There is a question that I happened to ask when I met high school students in the last two years. “Do you use ChatGPT to do homework?”. Yes, they answered me in most cases. Not only to write essays, but also to ask for explanations of difficult concepts or as a simple assistant to offer study advice. According to a research conducted by TGM Research on behalf of NoPlagio.it, in fact, AI in school is now a reality. Of the sample of just over a thousand students interviewed, 65% use it for homework and essays, 71% look for information, 60% use it to do homework, 33% to learn, 18% to answer tests, 21% as a personal assistant and 13% specifically to write essays.

In the year of the ban on smartphones in elementary and middle schools, artificial intelligence will be one of the main innovations of the new school year. The Minister of Education Valditara has in fact announced an experiment that will involve four regions, for a total of 15 classes, between middle and high schools. The idea is to use AI as a sort of tutor, able to help students starting from the history of classwork and tests.

The experiment will last two years and has the goal of understanding whether artificial intelligence can be an effective tool in school. From a technical point of view, it involves the use of software on Google Workspace. At the center, at least at the beginning, will be STEM subjects: scientific ones, to be clear. In concrete terms, the system should be able to identify students’ gaps, warn the teacher and propose personalized explanations and exercises.

How AI-Powered Tutors Work

AI-based tutors are one of the most discussed fields of application of this new wave of generative AI. The premises are there: the idea of ​​a personalized assistant that can support professors and students actually seems promising. Salman Khan, founder of the non-profit Khan Academy, recently predicted a future in which every student could have a personalized assistant at their disposal.

There are already some on the market. There are three applications that are more well-known than the others, also available on the Italian market. They are called Question AI, Answer.AI and Gauth: they allow you to photograph your homework to receive customized explanations. They are all applications of Chinese origin: a consequence of Beijing’s crackdown on the ed-tech sector on for-profit tutoring services, which has pushed companies towards internationalization. In the United States, there is also a Google product, called Socratic. The operation is more or less the same: you ask a question (even just by photographing an operation) and the system produces an answer, explaining the necessary steps step by step.

The Italian one is one of the first systematic experiments in the world of this technology. In an article published in the Harvard Gazette, a study conducted by Gregory Kestin and Kelly Miller on a physics course at Harvard is described, in which the use of an AI tutor surpassed the effectiveness of traditional teaching. On a sample of 194 students, the study revealed that those who used the AI ​​tutor made double the progress compared to their classmates who were followed by a teacher in class.

The risks: from hallucinations to the disappearance of the relationship

The truth today is that many of these experiences are mainly anecdotal. The effects of using artificial intelligence in learning are unknown. In one of the most cited scientific articles on the subject, a review of studies on the topic, it is noted that, according to the scientific literature, AI for education can have a positive effect. However, at the moment there seems to be a lack of reflections on pedagogical and ethical aspects. Many studies – say the researchers – focus too much on technical aspects (such as data analysis and the construction of predictive models) and neglect the educational side, that is, how these technologies can actually improve teaching and learning. Simply put: there is a lot of work being done on technology, but little on how to use it effectively for teaching.

Staying with the technology, AI tutors also have risks. In an article published in the New York Times, Ben Williamson, a fellow at the Centre for Research in Digital Education at the University of Edinburgh, points out that these generative tools could even have negative or “degenerative” effects on student learning. “There’s a rush to proclaim the authority and usefulness of these chatbots, but the evidence to support those claims is not there yet,” Williamson said.

The main risk is the rather well-known one of so-called hallucinations. We do not know how the system that the Ministry of the Interior will adopt works, but most AI tutors are built on systems that are very similar to ChatGPT. Which, often, tends to invent, to take for granted facts that are not certain.

The other risk is that of substitution. True, the Ministry has clarified that it will be a support. But what happens to the relationship between student and teacher when there is a system that responds 24 hours a day, provides simple explanations and does not ask for anything in particular in exchange? The issue, more cultural than anything else, is that it becomes a question of efficiency: that we can, in short, optimize the relationship with teachers, make it superfluous.