According to a recent study from the University of Pennsylvania, using theGenerative AI – i.e. artificial intelligence tools capable of generating text, images or videos, such as ChatGPT – without a critical spirit it improves our results immediately, but in the long term it damages us because it limits our ability to learn. The best way to use AI for learning in school or for study goals is by considering it as a digital tutorwhich does not directly give us the answer to the problem but helps us ask ourselves the most appropriate questions to arrive at the correct solution. In this way, not only can we exploit the capabilities of AI to optimize our study, but we also acquire the skills and abilities necessary to evaluate any errors in the answers.
Generative AI impacts student learning: The study
The study involved 1000 students of a Turkish high school to understand how the use of generative artificial intelligence affects learning. The students were divided into three groups:
- Control group: They didn’t have access to AI.
- GPT Basic: They could use a tool with a standard version of GPT-4 (one of the models behind ChatGPT), similar to ChatGPT.
- GPT Tutor: they could exploit a version of GPT-4 designed specifically to stimulate learning. This version was designed for give feedback and suggestions to students without giving them the answer directlybut guiding them step-by-step in solving the problem.
During this experiment, students first took a math lesson and then underwent two tests: a first test where the “GPT basic” and “GPT Tutor” groups could use AI and a second test without any kind of support, to evaluate how much they had actually learned.
The results of the study: if we use AI we have better results but we risk not learning
As you might imagine, groups with access to AI got it results significantly better in the first test compared to the group that could not use it (+48% with GPT Base and +127% with GPT Tutor). In the second testHowever, the situation has changed. When no group could use the AI anymore, the control group achieved the same results as the one using “GPT Tutor” and much better results compared to the GPT Base group (+17% for the control group).
These results show us three important things. The first is that if we use generative AI to help solve complex problems, our results improve. The second is that, if we limit ourselves to asking ChatGPT for solutions and copying the answers, we do not learn and develop the skills we need to evaluate whether the answers given by the AI are actually correct. The third is that the best way to use generative AI to learn is to think of it as a digital “tutor”. that doesn’t give us the answers, but there ask the right questionsto stimulate us to think. In fact, students who used “GPT Tutor” not only achieved significantly better results when they could use it, but also demonstrated that they had actually learned how to solve problems by obtaining the same results as the control group in the test without the support of AI tools.
How to use AI to learn and study
To learn with AI, therefore, it is better for us to use tools that adopt the Socratic methodthe same one used by the studio’s “GPT Tutor”. This approach does not give direct answers, but stimulates critical thinking through questions that help achieve independent understanding.
The model that uses this method more than any other and was designed specifically for learning is “EULER”. This new tool, created by EURECOM, never directly provides answers to users, but it entertains one conversation with them placing ever deeper and more complex questions. The template isn’t perfect yet and sometimes goes off topic with questions, but it’s a great start.
Another interesting example is NotebookLMby Google. We can upload documents and external sources and, through the “Study Guide”, NotebookLM generates 10 questions to guide us in our understanding of the text. We can also generate gods fake podcasts (for now only in English) in which two interlocutors discuss the ideas present in the documents we have uploaded, so that we can learn even if we have difficulty reading. It then allows us to summarize the text and highlight key facts, all automatically.
An additional way to use generative AI in a constructive way is to request feedback personalized on the work done. We can, for example, ask AI to evaluate an essay, a report, a project and provide comments on logic, structure and vocabulary. This can be useful for students to receive personalized and frequent feedback, but also for workers to evaluate if and how to improve their projects.
