The Only Question Teachers Should Ask Students in the Age of AI Remains the Same as Ever: “What Have You Learned?”

Reading a recent New York Times article about the nascent industry of humanizers and autotypers, “Student cheating is becoming impossible to detect in an AI era”, my response as an academic of more than three decades standing was not so much outrage as embarrassment. We’ve always known that some students will try to beat the odds, and paradoxically this has been made easier by schools and universities having allowed the perfect ecosystem to evolve: teachers obsessed with detecting texts written with AI, students trying to hide it, and companies happy to supply both sides. A perfect market for educational paranoia.

It’s no exaggeration to say that things have gotten out of hand: tools that rewrite an AI-generated text to make it look more human; programs that insert words into Google Docs one at a time, with pauses, typos and false corrections, all to make it look like the student was really writing. Apps that promise “credible bugs.” Detectors that claim to preserve academic integrity while offering, in the same product or in the adjacent ecosystem, the possibility of generating complete works. As one executive quoted by the NYT says, “bigger cat, bigger mouse.”

It’s not that AI has destroyed education: it has simply exposed many of its fictions. For decades we evaluated final products as if they were clear evidence of learning: an essay, a commentary, a report, a work. Whether these “tests” could be done by another classmate, by an older brother or by a brilliant student in Kenya, India or Nigeria did not matter. But if a machine can generate that product in seconds, the obvious question is not how we detected the machine, but why we have continued to evaluate in the same old way. AI hasn’t broken traditional assessment: it has revealed how fragile, routine, and easily automatable it has been for many years.

I’ve discussed the irresponsibility of sanctioning students based on using AI detectors, pointing out that OpenAI has withdrawn its AI classifier due to its low accuracy rate. Similarly Stanford HAI has documented biases against non-native English writers. Trying to turn an algorithmic probability into a disciplinary accusation is no more than pseudoscience with a nice interface. And now, in addition, we know that a whole industry is growing around these detectors to circumvent them. Is this the future we want for our education system? Or, as many educational institutions do, pretend that it doesn’t exist?

The only relevant question a teacher should ask a student is “what have you learned?” Students can use AI to copy without understanding the question they’ve been asked, just as they used to be able to copy from Wikipedia or from a classmate. But AI can also be used to ask for alternative explanations, practice, translate, revise arguments, generate examples, simulate a Socratic tutor or receive immediate feedback. Use does not define learning. It is defined by understanding, criteria, the ability to defend an idea, to contrast sources, to recognize mistakes and to improve.

Therein lies the enormous opportunity that we are wasting. AI has huge potential to personalize education to limits that we have never been able to reach. Each student learns at different paces, with different fears, with different abilities, with different family, linguistic and cognitive contexts. The traditional classroom has treated that diversity with a mixture of goodwill and insufficient resources. Well-managed AI can offer tailored explanations, unlimited practice, 24×7 availability, and non-judgmental interactions. For many students, being able to ask the same thing ten times without feeling embarrassed is not a luxury: it is a revolution.

This connects directly with CAST’s Universal Design for Learning: offering multiple ways to access, express, and engage with knowledge. AI can turn a complex text into a step-by-step explanation, into audio, into visual examples, into conversation, into graded exercises, or into one-on-one tutoring. It can help students with dyslexia, anxiety, high abilities, language difficulties or previous gaps. It does not replace teachers; it allows them to stop acting as a repeater of content and focus on what is truly human: criteria, accompaniment, context, ethics, demands and conversation.

Of course, this requires rules. UNESCO calls for people-centered policies. The US Department of Education has insisted on integrating AI with transparency, security and accountability. This isn’t about letting a machine decide for us or handing over student data to any platform, but nor is it about imposing blanket bans; responsible governments need to lay out clear guidelines on the uses and limits of AI.

The logical way forward is to rethink how we assess students’ performance. We need to think about the process, not just the end result. Prompts, drafts, iterations, discarded decisions, verified sources, oral defense, application to real cases, metacognitive reflection. In short, evaluate how to work with AI, rather than wishing it didn’t exist. Students need to be taught how to quote it, question it, correct it and use it without delegating thought. Anything else is simply encouraging students to learn to cheat better.

If using AI continues to be a cat-and-mouse game, teachers and students will both lose. What’s more, we will have wasted technology capable of making education more personalized, more accessible and fairer. AI should be the best news that education has received in decades. We still have time to stop treating it with suspicion, and instead start using it to improve how we teach and how students learn.

(En español, aquí)

—

This post was previously published on MEDIUM.COM.

—