It’s a challenge as old as education: adapting learning to each student rather than imposing the same lesson on both the best and the worst. To meet this challenge, education professionals began to dream, as early as the 1960s, of artificial intelligence (AI) capable of doing so at low cost and on a large scale. In recent years, the arrival of ChatGPT, available since 2022 and already used by teachers to personalize their exercises, has revived this hope.
However, it is not the conversational robot developed by the American company OpenAI that national education is currently deploying but an older technology, based on the precepts of “adaptive learning” – adaptive learningin English. After experiments carried out from 2020 with six companies, Gabriel Attal announced in 2023 the deployment of a digital tutor, Mia Seconde, with 800,000 high school students in second grade – an almost entire age group – in support of mathematics and French lessons from the start of the 2024 school year. Her successors at the Ministry of National Education have reduced this ambition: “Mia Seconde is only offered to a few tens of thousands of students at the moment,” observes Catherine de Vulpillières, the co-founder of EvidenceB, the company which won the call for tenders and is now developing the software.
To watch stripped down menus of the latter, we cannot guess how complex its machinery is. From the thousands of exercises with minute level variations that have been memorized, he chooses, for each student, only a few dozen exercises, in order to build a learning path that is neither too complex nor too simple. The course is personalized thanks to an initial level test, then thanks to an AI which observes the student’s response times, their errors and their rate of progress. “The hope is to promote a feeling of competence and to prevent the student from becoming demobilized,” analyzes Christophe Jeunesse, director of the education sciences department at Paris-Nanterre University.
“Adaptive learning”, an outdated technology?
Choosing software based on this technology may come as a surprise as the popularity of adaptive learning has been waning among digital learning experts since 2017 – according to a survey conducted by the research firm Opensesame. Media startups from the 2000s and 2010s, like Knewton and Amplify, were sold for very modest sums. “Too many deployed systems are too basic or poorly founded theoretically”points out Fien Depaepe, professor at the Center for Educational Psychology and Technology at the Catholic University of Louvain, in Belgium.
