From algorithmic bias to organizational accountability: what a course on trustworthy AI changed in a study of ethical AI adoption in Colombian hospitals

completed Trustworthy AI: Managing Bias, Ethics, and Accountability (Johns Hopkins University, Coursera, 16hours) because my dissertation asks how organizational culture shapes the ethical adoption of AI in ColombianLevel IV hospitals, and the course treats bias and accountability as problems that outlive the model. I adoptthree of its ideas: measurement bias and algorithmic drift, which give algorithmic bias a more central placein my theoretical framework; the human-baseline approach, which adds questions to my interview guide;and human control over the tool, the “human control button”, which I propose to examine as a characteristicof decision-makers. I reject the course’s U.S.-centred regulatory frame, because Colombia has no AI statutethat could carry a compliance-driven approach, and model-level fairness metrics, because my unit of analysisis the decision-maker, not the algorithm. I applied the course’s representativeness audit to open data fromReTHUS and DANE: three departments with 37.3 % of the population hold 60.1 % of specialist registrations,and specialist density falls as multidimensional poverty rises (Spearman ρ = −0.71). AI adopted by Level IVhospitals is likely to be validated where specialists are, not where they are missing. From an 80-study scopingreview I chose the gap on organizational culture as an antecedent of ethical AI adoption in middle-incomehospitals. No decision has been formally changed yet: the three reinforcements are under discussion withmy supervisor, and the main one concerns the framework, where algorithmic bias stops being one of threeethical problems and becomes a central axis.