# Timnit Gebru ## Overview Timnit Gebru is an Ethiopian-Eritrean American computer scientist, AI ethics researcher, and one of the most consequential critics of the tech industry's AI ideology. After co-leading Google's Ethical AI team, she was fired in December 2020 following her publication of research on the present-day risks of large language models—a moment that became a flashpoint for understanding how the tech oligarchy handles internal dissent around AI development. Gebru's work represents the most important counter-narrative to the [[TESCREAL]] worldview. While the TESCREAL framework emphasizes speculative future risks (superintelligence, existential threats), Gebru centers concrete present-day harms: algorithmic bias, surveillance capitalism, labor displacement, environmental devastation, and the concentration of power in the hands of a few corporations and billionaires. She has been a co-creator of the TESCREAL framework itself (with philosopher [[Émile Torres]]), using that analysis to expose how the ideology operates inside AI companies and shapes policy. Her trajectory—from Stanford computer vision researcher to Google insider to fired dissident to founder of the independent DAIR Institute—illustrates both the potential and the limits of dissent within corporate structures. Her firing also demonstrated that corporate "AI ethics" teams exist primarily as public relations infrastructure, tolerated only when they do not threaten the core business model of building larger, more profitable AI systems. ## Background and Early Career Gebru was born in Ethiopia during the Ethiopian-Eritrean War and grew up in Eritrea. She emigrated to the United States at age 15. She completed her PhD in electrical engineering at Stanford in 2017, advised by Li Fei-Fei, where her dissertation focused on computer vision and facial recognition. This early work in facial recognition research proved formative. Unlike much of the field, which treated recognition systems as neutral technical problems, Gebru's research began to surface questions about whose faces these systems worked for, whose they failed on, and what the deployment of biased systems into police departments and border enforcement meant for marginalized communities. This technical foundation directly informed her later activism. Before joining Google, Gebru worked as a researcher at Apple and Microsoft Research, continuing to investigate the social implications of computer vision systems. Throughout this period, she observed the extreme demographic homogeneity of the AI research field: the researchers, conference attendees, and industry employees were overwhelmingly male and overwhelmingly non-Black. ### Co-founding Black in AI (2017) In 2017, Gebru co-founded **Black in AI** with Rediet Abebe. The organization was created to increase Black representation in AI research and to explicitly address the field's structural exclusion of Black people. Black in AI holds annual workshops at major conferences like NeurIPS, creating community and professional development pathways for Black researchers in a field that had largely ignored them. The founding of Black in AI encapsulates one of Gebru's core critiques: the extreme demographic homogeneity of AI research is not a neutral fact but a structural problem. When the people building AI systems do not represent the people most affected by those systems, the resulting technologies encode and automate existing power structures and discrimination. Representation, from this perspective, is not a diversity initiative but a necessary precondition for building AI systems that don't replicate and amplify racial hierarchies. ## The "Stochastic Parrots" Paper and Google Firing (December 2020) The pivotal moment in Gebru's career came in December 2020, when she was fired from Google after publication of a paper that directly threatened the company's multi-billion-dollar investments in large language models. ### The Paper In March 2021 (submitted earlier), Gebru co-authored **"On the Dangers of [[Stochastic Parrots]]: Can Language Models Be Too Big?"** with Emily Bender, Margaret Mitchell, and Angelina McMillan-Major. The paper made several interconnected arguments: 1. **Environmental costs**: Large language models require massive computational resources to train, with enormous carbon footprints. These environmental costs are often obscured in industry literature and not counted in the calculus of whether such systems should be built. 2. **Trained on biased internet data**: LLMs are trained on internet text, which encodes human biases—racism, sexism, misogyny, antisemitism, transphobia. These biases become baked into the model's outputs in ways that are difficult to detect and remediate. 