Embedding Responsible AI in a Technical Machine Learning Course

Abstract

Machine learning (ML) courses in higher education usually emphasise mathematical foundations and algorithmic methods, while ethical and societal questions are often treated as separate or optional topics. This paper reports a small-scale practice-based evaluation of a blended design that embedded responsible AI within a Master’s-level ML and Data Mining course at a German university in winter semester 2025/26. The design combined fifteen on-campus lectures on core ML topics, including one lecture on fairness as a technical topic, fifteen optional weekly assignments discussed in tutorials, and an eight-module microlearning course on AI literacy and responsible AI. Completion of at least half of the microlearning and a mandatory practical assignment were required for exam eligibility. The study analyses an anonymous end-of-course survey, using only responses from students who explicitly consented to research use (n=35). Student perceptions were generally positive: 78% reported a good balance between theory and practice, 75% found the weekly assignments important for understanding, and 59% judged the microlearning to fit well with the on-campus lectures. For responsible AI, around two thirds reported greater confidence in identifying fairness issues, feeling better prepared to discuss ethics professionally, or perceiving ethics as integrated rather than added on. 75% intended to consider fairness and ethical implications in future ML projects, with no respondent disagreeing. Open responses suggested perceived transfer, with students naming hiring systems, facial recognition, student-risk prediction and credit or loan decision-making, and often using technical vocabulary such as training data, optimisation objectives and evaluation metrics. The paper does not claim causal impact, and findings should be interpreted in light of the small voluntary sample, the single-cohort single-institution setting, and the possibility of social-desirability bias on an end-of-course survey on ethics. Instead, it offers descriptive evidence that a lightweight microlearning layer, combined with a technical fairness lecture, can be a feasible way to connect technical ML learning with ethical reflection in practice. It also identifies revisions for the next iteration: reduce workload tension, add explicit ethical-reflection prompts to assignments, and include more practical coding work.