Research professor studying AI's social and material costs
Kate Crawford
Profile
Kate Crawford writes about what AI actually costs — not the cloud-shaped abstraction but the lithium mines, the data centre coolant, the underpaid annotators in Nairobi, the scraped photos no one consented to. She is Australian, a Research Professor at USC Annenberg, a Senior Principal Researcher at Microsoft Research New York, and the inaugural visiting chair of AI and Justice at the École Normale Supérieure in Paris. She is also one of the founders of the AI Now Institute at NYU, which more or less established the field of critical AI policy research before most of the current AI hype cycle started.
Her 2021 book Atlas of AI is the one to read. It argues that AI is neither artificial nor intelligent — it is an extractive industry, materially continuous with mining and logistics, and it concentrates power in ways the field’s own ethics committees rarely admit. The book was named a Financial Times book of the year, won three international prizes, and has been translated into twelve languages. If you’re a developer who has been shipping LLM features without thinking too hard about where the training data came from or what the inference bill looks like in megawatts, this is the book that will make you uncomfortable in a useful way.
Crawford’s other significant move is doing serious research as visual art. With Vladan Joler she made Anatomy of an AI System — a giant diagram tracing every input that goes into one Amazon Echo — which is in the permanent collections of MoMA and the V&A. Their follow-up Calculating Empires won the Silver Lion at the Venice Biennale in 2024. With artist Trevor Paglen she built ImageNet Roulette, an art piece that fed your selfie through ImageNet’s “person” categories and showed you the racist labels — a stunt that ended with ImageNet deleting 600,000 images.
She now leads the Knowing Machines Project, a transatlantic research lab investigating how AI training datasets are actually built. For developers, her work matters because it reframes the engineering questions: every model card, every benchmark, every “data quality” pipeline is also a labour question, an environmental question, and a power question. She is the most cited voice — alongside Timnit Gebru and Margaret Mitchell — for the case that the costs of frontier AI deserve a proper accounting.
Books
Key Articles & Papers
Excavating AI: The Politics of Images in Machine Learning Training Sets Anatomy of an AI System Calculating Empires: A Genealogy of Power and Technology Since 1500 Stop talking about AI ethics. It's time to talk about power. Artificial Intelligence Is Misreading Human Emotion The Trouble with Bias (NeurIPS 2017 Keynote)Controversies
Crawford’s critics — mostly from inside large AI labs — argue that Atlas of AI over-indexes on the costs and under-credits the technical capabilities, and that her framing risks discouraging the very research she wants to reform. She is also, depending on who you ask, in an awkward position as a senior researcher at Microsoft (a major AI vendor) writing the most prominent critique of AI’s industrial footprint. She has addressed the tension in interviews but it remains a fair thing to notice when reading her.
YouTube
Spotify Podcasts
