From trustworthiness to risk, from risk to responsibility. A responsibility-centered proposal for AI-based applications in healthcare

The AI Act approved by the European Parliament in May 2024 stands as the most important and complex regulatory instrument that has appeared so far, at least in the EU area. However, when compared to the directions that can be found in other drafts such as the Ethics Guidelines for Trustworthy AI of 2019, it marks a significant departure. Whereas before the focus was on developing principles that would allow for ethical-deontological lines of conduct for professions that have or will have to deal with AI, it is evident how the AI Act of 2024 is modeled on a risk-based approach. This paper critically examines the implications of this paradigm shift, particularly in terms of accountability and the neglect of interpersonal dynamics inherent in real-world work settings. Instead, in the risk-centric model employed by the AI Act, the regulatory emphasis primarily revolves around the continual reassessment and classification of AI applications deemed at-risk or potentially prohibited. Taking as a prime example for study the field of medicine, where the use of AI is growing more and more dramatically especially in diagnostic fields, the paper suggests an integrative model that could be useful in making up for those shortcomings. The proposed model is grounded in practice and underpinned by the incorporation of three guiding tenets: trust, transparency, and traceability, collectively referred to as the “3 T’s”.