This Student Does Not Exist: A Scoping Review of Synthetic Data for AI in Education

• In Event: AERA Poster Session Friday 7:45 am
  In Poster Session: ATL Poster Session 2: AI Integration and Outcomes in Postsecondary Settings

Abstract

Using synthetic data has become an appealing method for training AI models in various industries. In fields where data is expensive to acquire and subject to privacy concerns--like education--there are compelling reasons to use the technique. However, it also comes with risks and may not be suitable for certain contexts. These risks, and the underlying learning and learner theories that undergird them, remain underexplored and largely absent from less technically-oriented education discourses and literature. To address this, we performed a scoping literature review investigating the use of synthetic data in AI for education. Based on our analysis of these results, we then offer a preliminary framework for guiding risk-benefit analyses of using this technique in educational models and tools.

Authors

• Sarah Burriss, Vanderbilt University

• Golnaz Arastoopour Irgens, Vanderbilt University

• Dan Carpenter, North Carolina State University

• Ole Molvig