Secondment: Davide Cirillo
|
Institution of origin |
Machine Learning for Biomedical Research, Life Sciences Department, Barcelona Supercomputing Center (BSC) |
|---|---|
|
Host institution |
Psychology Lab, Institute of Applied Biosciences (INAB), Centre for Research & Technology, Hellas (CERTH) |
|
Initial objective |
Define evaluation metrics for Explainability, Fairness & Equity, and Intersectionality within the AHEAD AI assessment methodology. |
Results achieved
This secondment was carried out in May-June 2025. At the beginning, Dr. Cirillo participated in introductory exchanges with CERTH staff and became familiar with their approaches to AI evaluation in health and psychology. These initial interactions were particularly valuable in highlighting how a psychology-oriented research environment frames issues such as model assessment, interpretability, and human factors, adding an important complementary perspective to more technically driven approaches.
The main body of the work focused on the joint development of operational definitions and evaluation metrics for three key indicators: (1) Explainability, (2) Fairness & Equity, (3) Intersectionality. Explainability aims to make AI model decisions and recommendations transparent and understandable to foster trust and interpretability, Fairness & Equity focuses on identifying and mitigating biases in data and model development to reduce discrimination across attributes such as sex, gender, race, age, and socioeconomic status. Intersectionality addresses how overlapping and interacting forms of bias can compound and produce unequal outcomes.
A key strength of the secondment was the continuous interdisciplinary exchange between BSC’s expertise in machine learning and AI ethics and CERTH’s expertise in psychology and human factors. This collaboration helped ensure that metrics were consistently grounded in real-world healthcare needs, strengthening both their relevance and applicability. Alongside the technical work, the team also contributed to refining organisational aspects of the AHEAD secondments, which have benefited future activities across the project. A productive working relationship between BSC and CERTH was also established, laying the groundwork for continued collaboration in WP2.
The outputs of the secondment were presented during a consortium training session organised at BSC in June-July 2025. A further portion of the secondment will be completed during a future visit to CERTH, extending the work to additional indicators and contributing to the consolidation of the WP2 methodology.
