> ## Content Index
> Fetch the complete content index at: https://www.antoinebuteau.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Lessons from Timnit Gebru
- URL: https://www.antoinebuteau.com/lessons-from-timnit-gebru/
- Published: 2026-06-25T15:04:55.000Z
- Updated: 2026-07-18T21:50:55.000Z
- Description: Timnit Gebru studies algorithmic bias and AI’s social impact as a computer scientist. Her co-written Gender Shades and Stochastic Parrots link facial-recognition and language-model failures to data representation, corporate accountability, and the field’s ideology.
- Author: Antoine Buteau
- Tags: Profile, AI & Machine Learning Profiles

![Visual summary of operating lessons from Timnit Gebru.](https://www.antoinebuteau.com/content/images/2026/06/lessons-from-timnit-gebru-profile-infographic.webp)

## Lessons from Timnit Gebru

Computer scientist Timnit Gebru researches algorithmic bias and the social impact of artificial intelligence. She co-authored "Gender Shades," exposing facial recognition's high error rates for dark-skinned women, and "Stochastic Parrots," detailing the risks of large language models. This collection documents her statements on data representation, corporate accountability, and the ideologies driving modern tech.

### Part 1: Algorithmic Bias and Representation

1. **On The Global Majority:** "Existing measures of success in AI often fail to reflect the global majority, prioritizing the preferences of those with the power to shape technology." — [*Source: \[MIT*](https://mit.edu/?ref=antoinebuteau.com)*\]*
2. **On Dataset Skews:** "The high accuracy rates claimed by many companies for facial recognition were often based on datasets overwhelmingly composed of lighter-skinned subjects." — [*Source: \[Proceedings of Machine Learning Research*](https://mlr.press/?ref=antoinebuteau.com)*\]*
3. **On The Limits of De-biasing:** "People cannot simply reduce the harms caused by machine learning to dataset bias, because these technologies are not orthogonal to underlying social biases." — [*Source: \[Y Combinator*](https://ycombinator.com/?ref=antoinebuteau.com)*\]*
4. **On Systemic Reflection:** "Because AI models are trained on human-generated data, they often reflect, amplify, and cement existing societal biases rather than correcting them." — [*Source: \[The Gray Area: Is ethical AI possible?*](https://podcasts.apple.com/gb/podcast/is-ethical-ai-possible/id1081584611?i=1000593166307&ref=antoinebuteau.com)*\]*
5. **On Intersectional Disparities:** "In facial analysis, error rates for lighter-skinned men are often below 1%, while error rates for darker-skinned women can soar as high as 34.7%." — [*Source: \[Fast.ai*](https://fast.ai/?ref=antoinebuteau.com)*\]*
6. **On The Coded Gaze:** "The priorities, preferences, and sometimes prejudices of those in power become embedded directly into algorithmic systems." — [*Source: \[APA Online*](https://apaonline.org/?ref=antoinebuteau.com)*\]*
7. **On Datasheets for Datasets:** "The field needs standardized, rigorous ways of identifying and documenting dataset skews to prevent harm before systems are deployed." — [*Source: \[Cornell University*](https://cornell.edu/?ref=antoinebuteau.com)*\]*
8. **On Representation:** "It is impossible to build technology that serves everyone if the people building it do not represent the communities it impacts." — [*Source: \[Relativity*](https://relativity.com/?ref=antoinebuteau.com)*\]*
9. **On Technical Fixes:** "Technical solutions to bias are insufficient if the underlying institutions that create and deploy the AI are not fundamentally reformed." — [*Source: \[NYU*](https://nyu.edu/?ref=antoinebuteau.com)*\]*
10. **On Measuring Success:** "We must challenge what industry considers a successful AI model, particularly when that success only applies to a narrow demographic slice of the population." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*

