In 1968, Washington and philanthropy funded a nonprofit to build a television show for preschoolers; Sesame Street helped children stay on grade level for about $5 a child a year. AI could do the same for learning, but markets pay for what people reach for in the moment. Congress should fund competing nonprofit builders, require independent tests, share what works, and let proven teams earn a charter.
The Problem
A quarter of American teenagers now use ChatGPT for schoolwork, double the share a year earlier.1 Design decides whether tools like it build skills or replace them. In one randomized experiment with nearly 1,000 high school math students, a plain GPT-4 chatbot raised practice grades 48%, but once it was taken away, those students scored 17% worse than peers who never had it; a version designed as a tutor largely avoided the loss.2
A tutor judged by what a student can do without it serves a public interest no single buyer pays for. America has funded such technology before: the Children's Television Workshop, a nonprofit founded in 1968, was built on $8 million from the Carnegie Corporation, the Ford Foundation, the U.S. Office of Education, and the Corporation for Public Broadcasting.3 Preschoolers who could receive its Sesame Street were more likely to stay on grade level, especially in poorer counties, at about $5 per child per year in today's dollars.4 Three gaps keep AI's version from being built:
- Grants stop at the paper. NSF's 29 AI Institutes, funded at about $20 million each over five years, invest in "the foundational science behind AI."5 A paper about a promising tutor is not a tutor a school can use.
- Nonprofits rarely reach scale. Wikipedia draws nearly 15 billion views a month, and the Wikimedia Foundation says it is the only top-10 website hosted by a nonprofit.6 It can be done; it rarely is.
- No long-term home. The government sponsors 41 federally funded research and development centers for long-term needs; 26 serve the Energy and Defense departments, and none is sponsored by the Education Department.7
Why legislation: No law is needed to start a nonprofit; law is needed to fund one fairly and hold it to evidence. Existing vehicles fit poorly: NSF's AI Institutes are defined in law as research institutes,8 and federally funded research centers exist to serve a sponsoring agency's own mission.9 Congress has built nonprofits for public missions before, including the Patient-Centered Outcomes Research Institute, which funds peer-reviewed medical research,10 and the Foundation for Food and Agriculture Research, which may spend federal money only when private donors match it.11 When a technology shapes how children learn, the public has a stake in building the version that leaves them more capable.
The Solution
A four-step staircase: each step stands alone, and each step up adds public commitment and accountability. Awards go by open competition to independent nonprofits, never by earmark to a named organization, and every project needs a school, employer, clinic, or agency ready to use what it builds.
Step 1 — Open a competition. NSF, with the Education Department and NIH, funds nonprofit teams that pair engineers and designers with independent researchers and a delivery partner. Each team names one human outcome that matters outside the product, such as unaided reading or math, and each federal dollar is matched by a private one, the rule Congress set for agricultural research's nonprofit foundation.11
Step 2 — Test before scaling. Before the main trial, each team publishes its comparison group, primary outcome, possible harms, costs, and decision rule, and follows participants long enough to tell a crutch from a skill. Independent evaluators, not the product team, judge the results; null results are published, and donors get no veto. A tutor should be judged by what a student can do after it is gone.
Step 3 — Share what public money builds. Software, methods, and de-identified data built with federal funds carry open licenses and secure research access, so schools and companies can adopt what works. Each funded service needs a plan to keep running or to hand off to a district, agency, or other operator.
Step 4 — Let proven teams earn a charter. After five years, Congress may charter institutes with independently replicated results as standing centers modeled on federally funded research and development centers: chosen competitively, bound to "operate in the public interest with objectivity and independence," and renewed only after a review at least every five years.9 Scale should follow evidence; a nonprofit label is not evidence.
Where to start: Step 1 is the floor: a modest, competitive grant program. Step 2 is the heart of the proposal.
Administration and enforcement: NSF runs the competition with the Education Department and NIH and publishes selection criteria within 12 months; Congress appropriates a staged, five-year demonstration budget. Grant audits, milestone reviews, conflict-of-interest rules, and recovery of misused funds enforce the terms, and any charter or endowment requires its own act of Congress.
Risks and Mitigations
- Cronyism and capture: Open competition, published criteria, outside reviewers, recusals, and disclosed donor interests guard the awards, and no organization is named in law. Well-connected applicants will still write better proposals, which is why renewal depends on independent results.
- Crowding out the market: Funding is limited to outcomes that buyers underpay for, and open licenses let companies adopt what works. Some funded tools will compete with commercial products, and that tension remains.
- Duplication: Applicants must show why an existing university, company, or nonprofit cannot do the work; if one can, fund that route instead. The review burden is real but modest.
Similar Bills
Fit measures similarity to this proposal's mechanisms: High = direct precedent; Partial = useful component with material differences; Related = adjacent approach.
