The Department of Labor just rolled out a free seven-day, text-message AI literacy course called “Make America AI-Ready,” built with private partner Arist. A team of researchers including Arvind Narayanan and Hilke Schellmann put it under the microscope, and their verdict, published on Hacker News, lands somewhere between cautious approval and a punch list of fixes.
This is significant because it’s one of the first concrete moves from the Trump administration’s AI workforce agenda. The pitch: 10 minutes a day, SMS delivery, no app downloads, meant to push AI fluency to anyone with a phone. The reviewers call the design choice smart. It meets people where they are, skips the friction of account creation, and the pacing actually fits into a workday.
What the course gets right
The reviewers flag four genuine strengths:
- Accessibility through SMS. Maximizes reach to non-technical Americans.
- Verification first. The course hammers home that AI output must be checked, using sticky examples like looking up a restaurant only to discover it’s now a nail salon.
- Human accountability. A quiz scenario about a coworker submitting AI-generated fabricated stats puts responsibility squarely on the human. The point gets repeated throughout.
- Honest about limits. Hallucinations, training cutoffs, and the “AI predicts rather than knows” framing all show up. For a 101 course, the calibration is fair.
Where it contradicts itself
The sharpest critique is on privacy. The course closes with stern warnings: never share passwords, Social Security numbers, medical records, confidential work data, or income data with AI tools. The problem? Earlier lessons told users to do exactly that.
- Day 3 prompts users to upload photos, PDFs, or voice recordings.
- Day 4 calls it a “power move” to paste your resume and share monthly expenses.
- Day 5 suggests typing medical symptoms into AI to prep for doctor visits.
- Day 6 tells users to share their address to find nearby restaurants.
The reviewers don’t dismiss this as sloppy writing. They argue it exposes a real tension: AI gets more useful the more it knows about you, so blanket prohibitions don’t work. The fix they recommend isn’t a stricter rule. It’s more nuance, moved earlier in the course, with practical guidance on incognito chats, privacy settings, and threat models like prompt injection and training-data risks.
The pedagogy problem
The quizzes get flagged too. Questions about AI’s failure modes pair one obviously correct answer with absurd strawmen like “AI likes making things up to test you.” The reviewers say this kills the critical thinking the course claims to build. Better prompts would be open-ended: what do you do when your employer mandates AI tools that monitor your productivity, or how do you find out whether AI is scoring your loan application.
Why this matters now
Government-backed AI literacy is about to become a category. Other agencies and other countries will copy this template, which means the design decisions baked in here will shape how millions of workers first encounter AI. A course that quietly trains users to paste sensitive data into chatbots, then warns them not to on the last day, sets a bad default. So does a quiz format that rewards memorizing the “right” answer rather than reasoning about tradeoffs.
The reviewers also note something the Department of Labor of all agencies should have led with: the course barely addresses how AI is reshaping work itself. Monitoring, hiring algorithms, productivity scoring, displacement. For a workforce-readiness program, that’s a strange gap.
Practical takeaways
- For policymakers: AI literacy programs need privacy guidance grounded in threat models, not blanket prohibitions that the curriculum itself violates.
- For employers building internal training: Skip the right-wrong quiz format. Adults learn faster from scenarios with real tradeoffs.
- For practitioners: If you’re advising on AI curriculum, push for incognito-mode education and workplace-policy awareness before you teach prompt techniques.
The Make America AI-Ready course is a useful start, as the reviewers put it. The question is whether the 201 version closes the gaps or whether the 101 template gets cloned with its contradictions intact. Full breakdown at the original source.