3. **The illusion of meaning**: Large language models excel at generating statistically plausible text that feels coherent. But they lack understanding, intentionality, or truthfulness. The paper coined the term "stochastic parrots" to describe this: systems that produce text without comprehending it. 4. **Concrete risks to marginalized communities**: Without careful safeguards, deploying systems trained on biased data that lack understanding will cause concrete harms to the communities most affected by discrimination: people of color, low-income people, incarcerated people, immigrants. These systems might be used to deny loans, healthcare, employment, or legal representation to already marginalized people. The paper was not speculative. It was grounded in documented harms from existing systems and specific warnings about risks that were already being realized. ### The Suppression Google demanded that Gebru retract the paper or remove her name. When she refused, she was fired—though Google initially claimed she had resigned. The company disputed her characterization of the firing as retaliation, but the timeline and circumstances made clear what had happened: a paper that threatened a core part of Google's business model had been suppressed through employment termination. Jeff Dean, Google's head of AI, sent internal emails criticizing the paper and defending the company's approach to large language models. The internal reaction made clear that this was not a disagreement about scientific method or empirical rigor; it was a direct threat to the company's direction and investment strategy. Margaret Mitchell, Gebru's co-lead on the Ethical AI team, was also fired in February 2021 after defending Gebru and attempting to investigate the circumstances of her termination. Mitchell's firing cemented the message: Google's AI ethics team existed only insofar as it did not question the company's core business decisions. ### Broader Significance The firing became a watershed moment for understanding corporate AI ethics. It demonstrated several critical truths: 1. **Corporate "ethics" is performative**: Google had invested in hiring respected AI ethics researchers, building the Ethical AI team, and publicly committing to responsible AI. But when those researchers actually produced research that threatened Google's business interests, they were eliminated. 2. **The business model drives everything**: Google was simultaneously investing billions in exactly the kind of large language models that the "Stochastic Parrots" paper warned against. The paper was not being suppressed because of scientific flaws but because it pointed out the contradiction between Google's "responsible AI" branding and its actual business decisions. 3. **Dissent will not be tolerated**: For other employees and researchers working in AI, the firing sent a clear message: you can participate in "ethics" only if it does not actually challenge the core direction of the company. Real dissent results in termination. 4. **The field would proceed regardless**: The paper did not slow the development of large language models. [[OpenAI]] released GPT-3 in 2020. Google released its own competing models. The industry accelerated rather than reconsidered. The firing served to remove a voice of criticism from within the industry, not to change the industry's direction. ## The TESCREAL Framework In 2023, Gebru co-created the **[[TESCREAL]]** framework with philosopher [[Émile Torres]]. The framework identifies seven interconnected ideologies that together constitute a single coherent worldview dominating the tech elite: - **[[Transhumanism]]**: The belief that humanity should transcend biological limits through technology - **[[Extropianism]]**: The claim that technology will solve all problems and that we should reject any limits on technological progress - **[[Singularitarianism]]**: The prophecy of an imminent technological singularity that will transform humanity - **Cosmism**: The belief in cosmic immortality and the redemption of all beings through technology - **[[Rationalism]]**: A specific epistemology centered on logic and mathematics as the sole sources of truth, dismissing other ways of knowing - **[[Effective Altruism]]**: A philanthropic framework that claims to maximize good through reason and quantification - **[[Longtermism]]**: The belief that future people (not yet born) should be the primary moral consideration, potentially outweighing present suffering Gebru's contribution to this framework was unique: she brought the perspective of someone who had worked inside AI companies and observed how these ideologies operate in practice, not just in theory. While Torres provided philosophical and historical analysis, Gebru could point to concrete manifestations—how rationalism shaped Google's culture, how longtermism justified ignoring present-day harms, how effective altruism rhetoric was used to secure funding for AI companies. ### The Critique The TESCREAL framework, from Gebru's perspective, reveals the internal coherence of what appears as isolated ideologies. A person working at an AI company might encounter: - Longtermist arguments that future existential risks justify current power concentration - Rationalist epistemology that dismisses qualitative accounts of AI harm as insufficiently quantified - Effective altruist philanthropic rhetoric that funnels billions toward AI research (because AI is claimed to have high expected value) - Extropian ideology that treats technological acceleration as inherently good These are not independent philosophical positions but a unified system that serves a specific