### Part 2: The Myth of Objective AI

1. **On Sentience Claims:** "A lot of people want to imagine the machine is sentient and that there are no humans involved. That's part of a concerted effort to hide what's going on." — [*Source: \[Time*](https://time.com/?ref=antoinebuteau.com)*\]*
2. **On Hiding Human Labor:** "Framing AI as magical or objective serves to erase the massive amounts of human labor—often underpaid and precarious—required to train and maintain these systems." — [*Source: \[Tech Won't Save Us*](https://techwontsave.us/?ref=antoinebuteau.com)*\]*
3. **On Technological Determinism:** "AI development is a series of active human choices, not an inevitable technological progression that we are powerless to shape." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
4. **On Agency:** "It’s humans who decide whether all this should be done or not. We should remember that we have the agency to do that." — [*Source: \[Wikiquote*](https://wikiquote.org/?ref=antoinebuteau.com)*\]*
5. **On Unquestioned Innovation:** "We must question the fundamental premise of AI tools—specifically, whether a particular technology should be developed or deployed at all, rather than assuming all innovation is inherently beneficial." — [*Source: \[NYU*](https://nyu.edu/?ref=antoinebuteau.com)*\]*
6. **On AI as a Product:** "AI isn’t a standard IT product; it is a complex socio-technical system that requires oversight and caution rather than blind faith." — [*Source: \[Cybernews*](https://cybernews.com/?ref=antoinebuteau.com)*\]*
7. **On Naturalized Inequality:** "This hasn't happened with other technologies in the past, so why would we expect this would happen in the future?" — [*Source: \[Digital Democracies*](https://digitaldemocracies.org/?ref=antoinebuteau.com)*\]*
8. **On Problem Solving:** "Before implementing AI, leadership should always ask: What's the simplest solution to our problem, and why is AI better than that?" — [*Source: \[Daring to DAIR: Distributed AI Research with Timnit Gebru*](https://www.youtube.com/watch?v=wMmGZYVgM40&ref=antoinebuteau.com)*\]*
9. **On The Illusion of Neutrality:** "Algorithms are not neutral arbiters of truth; they reflect the institutional goals and subjective decisions of their creators." — [*Source: \[AI Weekly*](https://aiweekly.co/?ref=antoinebuteau.com)*\]*

### Part 3: The Dangers of "Stochastic Parrots"

1. **On Stochastic Parrots:** "Contrary to how it may seem when we observe its output, a language model is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data." — [*Source: \[Quote Investigator*](https://quoteinvestigator.com/?ref=antoinebuteau.com)*\]*
2. **On Lack of Meaning:** "These models generate text according to probabilistic information about how words combine, but they do so without any reference to actual meaning." — [*Source: \[Wikipedia*](https://wikipedia.org/?ref=antoinebuteau.com)*\]*
3. **On Amplifying Bias:** "Large language models risk amplifying the biases present in their massive, uncurated training datasets, reinforcing historical inequities at scale." — [*Source: \[Substack*](https://substack.com/?ref=antoinebuteau.com)*\]*
4. **On Environmental Impact:** "The race to build ever-larger models ignores the severe environmental cost and carbon footprint required to train and maintain them." — [*Source: \[Public Infrastructure*](https://publicinfrastructure.org/?ref=antoinebuteau.com)*\]*
5. **On Misleading Coherence:** "There is a profound danger in systems that mimic human coherence without possessing actual knowledge, as they can easily mislead users." — [*Source: \[E-Discovery Team*](https://e-discoveryteam.com/?ref=antoinebuteau.com)*\]*
6. **On the Rush to Deploy:** "The industry's rush to deploy AI without sufficient ethical safeguards creates a tension between profit motives and the public good." — [*Source: \[Vertex AI Search*](https://vertexaisearch.cloud.google.com/?ref=antoinebuteau.com)*\]*
7. **On Naming the Phenomenon:** "I think the 'stochastic parrots' phrase is a good one, right? When you see a parrot repeating what humans are saying, you don't assume it understands." — [*Source: \[Public Infrastructure*](https://publicinfrastructure.org/?ref=antoinebuteau.com)*\]*
8. **On Uncurated Data:** "Feeding models vast amounts of internet text without careful curation ensures that they will inevitably regurgitate the internet's worst elements." — [*Source: \[Digg*](https://digg.com/?ref=antoinebuteau.com)*\]*
9. **On Research Retraction:** "The controversy over the stochastic parrots paper demonstrated how corporate interests can suppress critical academic inquiry when it threatens their product roadmap." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
10. **On the Illusion of Understanding:** "When language models output seemingly intelligent text, humans naturally project meaning onto it, masking the model's fundamental lack of comprehension." — [*Source: \[Quote Investigator*](https://quoteinvestigator.com/?ref=antoinebuteau.com)*\]*

