NFDI4DS Lecture Series #8 / NFDI WG DSAI

by Mariana Vitti Rodrigues (Universidade Estadual Paulista "Júlio de Mesquita Filho")

Monday 23 Jun 2025, 09:00 → 10:00 Europe/Berlin

From Epistemic Opacity to Trustworthy Medical AI: Is transparency the pathway?

This presentation investigates the epistemological grounds for the ethical challenges brought about the growing use of automated decision support systems in diagnostic reasoning. Diagnostic reasoning, often described as a kind of abductive inference, can be understood as the process of generation and selection of plausible hypotheses to explain a patient’s set of signs and symptoms. The advances of machine and deep learning models for data analysis have promised an improvement in accuracy, performance, and efficiency of automated decision support systems to enhance the quality and speed of diagnostic reasoning. While the beneficial prospects of the use of Artificial Intelligence in healthcare are undeniable, their growing use in clinical settings challenge the type of trust attributed to results obtained by black box models, i.e., opaque computational models whose inner workings cannot be accessible to anyone due to their inherent complexity. Whereas regulations and ethical guidelines are pressing for the development of more transparent and interpretable algorithms to justify the rationale underlying automated forms of decision-making processes, the lack of conceptual clarity around the notion of opacity - and its correlated terms such as explainability, interpretability, and transparency - challenges the development of strategies to promote trustworthy AI. In this context, we will investigate, analyze, and compare forms of epistemic opacity present in machine learning models and human abductive inference in order to challenge the plausibility of requiring full-fledged transparent automated systems for building trustworthy Medical AI.

 

Speaker Bio:

Mariana Vitti Rodrigues is a researcher in Philosophy of Information, Science and Technology. She investigates epistemological and ethical consequences of the increasing automation of scientific practice enabled by the development of machine learning algorithms, emphasizing the analysis of strategies to overcome algorithmic opacity in bioinformatics and biomedical informatics.

 

The talk will be available via this link: https://tu-berlin.zoom.us/j/62739607385?pwd=YnVlMWpmVlF6UGQxMGhBVE1IME5ZUT09