function: justifying the concentration of power in the hands of people (mostly billionaires and tech executives) who can "solve problems" through technology that only they can control. Gebru's insight—developed through both TESCREAL and her independent work—is that what the tech elite call "AI safety" is actually a mechanism for consolidating power, not distributing it. "AI safety" research, as practiced by the TESCREAL community, focuses on making increasingly powerful AI systems "safe" in the hands of corporations and rationalist researchers. It does not consider whether those systems should be built at all, who should control them, or how to distribute power more broadly. ## DAIR Institute (Distributed AI Research) In December 2021, a year after her firing, Gebru founded the **DAIR Institute** (Distributed AI Research)—a nonprofit research organization pursuing AI ethics research entirely outside corporate control. The name itself encodes her thesis: AI research needs to be distributed, not concentrated in the hands of a few corporations (Google, OpenAI, Meta, Microsoft) controlled by a few billionaires. The centralization of AI research in corporate labs means that: 1. Research priorities are set by what's profitable, not what's necessary 2. Researchers who challenge corporate directions face retaliation 3. The power to decide how AI systems are deployed is concentrated 4. Communities affected by AI systems have no say in their design DAIR's research focuses on: - **Community-based AI research**: Including affected communities in the design and evaluation of AI systems rather than treating them as passive subjects - **Impact on marginalized communities**: Centering the perspectives and experiences of people most harmed by AI systems—people of color, low-income people, incarcerated people, immigrants, disabled people - **AI governance that centers affected populations**: Rather than governance designed by and for AI developers and corporate executives DAIR has become a model for independent AI research that maintains the rigor of academic work while centering justice and accountability. It represents a concrete alternative to corporate AI labs—not the only alternative, but an important one. ## The "[[AI Safety]]" vs. "AI Ethics" Schism To understand Gebru's role in this knowledge base, it is critical to grasp the fundamental split she represents in how to think about AI risk. ### The TESCREAL/"AI Safety" View The dominant perspective in [[Nick Bostrom]], [[Eliezer Yudkowsky]], the [[Rationalism]] community, and increasingly at AI companies like OpenAI is: - **The biggest AI risk is speculative future superintelligence** that could destroy humanity - Present-day harms (bias, surveillance, labor displacement) are real but secondary to existential risk - The solution is "alignment research"—making increasingly powerful AI systems aligned with human values - Paradoxically, this often means building more powerful AI systems, on the assumption that they can be made "safe" - Policy should prioritize long-term existential risk over near-term regulations ### The AI Ethics View (Gebru, Mitchell, Bender, et al.) The alternative perspective, held by Gebru and researchers like Emily Bender, Margaret Mitchell, Joy Buolamwini, Safiya Noble, and Ruha Benjamin: - **The biggest AI risks are happening right now**: Algorithmic bias systematically denying credit, employment, housing, and justice to Black people and other marginalized groups. Surveillance systems built into everyday technologies. Labor displacement concentrated in the Global South. Environmental devastation from massive computational resources. - Speculative future risks distract from concrete present harms and serve the interests of the people benefiting from current concentrations of power - The solution is accountability, regulation, community oversight, and crucially—slowing down or refusing to build certain systems - Policy should prioritize near-term protections for the communities most harmed by current systems ### Gebru's Deeper Critique Gebru goes further and argues that the TESCREAL "safety" framing is not just wrong but actively harmful. It: 1. **Redirects resources away from pressing problems**: Billions in funding go toward AI safety research aimed at future superintelligence, while almost nothing goes toward accountability for present-day algorithmic harms 2. **Serves the tech industry's interests**: More AI safety research means more AI research funding, more AI development, and more power concentrated in the hands of the tech industry 3. **Obscures the actual sources of power**: It treats AI as a neutral tool that might go wrong, rather than as a manifestation of existing power structures. The question is not whether AI can be aligned with human values, but: aligned with whose values? Whose preferences get encoded into AI systems? Whose communities get to decide? 