evidoria

← Back to browse

Good practice Imported

AI Adaptive Programming Tutor Nearly Doubles Learning Gains for Disadvantaged Students — University of Dunaújváros

Hungary · Dunaújváros · See the Hungary profile

Evidence: Randomised controlled trial Top 17% 67/100 · Ask Evidence Copilot about this practice

A 13-week controlled study at four Hungarian universities found a ChatGPT-based adaptive tutor using Bayesian Knowledge Tracing nearly doubled programming gains for 122 disadvantaged students (Cohen's d=1.40, p<0.001) versus standard instruction (d=0.74).

5.1 → 7.1 (Cohen's d=1.40, p<0.001)
Experimental group programming test score (pre → post) (2024–2025)
5.0 → 6.0 (Cohen's d=0.74, p<0.001)
Control group programming test score (pre → post) (2024–2025)
d≥1.00, p<0.001
Engagement effect size (behavioural, emotional, cognitive) (2024–2025)
n=122 (61 experimental, 61 control)
Study sample size
AI Adaptive Programming Tutor Nearly Doubles Learning Gains for Disadvantaged Students — University of Dunaújváros

Details

Maturity
Pilot
Promoter
University of Dunaújváros
Period
2024–2025
Region (NUTS)
HU211
Keywords
higher education, computer science, AI tutoring, digital divide, disadvantaged students

Context

Researchers at the University of Dunaújváros studied whether an AI adaptive-learning system could help close the digital divide for socially disadvantaged students learning computer programming, screening 122 students aged 18-21 across four Hungarian universities using a validated 15-item Social Condition Index.

Objectives

The study set out to test whether an AI system built on ChatGPT with Bayesian Knowledge Tracing, adapting task difficulty to each learner's knowledge, could raise programming learning gains for disadvantaged students compared with standard instruction.

Activities

Students were randomly assigned to an experimental group (n=61) using the AI adaptive tutor or a control group (n=61) receiving standard instruction over a 13-week period, with pre- and post-tests plus engagement measures.

Results

The experimental group's test scores rose from 5.1 to 7.1 (Cohen's d=1.40, p<0.001), nearly double the control group's rise from 5.0 to 6.0 (d=0.74, p<0.001); engagement was also significantly higher across behavioural, emotional and cognitive measures (all d≥1.00, p<0.001), and gains held regardless of students' socio-economic background.

Conclusions

The authors caution that the 13-week duration limits conclusions about long-term retention and that survey-based recruitment rather than full random population sampling may introduce selection bias, so replication over longer periods is warranted.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Short (< 1 year)
Staffing & skills
Education researchers to design and administer the ChatGPT-based adaptive tutor, Instructors delivering standard instruction to the control group in parallel

Conditions for success

  • A validated screening instrument (e.g. the 15-item Social Condition Index) to correctly target disadvantaged students
  • Sustained access to the AI adaptive tutor across the full 13-week study period
  • Random assignment maintained across cohorts to preserve comparability between groups

Common failure modes

  • The short, 13-week intervention window limits evidence on long-term retention
  • Survey-based recruitment rather than full random population sampling may introduce selection bias

Commonly funded by

National / regional programmes Erasmus+ Horizon Europe

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful