Kate Crawford

persona · active · confidence: medium · reviewed 2026-09-21 · tier: semiannual

Who she is

Affiliation: Senior Principal Researcher, Microsoft Research New York (co-founder of its FATE group — fairness, accountability, transparency and ethics) — role check run 2026-09-21: the title is confirmed on the MSR people page and on her own site; the v1 "[TBD: confirm current title — since ~2017]" is resolved. The ENS press release of 2019 listed her as plain "Principal Researcher", and USC's hiring note of 2021-03-17 already says "senior principal", so the promotion landed between those dates; no announcement located. Research Professor of Communication and STS (science and technology studies — how science and tools get shaped by people and power), USC Annenberg, since March 2021 (USC, 2021-03-17). Honorary Professor, University of Sydney (her About page). Inaugural Abeona–ENS Visiting Chair in Artificial Intelligence and Justice, Paris — inaugurated 18 September 2019 under a three-year ENS–Abeona partnership (ENS press release, 2019); v1's "2021–2022 [TBD]" was wrong, and her bios still carry the chair in the present tense. Lead PI of Knowing Machines (USC/NYU, sponsored by the Alfred P. Sloan Foundation; co-PIs Mike Ananny and Jason Schultz). Co-founder of the AI Now Institute (NYU, 2017, with Meredith Whittaker) — no longer affiliated: she appears nowhere on AI Now's About page, which lists Amba Kak and Sarah Myers West as co-executive directors; Wikipedia describes her as "co-founder and former director of research" and the institute as "now a fully independent institute." Member of the European Parliament's STOA advisory panel (per the Robert Bosch Academy page, which also lists fellowships May–Aug 2018 and May–Sep 2023; no date for the STOA seat). New in window: Distinguished Fellow, Weizenbaum Institute, Berlin, 2 August to 2 September 2026, where "she will work on her new book" (Weizenbaum, 2026-07-30) — no title or publisher announced. Australian-born, based in New York.
One-line position: AI is not artificial and is only weakly "intelligent" — it is an extractive industrial system built on labor, data, minerals, water, and energy, and treating it as anything else (a neutral tool, a coming superintelligence, or a purely software phenomenon) obscures the politics baked into every layer of the stack. In-window extension (her own words, Sept 2025): generative AI runs on "a different kind of Krebs cycle" — an extractive metabolism whose cultural output is "slop" and whose endpoint is "a new and dramatic metabolic rift" (a system that eats faster than the ground can regrow) — e-flux, September 2025. Signature-framing check: she still uses "neither artificial nor particularly intelligent" (Renewable Matter, 2024-01-16) and "the extractive industry of the 21st century" (TIME100 AI, 2023-09-07); the metabolic frame is added on top, not swapped in.

Discipline & technical bet

An STS scholar and researcher-artist, not a builder. She makes no technical wager in the sense of an architecture or a model; her bet is cartographic — that drawing the whole system (mines to data centers to disposal; datasets from scrape to label) changes what people are able to argue about. The instruments: Anatomy of an AI System (2018, with Vladan Joler — one Amazon Echo as a map of labor, minerals and data), Excavating AI (2019, with Trevor Paglen — ImageNet's "person" categories), Knowing Machines' dataset forensics (LAION-5B, "9 ways to see a dataset", interviews with 18 dataset creators), and Calculating Empires (2023–, with Joler — a 24-metre wall tracing technology and power since 1500). Her one empirical wager placed in the window is about training dynamics: that generative models fed their own outputs degrade — "AI will eat itself, then gradually collapse into nonsense and noise" (e-flux, Sept 2025). That is a testable claim, and the only one on this page that a lab could falsify in a quarter. The rest of her program is diagnostic: it says what AI is made of, not what it will do next.

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

A "technology realist" by her own label (Topol, 2024-05-12), which reads as: certainty that the material ledger is being hidden on purpose, and a low-grade anger at being asked to admire the cathedral without seeing the quarry ("We haven't invested this much money into an infrastructure like this really until you go back to the pyramids"). Wants to be true: that making the system visible is itself a form of power over it.

Hears

Marx via John Bellamy Foster (the "metabolic rift"), Weizenbaum's warnings about "powerful delusional thinking," the STS canon on classification, Joler and Paglen as co-authors who think in images, and the IEA/Bloomberg energy numbers she quotes from the press rather than from the underlying models. Notably absent from her input diet: frontier-lab technical papers, alignment literature, productivity economics.

Sees

A seat inside Microsoft Research during the largest infrastructure build in the company's history — the closest any critic on the bench sits to the hyperscaler ledger — plus curatorial access to the museums and biennales that now set the cultural frame for "what AI is." What she says she cannot see: the sites themselves — "Their status as high-security sites impacts our ability to visit these facilities and do the kind of fieldwork" (Harvard Design Magazine, 2026-05-20).

Incentive map

Re-derived this cycle, from income sources rather than carried forward. Microsoft pays her salary and hosts her group; Microsoft's 2025 emissions rose 25% on data-center buildout and it is contracting gas power (Fortune, 2026-07-09) — she cannot cheaply say "my employer's data centers are the problem," and the in-window record shows she names competitors instead (observable above; count it every cycle). USC and Sloan (Knowing Machines) are low-conflict academic money. The art market and prize circuit are new: six editions of Calculating Empires, a museum purchase, the EU-funded S+T+ARTS Grand Prize (2024), the Boghossian Prize (2024) and the Silver Lion (2025) reward the monumental, historical register — which is also the register least likely to produce a specific regulatory proposal, the bench's standing weak spot. European Parliament STOA panel: an "advises government" line, logged here as a conflict first and a credential second — the counterparty's live agenda is the AI Act's phased application (general application 2 August 2026, Art. 113). Paid keynotes via a speakers' bureau (listed as an exclusive Lavin speaker in this review's search results; not independently confirmed). What she can't say: that the productivity side of the ledger might justify the costs (it would dissolve the frame); that Microsoft's efficiency numbers (WUE 0.27) matter more than its totals; that x-risk is the priority. Rule-25 discount: moderate on numbers she quotes from the press (two drifted, one wrong this cycle), low on her documentary work (diagrams, dataset forensics), and a specific discount on any silence about Microsoft.

Theories aligned with

What she's reacting against

Where she overlaps / splits (with Rich)

Track record

Empirical vs normative

Weak spots / open questions

Rich's take

Delta log

2026-09-21 — v1→v2 migration + validation (bulk run)

Sources

migrated v1→v2 2026-09-21 · backup: _archives/kate-crawford.html.bak-20260921 · converts-from: people/kate-crawford.md · AI & Society domain