4. **Justifies non-action on regulation**: Longtermist arguments about future risk can be used to argue against near-term regulation. ("We can't slow down AI development because the long-term benefits outweigh the near-term costs.") From this perspective, [[TESCREAL]] is not a set of well-intentioned but mistaken beliefs. It is an ideology that serves to concentrate and justify power. The tech oligarchy benefits from the framing that AI's risks are technical (and thus solvable by more rationalist researchers) rather than political (and thus requiring redistribution of power). ## Key Publications and Public Interventions **"On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?"** (2021, with Emily Bender, Margaret Mitchell, Angelina McMillan-Major) - The paper that triggered her firing from Google - Foundational text for understanding present-day AI risks **Co-creator of the TESCREAL framework** (2023, with Émile Torres) - Provides the ideological analysis essential to understanding this knowledge base - See [[TESCREAL]] for full documentation **"Datasheets for Datasets"** (2021, with others) - Proposed standardized documentation for AI training data - Addresses the problem of opaque datasets used to train systems that affect people's lives - Calls for transparency about data sources, potential biases, and limitations **"Gender Shades"** (2018, with Joy Buolamwini) - Demonstrated significant racial and gender bias in commercial facial recognition systems - Showed that the systems worked well for lighter-skinned men and failed dramatically for darker-skinned women - Became a canonical reference for how ostensibly "neutral" AI systems encode discrimination **Extensive public commentary**: Gebru maintains a visible public presence through X/Twitter, media appearances, conference presentations, and essays. She has been willing to speak directly about corporate retaliation, the concentration of power in AI, and the inadequacy of corporate "ethics" measures. ## Impact of the Google Firing The December 2020 firing and the subsequent March 2021 publication of "Stochastic Parrots" catalyzed broader movements: 1. **Alphabet Workers Union**: The firing contributed to organizing efforts at Google/Alphabet, with workers demanding accountability and protection for researchers 2. **Debate about corporate AI ethics**: Widespread questioning of whether AI ethics teams inside corporations can function as genuine accountability mechanisms or are simply public relations 3. **Departure of other researchers**: The firing signaled that corporate AI labs would not tolerate serious dissent, leading to the departure of other AI ethics researchers from industry positions 4. **Demonstration of the contradiction**: The firing exposed the fundamental contradiction between corporate "responsible AI" branding and actual business incentives. Google advertises [[ethical AI]] while investing billions in exactly the systems Gebru warned against. 5. **Reference point for understanding TESCREAL**: The firing became a concrete illustration of how TESCREAL ideology operates inside tech companies. The rationalist dismissal of her concerns, the focus on speculative future alignment rather than present-day accountability, the invocation of long-term benefits to justify suppressing dissent—all reflected TESCREAL frameworks operating in practice. ## Connections to the Right-Wing KB Gebru is a **critic** of the systems and ideologies documented in this vault, not a participant. Her work provides essential counter-framing to the narratives of [[TESCREAL]], [[Rationalism]], [[Effective Altruism]], [[Longtermism]], and the tech oligarchs who promote these ideologies. Critical links: - **[[TESCREAL]]**: Her co-creation and primary tool for analyzing the tech elite - **[[Rationalism]]**: The epistemology she critiques as dismissing other ways of knowing and concentrating power - **[[Effective Altruism]]**: The philanthropic framework she analyzes as funneling resources toward tech industry priorities - **[[Longtermism]]**: The ideology she identifies as justifying current power concentration in service of speculative future benefits - **[[Nick Bostrom]]**, **[[Eliezer Yudkowsky]]**: The rationalist AI safety figures whose frameworks she directly challenges Additionally, note: - **Targeted harassment**: Gebru has been personally targeted by TESCREAL-aligned figures and rationalist communities, including online harassment campaigns - **[[Elon Musk]] attacks**: Musk has publicly attacked Gebru on X/Twitter - **Illustration of oligarchic power**: Her firing from Google demonstrates how the tech oligarchy handles internal dissent—not through open debate but through employment termination, denials, and bureaucratic obscurity ## See Also - [[TESCREAL]] — the framework Gebru co-created - [[Margaret Mitchell]] — co-lead at Google, also fired after defending Gebru - [[Emily Bender]] — co-author of "Stochastic Parrots" - [[Émile Torres]] — co-creator of TESCREAL framework - [[Effective Altruism]] — ideology Gebru analyzes as serving tech industry interests - [[Longtermism]] — the framing she identifies as justifying present-day power concentration - [[Rationalism]] — the epistemology she critiques - [[Black in AI]] — community organization she co-founded - [[DAIR Institute]] — independent research organization she founded - [[AI Ethics]] — the field shaped by her work and firing - [[Google AI Ethics Firing]] — the pivotal event in her career and in tech accountability