Can we learn with artificial intelligence?

There are 54 people divided into three groups, all given the same task: to write texts based on a set of common prompts. However, the first group is allowed to use artificial intelligence systems, whilst the second group can rely solely on online search engines, and the third group, the control group, must carry out the task without any digital aids. 

As a result, Although the essays produced are of good quality, the members of the group using AI show the lowest levels of brain activity recorded, and the participants struggle to recall the content of their own work. Furthermore, when asked to write a new text without the aid of artificial intelligence, the participants’ brain connectivity remains significantly weaker than that initially observed in the control group. According to the MIT researchers who conducted the experiment, once the help of AI was removed, these participants were unable to fully recover the patterns of brain activity that should characterise deep, original thinking.

"When we completely delegate a task to generative artificial intelligence systems, we lose contact with the ‘cognitive friction’ necessary for learning something new", explains Aldo Gangemi, a professor in the Department of Philosophy at the University of Bologna. "This cognitive friction is the obstacle to overcome that helps us understand and memorise new notions, and excessive use of generative artificial intelligence can make it disappear".

Desirable difficulties

Learning, in short, requires a certain amount of effort. And an AI system that removes every difficult step ends up leaving little room for acquiring new competences. “Several studies on learning highlight the paradox of ‘desirable difficulties’”, adds Chiara Lucifora, professor at Link Campus University. “These are conditions that make it harder to learn new things in the short term, but that over time lead to deeper, more solid knowledge: however, using generative artificial intelligence systems makes it possible to bypass these valuable obstacles”.

How, then, might an AI behave if it were designed not simply to provide answers, but to help people learn new things? Lucifora and Gangemi tried to imagine it, and propose a model (available online) in which AI becomes a collaborative tool, helping to pinpoint the problem to be solved and to co-create the most suitable solutions

"The goal is to arrive at an artificial intelligence that reasons together with us, instead of simply delivering the final answer", explains Gangemi. "A system of this kind is designed to stimulate reasoning by engaging in a dialogue with the user, to help them clarify and develop the idea they are working on”.

School and AI

The school and training environment is naturally the most suitable for the application of similar tools. And experiences in this field are multiplying. One example is Proffilo, which uses mini-games to assess students’ skills, thereby helping teachers to plan appropriate learning activities. Or there is MATHia, an AI-based maths tutoring programme for secondary school pupils.

"There is room for growth between what a person can do on their own and what they can learn with the right support", explains Martina Benvenuti, professor in the Department of Psychology at the University of Bologna. "And it is in this context that an AI system could be most useful: by offering a clue when needed, asking a question that helps to take the reasoning further, or suggesting that an error be considered from a different perspective".

In her view – developed together with colleagues Elvis Mazzoni and Mariagrazia Benassi – an artificial intelligence system designed for an educational context should become a “functional component” closely integrated with our way of thinking and our cognitive patterns.

“Just as using a pen to write helps develop motor-cognitive coordination and the construction of linguistic structures, a well-designed AI system can become a fundamental part of how a person thinks, plans, reflects and creates”, says Benvenuti.

A tailored dialogue

In this context, therefore, generative AI systems are taking on a new role: no longer mere providers of easy, generic answers, but collaborative tools for planning, exploring and learning. Also by making mistakes. In an educational context, for example, they can establish a tailored dialogue with pupils, asking questions and striking the right balance between challenges to be overcome and support for learning.

"We should use these tools as virtual assistants that stimulate creativity and critical thinking", Benvenuti confirms. "The wealth of information and ideas they provide can be a great help both at school and at work, but we must make active use of them by rereading, reflecting and looking for different perspectives”.

This is also because, whilst on the one hand the introduction of digital technologies is affecting our attention spans and memory, on the other it is also leading to the development of new cognitive abilities. Knowing where to retrieve information, organising different types of material, comparing sources and assessing the reliability of an answer are skills that are becoming increasingly widespread and important. Similarly, those who regularly use AI systems are learning in increasingly sophisticated ways when these tools are more or less useful, and consequently how to use them to achieve the desired result.

"Understanding when and what to delegate to artificial intelligence is an increasingly important skill to develop", explains Chiara Lucifora. "There is no need to become IT specialists, but it is important to understand how these systems work: knowing them better also allows us to distinguish between the tasks where our intervention is needed and those that can be entrusted to a machine".

Clear benefits

When faced with a request from a student who needs to prepare a research project, for example, an AI system might ask what problem they wish to tackle, what they already know, and which sources they have consulted. It could then suggest approaches, compare different interpretations, and ask them to justify the choices to be made. The final text would thus be the result of a collaborative effort, encouraging the student to reflect and learn new information.

"Before explaining something, a teacher tries to understand what the person in front of them already knows, and only then suggests they take the next step", says Aldo Gangemi. "We would like artificial intelligence to be capable of developing a process of this kind too: not by providing immediate answers, but by helping the user to plan, assess and better understand the problem at hand".

This configuration could be a specific mode of use to apply to existing models. But for such an approach to find a place in an educational context, it must offer clear benefits. If the aim is to use AI to look up practical information, the way forward is a quick, simple, and accurate answer. When studying, in educational activities or when faced with a complex problem, however, it becomes important to highlight the benefits of a system that asks us to stop and think.

"The model we have in mind is based on building knowledge with the support of technology, not on replacing learning", says Martina Benvenuti. "When students work in pairs or small groups, they need to discuss and negotiate a solution: artificial intelligence can also form part of this process, as a tool to use together and to discuss".

Ethical issues

However, introducing these tools into a school setting also raises crucial ethical issues. To tailor support to a student, a system must collect information on how they work and on their difficulties, and this makes the issue of confidentiality and data protection a key consideration. What's more, teachers must be able to intervene, correct or disregard the system's recommendations, whilst retaining responsibility for educational choices.

"Ethical issues should come before the development of AI systems, with a focus above all on transparency", Benvenuti confirms. "This means following a design and development process that begins with ethical considerations, and it means sharing and discussing these issues with users, so as to make the entire process transparent".

In short, for those who develop these tools, the way forward is to make them understandable, adaptable, and open to input from teachers. Teachers, meanwhile, should learn to use them with a specific teaching objective, giving students time to formulate ideas, make mistakes and discuss them. And even in workplace contexts, AI can help carry out complex activities, provided people retain the abilities to assess the results and decide how to use them.

“The choice we face today is not whether to use or not use artificial intelligence systems: the choice is between systems capable of strengthening our cognitive abilities and systems that risk replacing them”, Gangemi adds. “The tools, evidence and theoretical foundations needed to make the best choice are already at our disposal: all we have to do is decide how to apply them”.