Jaron Lanier

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

Who he is

Affiliation: Prime Unifying Scientist ("Microsoft's Octopus"), Office of the Chief Technical Officer, Microsoft — role check run 2026-09-21 against Microsoft's own people page, which carries that exact title; the office he sits in still exists and its head, CTO Kevin Scott, was testifying under that title on 2026-05-13 (GeekWire). Wikipedia's timeline (secondary): joined Microsoft 2006, Microsoft Research from 2009. No job change, no departure, no new employer found in a dated search of the window (2026-06-16 → 2026-09-21). Founder of VPL Research (early 1980s; bankrupt 1990) — the company whose team built "the first implementations of multi-person virtual worlds" (late 1980s, per Microsoft's page). Author of You Are Not a Gadget (2010), Who Owns the Future? (2013), Dawn of the New Everything (2017), Ten Arguments for Deleting Your Social Media Accounts Right Now (2018). Musician (Philip Glass, Jon Batiste, T Bone Burnett per his agency bio). Born 1960-05-03; lives in Berkeley.
Other hats found this cycle (v1 listed none): (1) elected to the Authors Guild Council on 2023-04-05 — the Guild sued OpenAI (Sept 2023) and added Microsoft as a defendant (Dec 2023); on StarTalk (May 2026) he put it himself: "I'm both on the board of the Authors Guild and I'm a prime scientist at Microsoft, so I sued myself... I call it capitalist yoga." The Guild's Council page as fetched 2026-09-22 lists his whole 2023 cohort (Bowen, Castellani, Gay, Jacob, Maum, Sealey) but not him — see Incentive map for how this is handled. (2) Co-author of a Microsoft research paper on training-data attribution (arXiv, 2026-05-07). (3) Attached to Natasha Lyonne and Brit Marling's AI-made feature Uncanny Valley (Asteria studio; announced 2025 per his agency bio — the Hollywood Reporter primary was paywalled at fetch, so the exact date is unverified). (4) Working New Yorker essayist: "There Is No A.I." (April 2023), "How to Picture A.I." (2024), "Where Will Virtual Reality Take Us?" (2024), "Your A.I. Lover Will Change You" (2025-03-22).
One-line position: The people whose data trains AI should be paid for it — "data dignity" is the structural fix for an economy where tech companies extract value from human creativity and labor without compensation, and AI makes this extraction faster and more total. The 2023–26 restatement sharpens it: there is no A.I. — a model is "a collaboration of the people whose actions and data are combined to create it" (Existential Hope, 2026-05-13), so the black box is opened by revealing the people, and paying them.

Discipline & technical bet

Computer scientist by training and temperament, public intellectual by output. For a decade the bench could file him as a purely normative voice — an economist of data with no engineering wager. That changed in the window. TechCrunch reported (2025-03-21) a Microsoft research initiative on "training-time provenance" — training networks so their outputs' sources "can be efficiently and usefully estimated" — with Lanier "reportedly involved." On 2026-05-07 he appeared as third author on "DataDignity: Training Data Attribution for Large Language Models" (Li, Banburski-Fahey, Lanier; all Microsoft). The paper frames the problem as "pinpoint provenance": given a prompt and a model's answer, rank the documents that best support the answer. It ships a benchmark (FakeWiki, 3,537 fabricated Wikipedia-style articles) and two methods; the supervised one lifts Recall@10 (how often the true source lands in the top ten guesses) from 35.0% to 52.2% across nine open-weight models. The paper's own limits: document-level only, not sentence-level; a fabricated corpus, not a production training set. So the concrete technical bet is now explicit: attribution can be made cheap enough to hang a payment system on. Half right today — it works on a toy corpus, one time in two. That is the number to watch, and it is his number.

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

A builder's grief: he helped make the thing and watched it become "the behavior mod machine." Wants to be proven right that the fix is economic, not spiritual — that people can be paid rather than pitied. Fears a world of "good pets" on basic income more than he fears a superintelligence; the Stalinist line is the tell.

Hears

Microsoft's Office of the CTO (Kevin Scott) and Microsoft Research colleagues (Weyl's Plural Technology Collaboratory; the Li–Banburski-Fahey attribution group); the Authors Guild and working writers; Bay Area musicians and the Berkeley humanities circuit (Townsend Center, CDSS). Downstream of Norbert Wiener on cybernetics (his own citation, StarTalk), Ted Nelson on two-way links (a debt he has long acknowledged; not re-sourced this cycle), and his own 2006 "Digital Maoism." Almost no exposure to the alignment or scaling communities on their own terms.

Sees

Two things nobody else on the bench sees at once: the inside of the company that is both a lead defendant in the training-data litigation and the operator of a licensing marketplace — and the inside of the plaintiff class. He knows what attribution actually recovers because his co-authors measured it (one time in two, at document level), and he knows what authors actually get because he watched the $3,000 checks get contested.

Incentive map

Re-derived from scratch this cycle; v1's single line ("working at Microsoft creates a tension") was correct and far too small.

Theories aligned with

What he's reacting against

Where he 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/jaron-lanier.html.bak-20260921 · converts-from: people/jaron-lanier.md · AI & Society domain