Raw capture — raw/private-capture-of-public-genius.md
url: https://www.wysr.xyz/p/the-private-capture-of-public-genius title: The Private Capture of Public Genius author: Cameron Russell Armstrong date_fetched: 2026-07-08 date_published: 2026-07-02
By Cameron Russell Armstrong, published in Wysr, July 2, 2026. Essay I of "Upstream of Everything."
The essay opens with AT&T's 1956 consent decree, which forced the company to license all 7,820 unexpired patents royalty-free. Armstrong notes that AT&T was then the largest private company in the world, employing 746,000 people, with Bell Labs having already produced the transistor, solar cell, information theory, and radio astronomy.
Intel co-founder Gordon Moore is quoted describing this decree as one of the most important developments for the commercial semiconductor industry, saying it "allowed the merchant semiconductor industry to really get started in the United States." Armstrong argues this "started the growth of Silicon Valley."
The author likens AT&T's regulated monopoly to a carefully engineered rice paddy. Because AT&T's profits were tied to its capital base (through a regulated ~7% return), "Every dollar spent on research at Bell Labs did two things at once" — it was a recoverable cost and it expanded the rate base, creating a self-reinforcing cycle of investment.
Armstrong argues that frontier AI models are trained on "the collective experience, knowledge [and] learnings of humanity" (quoting Sam Altman). The essay describes this as "the private capture of public genius" — a compression of massive amounts of publicly generated data into private model weights.
The author reports that Anthropic's annualized revenue grew from $87M in January 2024 to $47B by May 2026, calling it "the fastest-compounding enterprise software company in history."
The essay categorizes goods by excludability and rivalry: private goods, club goods, common-pool goods, and public goods. The author argues the frontier labs treat internet data as a public good via fair use claims, but contends that "granting access is not the same thing as giving license."
Key rulings discussed: - Bartz v. Anthropic (June 2025): Judge Alsup ruled training on legally acquired books was "quintessentially transformative" but training on pirated books was not. Anthropic settled for $1.5B. - Kadrey v. Meta: Judge Chhabria found LLM training transformative but criticized plaintiffs for insufficient evidence of market harm. - The U.S. Copyright Office issued a 2025 non-binding report concluding public availability doesn't inherently allow fair use training.
The author contrasts human reading with AI training: "A person who read ten thousand books" returns sediment to the cultural delta slowly; a model "becomes a printing press that prints more printing presses" at industrial volume. The essay traces the origins of modern copyright to the Statute of Anne (1710), which followed the Gutenberg press by 270 years, warning that today's reckoning won't allow such a delay.
Armstrong argues the labs' claim that individual training inputs are "worthless" (since removing any single one barely affects the model) is doublespeak. Shapley value — the leading method for valuing training inputs — is described as computationally infeasible at frontier scale. The author concludes: "you cannot pay people in proportion to their contribution because no administrable, specific share exists." The conclusion drawn is that payment is still owed, just not proportionally.
The central policy proposal: frontier labs pay "a fixed share of gross revenue into a public fund" distributed equally to every eligible American. The author argues this avoids the measurement problem that would "die ten thousand quibbling deaths in the courtroom."
Analogies offered include: - Alaska Permanent Fund (distributing resource wealth no individual can separately claim) - Superfund (holding polluters accountable even when legal at the time, without tracing each barrel)
The royalty is described as "restitution" — not welfare, charity, or a tax. The author states: "This is the private capture of public genius at civilizational scale."
Drawing on Elinor Ostrom's eight conditions for sustainable commons governance, the essay notes the internet satisfies almost none of them: "Its boundaries are hazy, fueled by everyone and fenced by no one." The author warns of an "attribution collapse" and a "cyberindustrial runoff disaster poised to spoil the entire delta."
The piece closes with a note that the royalty is one piece of a larger system and doesn't replace copyright claims or private licensing where ownership is clear. The final framing: frontier labs seek "the privilege and power of a utility without accepting the public obligations that come along with it."