Skip to Content
Academic Mode
Research Brief

Does GenAI Help Those Who Need It Most?

Estimating Heterogeneous Treatment Effects of Generative AI on Student Performance in Danish High Schools.

Collaborators

Status

Proposal

Grade

10

Submitted

Jan 2026

Summary

As Generative AI becomes ubiquitous in education, a critical question emerges: Does it level the playing field or exacerbate existing inequalities? This project, submitted as a research proposal in January 2026, proposes an investigation into the heterogeneous effects of GenAI usage across socioeconomic backgrounds. By linking detailed survey data with administrative grade records from Danish upper-secondary schools, we aim to uncover whether AI acts as an 'equalizer' or if it deepens existing divides.

Research Questions

01

What are the heterogeneous effects of GenAI usage on the academic outcomes of students?

02

Does the introduction of GenAI in education reduce or deepen existing socioeconomic inequalities?

03

Do students from different socioeconomic backgrounds use GenAI at different rates, for different subjects, and for different purposes?

Methodology

Model Framework

Causal Forest (Machine Learning Inference)

Identification Strategy

Estimating Conditional Average Treatment Effects (CATE) to capture non-linear heterogeneity.

Data & Frequency

Primary Survey Data + Administrative Grade Records (Pre/Post GenAI) for ~800 students.

Estimation Approach

Difference-in-Differences setup within a Causal Forest framework to control for high-dimensional covariates.

[ Transmission Pathway Diagram: Model Architecture ]

Hypotheses & Expectations

  • Research Proposal: This project is a research design without empirical results yet.

  • Hypothesis: We expect the impact of AI to vary significantly based on student background and usage patterns.

  • Focus: Differentiating between 'high-quality' (tutor-like) and 'low-quality' (shortcut) usage is key.

  • Ethics: The project addresses fairness and the long-term consequences of AI in the educational system.

Bibliography

  • Thomas, A. (2024). AI-assisted tutoring and student performance. Journal of Educational Psychology (Forthcoming).
  • Wager, S., & Athey, S. (2018). Estimation and Inference of Heterogeneous Treatment Effects using Random Forests. Journal of the American Statistical Association.
  • Knittel, M., & Stolper, S. (2024). Functional Heterogeneity and Causal Inference.
  • van de Mortel, T. F. (2008). Faking it: social desirability response bias in self-report research. Australian Journal of Advanced Nursing.

Technology

LaTeX

Data & Tools

Sources

Primary Data Collection (Survey), Administrative Systems (Lectio/Ludus).

Software

LaTeX (Writing).

Documents

Download Thesis (PDF) Full Text • 2.4 MB
Course Page
Reference
Jørgensen, Anton M. E., Chen, Sheng Ye Michael & Nielsen, Jonas Amasa Skov (2026). Does GenAI Help Those Who Need It Most? Research Proposal, University of Copenhagen.

Relevance

Evidence-based regulation: Should schools ban, restrict, or embrace GenAI?

Equity in Education: Identifying groups that need targeted support for 'AI Literacy'.

Future Skills: Understanding how early adoption affects human capital accumulation.

Cite this work

@misc{does_2026,
  author = {Jørgensen, Anton M. E., Chen, Sheng Ye Michael & Nielsen, Jonas Amasa Skov},
  title = {Does GenAI Help Those Who Need It Most?},
  year = {2026},
  url = {https://antonebsen.dk/projects/does-genai-help-those-who-need-it-most?}
}
Jørgensen, Anton M. E., Chen, Sheng Ye Michael & Nielsen, Jonas Amasa Skov (2026). Does GenAI Help Those Who Need It Most? Research Proposal, University of Copenhagen.