Education Put to the Test by the Machine

The educational field has long been fond of technology. Think of overhead projectors, the Internet, or videoconferencing. Moreover, research on AI in education (AIED) began more than 50 years ago. But it was in the 2010s that AI-driven personalized learning truly took off. Khan Academy, for instance, leveraged large volumes of learning data to adapt content to individual needs.

What’s new today is that the most widely used tools - ChatGPT, Copilot, Gemini, among others - are not specifically educational. And yet, they are flooding classrooms on a massive scale, often without clear guidelines or supervision.

It is risky to avoid the subject, because generative AI, whether we like it or not, is infiltrating everything. It powers search engines, sorts our emails, corrects our texts, recommends content, and predicts our behaviors. Education, in particular, is being put to the test because the use of generative AI carries the risk of harming learning. In addition, its omnipresence is discreet. In many cases, the contribution of a generative AI tool to a student’s work escapes any reliable detection. Some institutions are considering reinforcing control mechanisms - at the cost of a “pedagogy of suspicion,” which runs counter to any educational relationship based on trust and responsibility.

It’s clear that binary responses “allow or forbid” don’t work. We cannot allow practices that undermine learning and the development of cognitive skills, but we also cannot prohibit something that has become ubiquitous. Teachers and policymakers must walk a tightrope.

When it comes to generative AI in education, everything depends on how it is used, for what kind of learning, and in what context. Let’s take ChatGPT, since it is the most widely used tool. A true cyber Swiss Army knife, it promises a thousand uses. Originally designed to simulate human conversation, it can summarise a text, translate it, provide feedback, create exercises, ask questions about a course, solve homework problems, write code, or generate fictitious data. Some of these uses may be beneficial in certain contexts, others detrimental. Like a Swiss Army knife, ChatGPT is neither good nor bad in itself. From a purely educational standpoint, both praise and distrust depend on what one chooses to look at.

Only if they choose to. To be “doomed” or “destined” implies having no choice, but that’s not the case here. No one is condemned to rest on their laurels. However, learning does not happen without effort. One does not become a lawyer, doctor, or mathematician without long, sometimes arduous practice of one’s professional skills. Generative AI can be used as a lever for efficient learning, but using it as a substitute for effort can only stunt the learning process. Down that path, the outlook darkens. Most of our students genuinely want to learn - we must help them use generative AI wisely.