### Part 4: Big Tech and Profit Motives

1. **On The Gold Rush:** "It feels like a gold rush. In fact, it is a gold rush. And a lot of the people who are making money are not the people actually in the midst of it." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
2. **On Corporate Self-Regulation:** "I think it made it really clear that unless there is external pressure to do something different, companies are not just going to self-regulate." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
3. **On The Need for External Pressure:** "We need regulation and we need something better than just a profit motive to guide the development of artificial intelligence." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
4. **On Dissent in Big Tech:** "Big tech is consumed by a drive to develop AI and they don’t want someone like me who’s going to get in their way." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
5. **On Corporate Priorities:** "The primary incentive for major technology companies is to maximize shareholder value, which frequently conflicts with the time and resources required for ethical AI research." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*
6. **On Suppressing Criticism:** "When companies silence their internal ethics teams, they send a clear message that critical research is only welcome when it aligns with public relations." — [*Source: \[Time*](https://time.com/?ref=antoinebuteau.com)*\]*
7. **On Concentration of Power:** "The current trajectory of AI development centralizes immense power within a handful of Silicon Valley corporations, marginalizing alternative voices." — [*Source: \[Daring to DAIR: Distributed AI Research with Timnit Gebru*](https://www.youtube.com/watch?v=wMmGZYVgM40&ref=antoinebuteau.com)*\]*
8. **On Public Accountability:** "Without robust legal and regulatory frameworks, the public is forced to rely on the goodwill of corporations that are fundamentally designed to seek profit." — [*Source: \[Medium*](https://medium.com/?ref=antoinebuteau.com)*\]*
9. **On the Firing Aftermath:** "I was not in thinking mode. I was just in action mode, like: 'I need a lawyer and I need to get my story out; I wonder what’s happening to other people?'" — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*

### Part 5: The "TESCREAL" Ideologies

1. **On the TESCREAL Bundle:** "TESCREAL ideologies—from Transhumanism to Longtermism—form an interconnected, overlapping group that heavily influences Silicon Valley's vision of the future." — [*Source: \[Truthdig*](https://truthdig.com/?ref=antoinebuteau.com)*\]*
2. **On Eugenics Roots:** "The normative framework that motivates much of the goal to build AGI is rooted in the Anglo-American eugenics tradition of the twentieth century." — [*Source: \[DAIR Institute*](https://dair-institute.org/?ref=antoinebuteau.com)*\]*
3. **On Discriminatory Attitudes:** "Many of the very same discriminatory attitudes that animated eugenicists in the past—racism, xenophobia, classism, ableism, and sexism—remain widespread within the movement to build AGI." — [*Source: \[First Monday*](https://firstmonday.org/?ref=antoinebuteau.com)*\]*
4. **On Distracting from Real Harm:** "When someone talks about the 'existential risk' posed by artificial general intelligence, look under the hood. It is often a distraction from the immediate harms of algorithmic bias and labor exploitation." — [*Source: \[DAIR Institute*](https://dair-institute.org/?ref=antoinebuteau.com)*\]*
5. **On Ideological Influence:** "It's really important for people to understand what this bundle of ideologies is, because it's become so hugely influential, and is shaping our world right now." — [*Source: \[Truthout*](https://truthout.org/?ref=antoinebuteau.com)*\]*
6. **On Effective Altruism:** "Silicon Valley EA types are convincing themselves that the way in which they're exploiting people and causing harm is the best possible thing they can be doing in the world." — [*Source: \[Effective Altruism Org*](https://effectivealtruism.org/?ref=antoinebuteau.com)*\]*
7. **On Ineffective Altruism:** "Some of my collaborators had the idea of creating an 'ineffective altruism' movement. Where one works on boring useless things like, I dunno, poverty. All this would be funny if this were a fringe thing." — [*Source: \[Join Reboot*](https://joinreboot.org/?ref=antoinebuteau.com)*\]*
8. **On Power Seeking:** "These utopian ideologies often operate as power-seeking behavior dressed up as altruism, prioritizing a hypothetical distant future over marginalized groups living in the present." — [*Source: \[Johannes Kleske*](https://johanneskleske.com/?ref=antoinebuteau.com)*\]*
9. **On Merging with Institutions:** "I guess it was only a matter of time when the religion of effective altruism, worshipping so-called artificial general intelligence, merged with entrenched religious institutions." — [*Source: \[Bluesky*](https://bsky.app/?ref=antoinebuteau.com)*\]*

