<![CDATA[Misconceptions about AI ethics and algorithmic bias remain prevalent among vocational high school students, particularly the belief that AI systems are inherently objective and neutral. This study investigates whether the Think–Pair–Share (TPS) cooperative learning model is more effective than conventional lecture-based instruction in reducing misconceptions and improving conceptual understanding of AI ethics and bias. The study addresses the need for empirically validated pedagogical strategies in secondary-level AI ethics education. A quasi-experimental non-equivalent control group pretest–posttest design was employed involving 68 eleventh-grade vocational students (34 experimental; 34 control). The experimental group received four weeks of TPS-based instruction, while the control group received lecture-based instruction covering identical content. A validated two-tier diagnostic test (20 items; α = 0.87) measured misconceptions across five domains: algorithmic bias, data representativeness, fairness and discrimination, transparency and accountability, and human oversight. Data were analyzed using paired and independent samples t-tests, normalized gain scores, and Cohen’s d effect size. The TPS group demonstrated significantly higher normalized gain (M = 0.63, SD = 0.12) compared to the lecture group (M = 0.32, SD = 0.15), t(66) = 9.14, p < .001, with a large effect size (d = 1.45). The greatest misconception reduction occurred in the fairness and discrimination domain (70%). Both hypotheses were supported. The study was limited to one school and short-term intervention duration, restricting generalizability and long-term retention analysis. This study provides empirical evidence supporting TPS as an effective instructional strategy for AI ethics education in vocational contexts and contributes a validated diagnostic instrument for measuring AI bias misconceptions. ]]>
