Angelina McMillan-Major wrote "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?", a paper arguing that very large language models carry risks including environmental cost, hard-to-audit training data, and the encoding of social bias [3][9]. Mitchell published parts of this work under the name "Shmargaret Shmitchell" [3]. The paper became a flashpoint in a dispute with Google, and Gebru's employment ended in December 2020 under contested circumstances [3][9].
In January 2021 Google locked Mitchell's corporate account and opened an investigation [3]. On February 19, 2021, the company dismissed her. Accounts of why differ, and the following characterizations are attributed to their sources.
Google's position. In a statement, a Google spokesperson said: "After conducting a review of this manager's conduct, we confirmed that there were multiple violations of our code of conduct, as well as of our security policies, which included the exfiltration of confidential business-sensitive documents and private data of other employees" [10][11]. Reporting at the time described the company's concern as centering on Mitchell's use of an automated script to search through her email, which Google said involved moving files outside the company [10][11].
Mitchell's position and that of her supporters. Mitchell and others described the file activity differently, saying she had been gathering material that documented the treatment of Gebru. She publicly defended Gebru, and after her own dismissal she said she had "tried to use my position to raise concerns to Google about race and gender inequity, and to speak up about Google's deeply problematic firing of Dr. Gebru" [10][11]. She also called the outcome "devastating" and said she hoped that speaking out would advance the cause of ethical AI [10]. In later interviews, Mitchell and Gebru said they had effectively been forced out after the language model paper [9][12].
The two departures drew wide coverage and prompted debate about academic freedom and corporate research, including criticism from outside researchers and some Google employees [9][11][12].
Mitchell joined Hugging Face, an open-source AI company, in late 2021 as a researcher and chief ethics scientist [2][12]. There she works on responsible model development, evaluation, and the handling of training data, and she has described a multi-year plan to build ethics practices into how models are made and shared rather than treating ethics as a review at the end [12].
She took on data governance for BigScience, a large international research collaboration coordinated by Hugging Face that trained the multilingual large language model BLOOM [13][14]. The project released BLOOM openly under a Responsible AI license that placed conditions on certain uses, an approach Mitchell and collaborators presented as a way to combine openness with limits on harm [13][14].
Mitchell argues that many harms from AI are concrete and present, such as biased outputs and opaque training data, and she has been skeptical of framing that treats AI mainly as a distant existential threat. She has suggested that talk of AI ending the world can serve the interests of the companies promoting the technology by drawing attention away from nearer problems [9][2]. She links the quality of AI systems to who builds them, contending that a lack of diversity and inclusion in AI organizations tends to produce worse technology [9].
Her continuing work emphasizes documentation, transparency, careful data practices, and evaluation methods that surface differences across groups, themes that run from the original model cards paper through her work at Hugging Face [1][7][9].
In 2023 Mitchell was named to the inaugural TIME100 AI list of the most influential people in artificial intelligence [9][15]. Her work on Google Cloud Model Cards was recognized with a United States Department of Defense Tech Spotlight Award associated with then Secretary of Defense Ash Carter [1][8]. She is widely cited as one of the founders of the modern field of AI ethics and is a frequent speaker and commentator on fairness and accountability in machine learning [1][2][9].