### Part 6: Diversity and Institutional Racism

1. **On Being the Only One:** "I grew tired of being the only Black person in a large pool of engineers. Researchers from marginalized backgrounds have to take a stand for our own people." — [*Source: \[Infinite Family*](https://infinitefamily.org/?ref=antoinebuteau.com)*\]*
2. **On Building Support Networks:** "I’ve learned that we need to have a network, we need to stand up for each other in order to survive in spaces that weren't built for us." — [*Source: \[Sage Journals*](https://sagepub.com/?ref=antoinebuteau.com)*\]*
3. **On Creating Islands of Safety:** "I wanted some amount of autonomy, some amount of ability to create a space for people in marginalized groups—an island of safety where our work could thrive." — [*Source: \[Sage Journals*](https://sagepub.com/?ref=antoinebuteau.com)*\]*
4. **On the Message Sent by Firings:** "Young women of color, I want them to know that these moments are necessary. What's painful to me is the message that this sends: if this is happening to me, what's happening to other people?" — [*Source: \[Medium*](https://medium.com/?ref=antoinebuteau.com)*\]*
5. **On Structural Barriers:** "We can talk about the ethics and fairness of AI all we want, but if our institutions don't allow for this kind of work to take place, then it won't." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*
6. **On Exhaustion:** "The constant battle to justify the importance of diversity and ethical oversight to leadership takes an immense toll on researchers of color within corporate environments." — [*Source: \[Scribd*](https://scribd.com/?ref=antoinebuteau.com)*\]*
7. **On Performative Diversity:** "Companies frequently use diversity initiatives as public relations tools while simultaneously dismissing or penalizing the actual substantive work produced by Black researchers." — [*Source: \[Engadget*](https://engadget.com/?ref=antoinebuteau.com)*\]*
8. **On the Burden of Proof:** "Marginalized researchers are routinely subjected to higher burdens of proof and harsher scrutiny when publishing critical work compared to their peers." — [*Source: \[Washington Post*](https://washingtonpost.com/?ref=antoinebuteau.com)*\]*
9. **On Institutional Change:** "At the end of the day, achieving equitable technology requires deep institutional and structural change, not merely token representation on engineering teams." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*

