In France, mathematics become a field of inequality from the start of schooling, with differences that are widening over the years and eventually weigh heavily in international assessments. While teachers face increasingly heterogeneous classes, a French start-up is based on artificial intelligence at school to offer an individualized approach to learning, focused on the real needs of students. At the intersection of neuroscience and educational technologies, this initiative sketches a new way of thinking about the pedagogy of fundamentals.
The fractions, in particular, represent a frequent break in the understanding of the students. Difficulty understanding the notion of part of a whole, fractional logic not very intuitive, specific vocabulary … This mathematical field requires careful support, all the more difficult to ensure in heterogeneous classes or in priority education. For some students, this is where confidence in their abilities is starting to crumble.
An AI at the service of teachers, not in their place
Faced with this observation, the French start-up Evidenceb offers an original technological solution. Called Adaptiv'Fraction, their module fits into a wider program called Adaptiv'Math, designed for primary school. This module is based on the joint contributions of artificial intelligence and cognitive sciences, in order to automatically adapt the exercises at the real level of each student. The idea is simple. Instead of offering a unique route, all exercises are adjusted in real time depending on the student's responses.
Unlike certain received ideas, this tool is not designed to replace the teacher. Catherine de Vulpillières, co -founder of Evidenceb and former teacher in preparatory class, insists on this point. The module does not change the course, it only replaces the training phase, where the need for differentiation is the most glaring. The teacher retains all of his role, by adding personalized digital support.
The development of the interface relied on work carried out with specialized laboratories, particularly in cognitive sciences and computer science. The Flowers Laboratory of Inria Bordeaux and the Lapsydé of Paris Cité University thus contributed to designing a tool that is both rigorous scientifically and motivating for children. The attention paid to user experience, visual clarity and the fluidity of the routes is designed to avoid cognitive overload, while promoting commitment.
Can artificial intelligence at school really fill the gaps?
To assess the effectiveness of the module, researchers have conducted two studies with 555 students in 31 CM1 classes, including nine in priority education. As the research article explains, the teams have rigorously framed these experiments. They divided each class into two groups: an experimental group using adaptiv'fraction one hour per week for five weeks, and a control group according to the usual courses without the module. The researchers then brought tests to the students before and after this period, on two levels of difficulty.
The preliminary results are striking. In a test of five questions on the fractions, students who used the module increased by 36.6%, against 17.5% for the group without AI. On a more complete test of 24 questions, only half equipped with adaptiv'fraction shows a notable progression, reaching 8% improvement. This doubling of performance seems to indicate that the personalization provided by AI plays a key role in assimilating concepts.
This progress remains unequal. Students in difficulty advance slower, but they do not drop out. They manage to reduce the gap with their comrades, even if the pace is slower. The complexity of the exercises does not brake them, because it adapts to their level. Notable, girls have progressed four times faster than boys with this module. Before its use, it was however the opposite. For researchers, this difference would be explained by a pace of consolidation better suited to certain female cognitive profiles.
Technology will not solve all the problems of the school, but it can correct certain inequalities. Here it offers concrete help. We must now analyze the data with rigor and validate the long -term effects. A wider reflection must follow, to consider a large -scale implementation. If the results are confirmed, artificial intelligence at school could become discreet, but very effective support.
