Innovation Lab : Explainable and validated AI
Question:
What technologies and methodologies are needed to develop robust, transparent, and clinically validated AI algorithms for predictive healthcare tailored to individual patient profiles?
Proposed solution:
To develop transparent AI algorithms for predictive healthcare, one approach could consist in creating a LLM-based AI agent to provide explainability for predictions generated by multiple AI tools. This agent would serve as an intermediary to communicate and justify algorithmic outputs to clinicians (and to patients). This would reduce barriers to adoption, enhancing interpretability, and building trust in AI-assisted decision-making processes.
The AI agent should be developed and validated by a group of transversal experts in accordance with international standards (e.g., ISO guidelines for medical software and AI), ensuring clinical relevance and compliance. Its training pipeline could incorporate Generative Adversarial Networks (GANs) combined with human-in-the-loop feedback to refine outputs iteratively and improve contextual accuracy and usability.
