Episode Summary
Executive Summary: Timnit Gebru discusses founding DARE, a distributed AI research institute built to challenge exploitative, centralized models of AI research and funding. She reflects on her firing from Google, the dangers of large language models, why ethics work inside big tech can become a fig leaf, and how DARE aims to prioritize interdisciplinary, community-led research—especially for Africa and the African diaspora.
Main Topics: Why DARE exists (Priority: 5/5): Gebru explains DARE as an independent, distributed research institute meant to avoid power consolidation, brain drain, and exploitative research norms while enabling more context-specific work. Firing from Google and its aftermath (Priority: 5/5): She recounts being 'resignated' from Google after coauthoring the large language models paper, describing harassment, disruption, and the difficulty of processing the event amid the pandemic and conflict in Ethiopia. Institution building and incentive structures (Priority: 5/5): A central theme is that fairness cannot be achieved without the right institutions, labor conditions, and incentives; she argues current academia and industry often reward the wrong behavior. Critique of big tech ethics teams (Priority: 4/5): Gebru argues internal ethics groups in large tech firms often function as public-relations shields against regulation rather than as mechanisms for real change. Community-led and distributed research (Priority: 5/5): DARE is framed as interdisciplinary and bottoms-up, centering lived experience, especially from Africa and the African diaspora, and avoiding 'parachute science' or extraction from affected communities. Research projects and practical examples (Priority: 4/5): She cites ongoing work such as spatial apartheid analysis in South Africa and data stewardship models as examples of research that is socially grounded and operationally different from mainstream AI work.
Key Arguments: Fairness cannot be reduced to a mathematical metric; it depends on institutions, incentives, and the broader social system in which technology is built and used. Large companies often use ethics research strategically to defend themselves against regulation, not necessarily to improve outcomes for impacted communities. Academia can be just as structurally constraining as industry because tenure pressure, publish-or-perish norms, and dependence on advisors create strong disincentives to speak out. A distributed institute allows researchers to stay in their communities and contribute relevant expertise without forcing migration to Silicon Valley. Research should be led with and by communities affected by AI systems, rather than extracting data, stories, or labor from them without compensation or recognition. Alternative organizational models already exist in grassroots networks and community projects, and DARE aims to learn from and extend those models. Funding shapes the agenda: if money comes from warfare, surveillance, or billionaire philanthropy, it influences the kinds of technologies and research that get produced.
Data Points: Google appearances on show: about the 4th time - Sam Charrington notes Gebru has been on the podcast multiple times. Episode count reference: episode 88 - Her first appearance on the show is recalled as January 2018, episode 88. Current podcast episode estimate: around 588 - Host jokes the show is now around episode 588. Google firing timeline: middle of the pandemic - Gebru says she was fired while the pandemic was ongoing. Ethiopia conflict timeline: a war that just started in Ethiopia - She links her firing to a period when war had just begun in Ethiopia. Founding funders mentioned: MacArthur, Ford, Kapoor Center, Rockefeller Foundation, Open Society Foundation - Initial funding sources for DARE. Major funding concern: 100% grant-based funding risk - She worries about an institute depending entirely on grants. Research team size mentioned: 2 research fellows plus 1 likely full-time hire soon - Gebru describes DARE's early staffing state. Large language model paper year: 2022 implied context after prior events - She refers to writing the paper on dangers of large language models after internal Google concerns rose. Graduate student stipend figure: $36,000 a year - She cites a tweet mentioning graduate student compensation to illustrate exploitative academic labor. Spatial apartheid origin year: 1950 - She explains the Group Areas Act in South Africa. Black in X network growth: multiple networks across disciplines - She notes the expansion from Black in AI to many other identity-based scholarly networks. Red line burn-rate example: October - She mentions a financial forecast where the institute could run out of money by October.
Pivotal Quotes: "if you don't have the right institution and the right structure, there's just no way that you can do things quote unquote fairly." — Timnit Gebru: Her core thesis on why fairness requires institutional change, not just better models. "I do think that this is their goal. But so the people inside could know that and try to fight that, right?" — Timnit Gebru: On why internal ethics teams at big tech may function as defenses against regulation, while still leaving room for worker organizing. "I want to call it dare. Like, does it sound weird? You know, it's like, no, it's cool. So that's dare." — Timnit Gebru: Explaining the naming and founding of the Distributed AI Research Institute.
Implications: The interview argues AI accountability needs independent institutions, not just internal ethics roles. For researchers, it highlights collective organizing, community-led work, and diversified funding as the path to more credible, just AI development.