### Part 7: Labor Exploitation and Power Structures

1. **On Invisible Labor:** "The creation of AI relies on an invisible underclass of data annotators and content moderators who are chronically underpaid and exposed to traumatic material." — [*Source: \[Time*](https://time.com/?ref=antoinebuteau.com)*\]*
2. **On the Global South:** "AI systems are typically designed by and for a narrow demographic in the Global North, while disproportionately exploiting labor and extracting data from the Global South." — [*Source: \[Public Infrastructure*](https://publicinfrastructure.org/?ref=antoinebuteau.com)*\]*
3. **On Hiding the Humans:** "Portraying AI as autonomous and magical intentionally obscures the massive human labor infrastructure that actually makes these systems function." — [*Source: \[Tech Won't Save Us*](https://techwontsave.us/?ref=antoinebuteau.com)*\]*
4. **On Resource Extraction:** "The current model of AI development closely mirrors historical resource extraction, taking data without consent and returning little value to the communities that generated it." — [*Source: \[Racism and Technology Center*](https://racismandtechnology.center/?ref=antoinebuteau.com)*\]*
5. **On Concentrating Wealth:** "The financial gains from the AI 'gold rush' are overwhelmingly captured by executives and investors, not the workers performing the foundational data labor." — [*Source: \[The Guardian*](https://theguardian.com/?ref=antoinebuteau.com)*\]*
6. **On Decentralizing Expertise:** "We must actively work to fracture the concentration of power in Silicon Valley and build instead a decentralized and local base of technological expertise." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*
7. **On Surveillance and Control:** "We must oppose a technological future that is used for surveillance, warfare, and the centralization of power by Silicon Valley monopolies." — [*Source: \[Dr. Timnit Gebru on the Smoke and Mirrors of AI Hype*](https://www.youtube.com/watch?v=QdB7BcPgfQ4&ref=antoinebuteau.com)*\]*
8. **On Rejecting Inevitability:** "There is nothing inevitable about the current exploitative structure of the tech industry; it is the result of deliberate business models that can and should be challenged." — [*Source: \[Digital Democracies*](https://digitaldemocracies.org/?ref=antoinebuteau.com)*\]*
9. **On Empowering Workers:** "Protecting the public from harmful AI requires protecting the rights, wages, and working conditions of the laborers who build and maintain these datasets." — [*Source: \[Tech Won't Save Us*](https://techwontsave.us/?ref=antoinebuteau.com)*\]*

### Part 8: Building Alternatives and Community-Centered AI

1. **On Rethinking Institutions:** "If we had the opportunity to pursue this work from scratch, how would we want to build these institutions to ensure they actually serve the public?" — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*
2. **On Community Benefit:** "If we want AI that genuinely benefits our communities, then we must critically ask what kind of processes and methodologies we should follow to get there." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*
3. **On the Mission of DAIR:** "The goal of independent research is to foster a technological future that serves our communities instead of one that centralizes power." — [*Source: \[DAIR founder Timnit Gebru on potential harms of AI*](https://www.youtube.com/watch?v=b1x500Ic-mw&ref=antoinebuteau.com)*\]*
4. **On Independent Research:** "For ethical AI research to truly flourish, it requires structural independence completely separated from the influence and financial incentives of large tech corporations." — [*Source: \[Cybernews*](https://cybernews.com/?ref=antoinebuteau.com)*\]*
5. **On Centering the Margins:** "We must encourage tech companies to consider all perspectives—especially those from marginalized groups—when designing products and services that shape society." — [*Source: \[Quartz*](https://qz.com/?ref=antoinebuteau.com)*\]*
6. **On Local Expertise:** "True technological equity means empowering local communities to govern and develop their own tools, rather than having solutions imposed from the outside." — [*Source: \[Stanford University*](https://stanford.edu/?ref=antoinebuteau.com)*\]*
7. **On Reframing Innovation:** "Innovation should be measured not by the scale of the model or the processing power used, but by the tangible positive impact the technology has on vulnerable populations." — [*Source: \[DAIR Institute*](https://dair-institute.org/?ref=antoinebuteau.com)*\]*
8. **On Collective Action:** "Addressing systemic technological harms requires collective organizing and solidarity, rather than relying on the isolated efforts of individual researchers." — [*Source: \[Public Infrastructure*](https://publicinfrastructure.org/?ref=antoinebuteau.com)*\]*
9. **On the Value of Slowness:** "Building ethical, community-rooted technology requires moving slowly, gathering consensus, and respecting the time it takes to understand potential downstream consequences." — [*Source: \[DAIR Institute*](https://dair-institute.org/?ref=antoinebuteau.com)*\]*
10. **On a Better Future:** "A different tech ecosystem is possible—one where human dignity and community well-being are prioritized over the relentless pursuit of scale and artificial general intelligence." — [*Source: \[Truthout*](https://truthout.org/?ref=antoinebuteau.com)*\]*