Ethics by design: principles and questions during development

A practical compass is the four principles of medical ethics: autonomy, justice, beneficence, and non-maleficence. The "ethics by design" approach, as understood by the European Commission, means preventing and addressing ethical issues already in the early phases of development, not retrospectively. There is no single recipe; requirements change with the project phase and the context of use. Therefore, at each step let us ask whom the system serves, what assumptions we build into it, how we measure benefit, and who bears responsibility.

When data mislead: examples of bias and the AI life cycle

A cautionary example is a 2019 study that revealed a widely used risk-prediction algorithm in the USA disadvantaged patients of color because it was trained on costs rather than health status. Patients with the same "risk" according to the model were actually doing differently—the people of color were sicker because they had historically drawn on fewer resources. No one noticed that the data themselves bore traces of systemic inequality, thereby distorting the tool’s performance. Similar questions are raised by end-of-life prediction models: if an algorithm forecasts one year until death, what is the accuracy, how should it be communicated, and what if the physician disagrees with the output.

This is where the difference between equality and equity comes into play: equality gives everyone the same, equity evens out unequal starting conditions. For example, wearables with optical sensors that use red or yellow light may record less accurate data from darker skin, which calls for technical adjustments and broader testing. The aim is for solutions to work fairly for entire target populations, not just a subset. An ethical approach therefore follows the entire AI life cycle—from design through development and deployment to ongoing monitoring of impacts on users and society.