Research actions under this topic should include all the following activities, ensuring multidisciplinary approaches and a broad representation of stakeholders in the consortia (e.g. industry, academia, healthcare professionals, patients):

• Develop virtual assistant solutions based on new or optimised trustworthy and ethical Generative AI models, augmented by other AI tools to support healthcare professionals. The models should leverage extensive and diverse multimodal health and research data, public knowledge, and reliable healthcare systems information relevant for healthcare settings. Examples can include electronic health records, medical imaging, genomics, proteomics, molecular data, laboratory results, patient information (including on safety), and/or unstructured health data (the applicants may choose any type of available large-scale data). The development and training of the models should take place in multinational consortia and federated governance approaches should be considered. The applicants should demonstrate how the project goes beyond combining existing data and generates new specific knowledge to improve clinical decision making.
• Demonstrate the added-value and clinical utility of the virtual assistant solutions in at least two healthcare use cases in different medical fields and unmet needs showing e.g. improved care management and efficiency, prediction of potential patient-specific therapeutic strategies and outcomes, etc. The applicants should provide evidence of high maturity technology for the use cases and assess the relative effectiveness of the solutions compared with standard of care, including on why these solutions would be superior to other AI tools and would deliver better outcomes. They should actively engage healthcare professionals as end users, and other stakeholders such as patients, caregivers in the development and testing of the solutions, ensuring that diverse perspectives and intersectional considerations are integrated throughout the process. Training and education activities for healthcare professionals should be organised.
• Develop a regulatory strategy/interaction plan with regulators (including in the area of Health Technology Assessment) for generating evidence, where relevant, in a timely manner. Consider also the potential for future regulatory impact of the results and sustainability aspects.
• Develop or adapt existing methodologies for continuous assessment of the developed solutions. The methodologies should demonstrate technical robustness, healthcare utility and trustworthiness of the Generative AI-based solutions, by adopting:
  • Appropriate metrics for evaluating alignment with human values, ethical principles and the intended purposes of Generative AI models, performance including testing their technical robustness and clinical utility, as well as their model intelligibility, in view of ensuring AI trustworthiness.
  • Appropriate solutions to identify and mitigate potential bias of the models (e.g. representativeness of the data, bias of the trainer, bias of training and validation data, algorithmic discrimination and bias including gender bias etc.).
  • Appropriate techniques to discover and demonstrate explainability of model reasoning, increase users’ trust, and address the black box element, thus further enhancing transparency, model explainability and alignment.
  • Methods to systematically address and assess ELSI (Ethical, Legal and Societal Implications), including data privacy concerns and risk of discrimination/bias (not limited to sex, gender, age, disability, race or ethnicity, religion, belief, minority and/or vulnerable groups). The implication of medical errors originated from AI-assisted decision-making and the effects on potential legal liability for healthcare professionals should be explored.

All proposals should demonstrate EU added value by focusing on the development and/or use of trustworthy Generative AI models developed in the EU and Associated countries, involving in the consortium EU industrial developers, including leading-edge startups when possible. An open-source approach is encouraged when technically and economically feasible. Successful proposals are encouraged to utilise the resources offered by the AI factories, when relevant and in accordance with the specific access terms and conditions.