AI in Learning Assessment: Reducing Bias or Reproducing Inequalities? A Comparative Study of Human and AI Scoring In French as a Foreign Language
TL;DR
This article analyzes whether automated assessment-and more specifically, assessment using AI-can correct biases or whether, on the contrary, it risks introducing new ones and recommends a hybrid approach combining AI and human teachers as a balanced solution.
Abstract
Abstract: Education has not been left behind in the face of change; as new technologies have significantly transformed the educational landscape. The integration of digital technology into teaching practices has opened the door to more diverse learning methods, greater access to resources, and more flexible and personalized instruction, creating a forum for discussing how to leverage artificial intelligence as a catalyst to enrich the educational experience. Assessment, as an essential component of the teaching-learning process, often remains on the sidelines of these developments. It is our responsibility to carefully plan its design, administration, grading, and analysis in order to fully leverage its central role in language learning and minimize its well-documented biases. This article analyzes whether automated assessment-and more specifically, assessment using AI-can correct these biases or whether, on the contrary, it risks introducing new ones. The study is based on an experiment conducted with 90 Higher Education Cycle (Semester 1), whose exams were graded by two teachers and an AI grader. Significant discrepancies emerged. While the AI reduces some of these biases and brings standardization and consistency to grading, algorithmic biases and a lack of consideration for qualitative dimensions also emerged. Consequently, a hybrid approach combining AI and human teachers is recommended as a balanced solution. Keywords : Educational assessment; Artificial intelligence ; Assessment bias ; ChatGPT5