Scientists find that neural networks boost student performance through engagement

AI works through interest, not through ready-made answers

Educational neural networks — chatbots, intelligent tutors and writing assistants — improve students' academic performance, but indirectly: first they draw learners into the learning process, and it is engagement that then brings good grades. Researchers from Tongliao and Taiyuan universities came to this conclusion after surveying 438 university students in China's Shanxi province; the work was published in the journal Frontiers in Psychology. The study is reported by Gazeta.Ru.

What the numbers showed

In an online survey, students indicated how often they use neural network tools, how much effort they put into their studies, and how they assess their own progress. By engagement, the authors meant the willingness to devote energy, interest and attention to classes. The responses were processed using structural equation modeling, which builds chains of factors influencing one another. The link between neural networks and engagement proved notable (correlation coefficient 0.52), with grades it was more modest (0.36), while engagement, in turn, is closely tied to academic performance (0.47). The direct effect of AI on grades is weak (0.18), whereas the indirect one, through engagement, amounts to 0.21 out of a total effect of 0.39 — that is, more than half. The benefit of neural networks for learning, according to the study, is realized primarily through interest and immersion, not through quick answers.

The mechanism and limitations of the conclusions

The authors explain the mechanism as follows: instant feedback and step-by-step explanations reduce unnecessary mental load, help students not to give up on a difficult task and to work through it on their own. 'The value of AI for education cannot be fully understood from a purely technological standpoint,' the researchers write, noting that the effect depends on whether the student's thinking remains engaged in working with AI or is replaced by ready-made answers. The work adds an argument to the debate over whether chatbots should be seen as a cheat sheet or a tool: critics believe that students delegate homework to neural networks and unlearn how to think. At the same time, the conclusions are based on a single survey, and the authors, as Gazeta.Ru notes, point to the need for verification through long-term observations; until then, neural networks should be regarded as an amplifier of interest in learning.