AI-Adaptive Programming Education for Socially Disadvantaged Students — A 13-Week Study Across Four Hungarian Universities

A research team from the University of Dunaújváros, Óbuda University, John von Neumann University, and Budapest University of Economics and Business ran a quasi-experimental study to test whether an AI adaptive learning system could narrow programming-skill gaps for socially disadvantaged students. 122 students aged 18-21 across four Hungarian universities took part, split into a 61-student experimental group and a 61-student control group. The study ran over 13 weeks during 2024-2025 and was published in TechTrends (Springer) in 2025.

The experimental group used a ChatGPT-based adaptive tutor employing Bayesian Knowledge Tracing to model each learner's knowledge state and adjust task difficulty in real time. The control group received the same 13-week curriculum, covering control structures, functions, loops and algorithms, via traditional instruction. Both groups were taught by the same instructor with identical labs, hardware and internet access, isolating the AI tutor as the variable of interest.

Test scores rose from a mean of 5.1 to 7.1 in the experimental group (d=1.40) versus 5.0 to 6.0 in the control group (d=0.74), both p<0.001. Engagement also increased across behavioural (d=1.10), emotional (d=1.00) and cognitive (d=1.40) subscales (Cronbach's α=0.89), all p<0.001. The experimental group's gains were substantially larger than the control group's across every measured outcome.

The authors note that the 13-week duration 'limits any conclusion about long-term retention', and that participants were recruited via survey response rather than full random sampling, raising a risk of selection bias. The study covers a single subject, introductory programming, and has not yet been replicated or scaled beyond these four universities.