Topic-relevant experience divides opinion before discussion: a case-method pilot in AI-ethics education

Abstract As AI systems are delegated consequential decisions such as hiring, AI-ethics education is widely regarded as urgent yet under-developed. We propose the business-school case method as a pedagogy for AI ethics and report a single-site pilot ( $$N=20$$ students) built around an original case modeled on real recruiting-AI incidents (the Rikunabi affair and Amazon’s abandoned hiring tool). We ask how opinions change across preparation and discussion, and which attributes are associated with divergence that could inform grouping. Opinions measured after pre-reading (T1), individual preparation (T2), and group discussion (T3) were analyzed with a fully specified mixed ANOVA, simple-effects decomposition, and permutation tests. Prior topic-relevant experience (having job-hunted) was associated with opinion: experienced students were markedly more in favor of AI-based applicant evaluation ( $$F(1,16) = 5.03$$ , $$p =0.039$$ ). For major field of study we detected no association, but that comparison was severely under-powered and is reported as inconclusive rather than null. The experience difference was a pre-discussion phenomenon: large and significant at T1 and T2, but no longer detectable at T3, where the estimate is near zero yet imprecise (95% CI [ $$-1.39$$ , 1.31]). The narrowing was concentrated in the interval containing the discussion and was corroborated on a four-item composite (difference-in-differences $$p =.023$$ ), though this evidence remains exploratory. Cohort-level dispersion barely changed: discussion coincided with a loss of the experience-based patterning of opinion without erasing disagreement. We interpret this as a deliberative dynamic, frame the grouping implication as a hypothesis for future testing, and report the study transparently as an exploratory pilot with no control condition.