Federal
| Proposal or bill | Relevant provisions and fit | Fit |
|---|---|---|
| National AI Research Institutes, 15 U.S.C. §9431 Enacted Jan. 1, 2021 (P.L. 116-283, Div. E, §5201) |
Authorizes NSF and other agencies to fund AI research institutes focused on sectors including education and health, with required public-private partnerships. Precedent for Step 1's competitive, partnered awards; research institutes rather than builders with delivery duties. | Partial |
| Patient-Centered Outcomes Research Institute, 42 U.S.C. §1320e Enacted Mar. 23, 2010 · Amended Dec. 20, 2019 |
A statutory nonprofit, "neither an agency nor establishment" of the government, that funds peer-reviewed comparative outcomes research. Governance model for Steps 2 and 4; health research rather than technology building. | Partial |
| Foundation for Food and Agriculture Research, 7 U.S.C. §5939 Enacted Feb. 7, 2014 (P.L. 113-79, §7601) |
A nonprofit given $200 million in federal funds that it may spend only when matched by equal non-federal money. Direct model for Step 1's match; agricultural research. | Partial |
| Foundation for the National Institutes of Health, 42 U.S.C. §290b Enacted Nov. 16, 1990 (P.L. 101-613) |
A statutory nonprofit that raises private money to support NIH's mission. Public-private model for Step 1; supports an agency rather than independent builders. | Partial |
State
| Proposal or bill | Relevant provisions and fit | Fit |
|---|---|---|
| New York — Empire AI (A8808C, Part TT) Chapter 58 · Signed Apr. 20, 2024 |
Economic Development Law §361 creates a state-owned AI research and computing facility at SUNY Buffalo, run by a nonprofit consortium, for "ethical and public interest uses" of AI. Shared-infrastructure analogue for Step 3; research capacity rather than builders of public tools. | Related |
| Massachusetts — Chapter 238 of 2024, item 7002-8070 Approved in part Nov. 20, 2024 |
Capital grants through the Massachusetts Technology Park Corporation for applying AI to "public policy problems" and state industries. Public-funding analogue for Step 1; no independent-evaluation or publication duties. | Related |
| California — SB 53, CalCompute (Gov. Code §11546.8) Chapter 138 · Approved Sept. 29, 2025 |
Creates a consortium to design a public computing cluster for "research and innovation that benefits the public," operative only upon appropriation. Infrastructure analogue for Step 3; not an outcomes institute. | Related |
What this adds: Existing law funds AI research institutes and public-private foundations for health and agriculture. This proposal adds a competitive, privately matched program for nonprofits that build and maintain technology, judged by independent evidence of human outcomes, with a path to a chartered center only after results replicate.
Notes
-
Olivia Sidoti, Eugenie Park, and Jeffrey Gottfried, "About a Quarter of U.S. Teens Have Used ChatGPT for Schoolwork – Double the Share in 2023," Pew Research Center, January 15, 2025. Self-reported; survey of 1,391 U.S. teens ages 13–17, September–October 2024; 26%, up from 13%. ↩
-
Hamsa Bastani et al., "Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics," Proceedings of the National Academy of Sciences 122, no. 26 (2025). One field experiment with two GPT-4 tutors; results are specific to its tools and setting. ↩
-
Joan Ganz Cooney Center at Sesame Workshop, "Joan Ganz Cooney," accessed September 2026. The Carnegie Corporation "partly" financed the project; the rest of the $8 million came from the other three funders. ↩
-
Melissa S. Kearney and Phillip B. Levine, "Early Childhood Education by Television: Lessons from Sesame Street," American Economic Journal: Applied Economics (January 2019). Figures from the authors' NBER Working Paper 21229 (revised 2016): gains in grade-for-age status, strongest for boys, Black children, and poorer counties; long-run effects inconclusive. ↩
-
U.S. National Science Foundation, "National AI Research Institutes," accessed September 2026. ↩
-
Wikimedia Foundation, homepage, "By the numbers," accessed September 2026. The foundation's own figures. ↩
-
National Center for Science and Engineering Statistics, Master Government List of Federally Funded R&D Centers: FY 2026, current as of February 2026. Sixteen are sponsored by the Energy Department and ten by the Defense Department. ↩
-
15 U.S.C. §9431(b)(2), defining an institute as "an artificial intelligence research institute." ↩
-
Federal Acquisition Regulation §35.017(a)(2): FFRDCs perform tasks "integral to the mission and operation of the sponsoring agency" and must "operate in the public interest with objectivity and independence." §35.017-1(e) limits sponsoring agreements to five years, renewable after review. ↩ ↩2
-
42 U.S.C. §1320e, establishing the institute as a nonprofit corporation with a peer-review process for primary research. ↩
-
7 U.S.C. §5939(g): the foundation may use its $200 million "only to the extent that the Foundation secures an equal amount of matching funds from a non-Federal source." ↩ ↩2