Trustworthy AI
Artificial Intelligence has significant potential to accelerate biomedical research, but its use in health-related settings requires particular care. Because biomedical AI outputs can shape scientific hypotheses, disease models and future approaches to diagnosis and treatment, researchers need to understand their reliability, uncertainty, limitations and potential biases.
These challenges are particularly relevant for Generative AI, which can produce plausible sounding but incorrect or biased outputs and may rely on correlations that do not reflect underlying biological mechanisms. Trustworthy AI must therefore combine scientific reliability with transparency, fairness, privacy, explainability and meaningful human oversight.
In Europe, these principles are reinforced by an evolving ethical and regulatory framework, including the EU AI Act, the General Data Protection Regulation (GDPR) and the EU Ethics Guidelines for Trustworthy AI. Building on these European principles, AIRIS integrates responsible AI throughout its research and development.
Trustworthy AI in AIRIS
This commitment to trustworthy and ethical AI is reflected across the development and use of AIRIS’ Generative AI tools, from model design and evaluation to data governance and human oversight. By grounding models in biological and causal mechanisms, the project aims to improve robustness, reduce implausible outputs and make results easier to interpret and verify. A dedicated evaluation framework will assess scientific validity, accuracy, robustness, fairness, explainability and usability, alongside legal and ethical compliance, while transparent documentation will help researchers understand the evidence and uncertainty behind AI-generated outputs.
Privacy, transparency and responsible use are addressed throughout the project. Data will be handled in line with GDPR and FAIR principles, while bias monitoring, ethical impact assessments and external oversight will help identify and address emerging risks. Researchers, clinicians and other stakeholders will contribute to the design and evaluation of the tools, supported by training and guidance for their critical and responsible use.