Conference Program

The paper presents early findings of an empirical research, within an exploratory framework, involving assessment processes carried out by generative artificial intelligence (AI) and human evaluators (HI). Research questions wonder whether I) between results of semi-structured test assessments carried out by HI and results of assessments carried out by AI, lies a discrepancy that could be considered critical – greater than the one found between human evaluators in the previous research phase – and II) if the upload of teaching materials, integrated with AI-HI interaction, have a significant relation with discrepancy reduction. Objectives of research are then to control the existence of such critical discrepancy and the relation between teaching materials upload and discrepancy reduction. The research hypothesis states that I) a critical discrepancy exists and that II) it can be reduced with teaching materials uploading, commented by human evaluators. Population consists of Masters’s Degree students, the sample being composed by those who have carried out in groups, as in itinere or summative assessment, a semi-structured project work during academic years 2024/2025 and 2025/2026.

Awaiting further investigation, initial results describe the discrepancy as critical and seem to indicate that AI may be subject to docimological bias already attested for human evaluators as adoption of implicit criteria, failure to identify intermediate performance levels, lack of consistency over time, in addition to specific AI-related biases. This line of research is established in defence of a responsible and inter-subjective assessment practice, promoting quality and transparency in democratic pedagogy.