Universities across the country are racing to make their students "AI fluent." Ohio State promises every graduate will be fluent in AI applications in their field. Other schools are rolling out AI literacy programs, frameworks, and requirements.

There's just one problem: nobody actually knows what AI fluency means yet.

The rush to teach what we don't understand

According to Justin Reich from MIT, we're seeing a familiar pattern play out. A new technology emerges, experts hastily define skills around it as a "literacy," and schools rush to teach it before anyone really understands what needs to be taught.

We've been down this road before with tablets, smartphones, and social media. The results? Students who can't organize files on their computers. Teens who scroll mindlessly for hours. High schoolers who can't distinguish credible sources online.

The pattern is clear: when we rush to teach technology skills before understanding them, we fail students.

The real challenge: evaluation, not generation

Here's what makes AI different and harder to teach: making AI generate content is easy. The real challenge is knowing when that content is good or garbage.

Can students distinguish between a hallucination and a genuinely useful insight? Can they evaluate AI output critically? Can they spot when the model has confidently delivered complete nonsense?

Reich argues that the students best prepared to evaluate AI output won't be the ones with "AI fluency training." They'll be the ones with deep knowledge in their actual disciplines – exactly what colleges have been trying to build for centuries.

Think about it: you don't need special AI training to know if a historical claim is accurate or an engineering calculation makes sense. You need to actually understand history or engineering.

What works: leading with uncertainty

Instead of rushing to proclaim what AI literacy is, Reich suggests we should lead with our uncertainty.

We should experiment. We should explore where AI proves genuinely useful in different fields. We should figure out together – educators and students – what practices actually help versus harm learning.

Most importantly, we should avoid the trap of confusing speed with learning. Yes, AI can make students produce work faster. But does faster mean better? Does it mean they've actually learned anything?

The Flintolabs approach: learning by doing, not by decree

At Flintolabs, we focus on what we know works: hands-on experimentation with real projects. Our students don't memorize frameworks about AI – they build actual applications, make mistakes, debug problems, and figure out through experience where AI helps and where human thinking is irreplaceable.

When you build a face filter app, you learn computer vision by wrestling with it, not by studying someone's definition of "AI literacy." When your chatbot says something weird, you learn about the limitations of natural language processing firsthand.

This approach embraces uncertainty. It treats AI as a tool to explore, not a subject to master according to some predetermined checklist.

The bottom line

Any program or school claiming they've definitively figured out "AI fluency" is, frankly, pretending. The technology is changing too fast. The research on how to teach with it effectively is still emerging. The very definition of what students need to know keeps shifting.

What we can do is create environments where students experiment safely, build real things, develop critical thinking, and gain deep domain knowledge in their fields. These are the foundations that will serve them regardless of how AI evolves.

The future doesn't belong to students who completed an "AI fluency" checklist. It belongs to those who learned how to think critically, evaluate carefully, and adapt continuously – with or without AI.

References

Reich, Justin. "Stop Pretending You Know How to Teach AI." The Chronicle of Higher Education, November 2025, https://www.chronicle.com/article/stop-pretending-you-know-how-to-teach-ai