The proposals shall focus on all following aspects:

• Enrich the EOSC federation with Generative AI tools for evaluating research data quality, ensuring trustworthiness across the European network of trusted repositories, accessible by humans, machines, and Generative AI services: formulate protocols and policies to facilitate effortless data access, processing, and provenance updates within EOSC’s repository and service network.
• Support European research infrastructures to improve the FAIRness of their data, so that they are ready to be combined with data of infrastructures in scientifically neighbouring domains, in order to provide Generative AI-ready data.
  • conduct pilots to validate the effectiveness and accuracy of the Generative AI-driven data quality evaluation methods, iteratively improving and refining them based on feedback and real-world use cases and removing the potential biases inherited from the training data.
• Run community engagement and support programmes for implementing Generative AI in scientific workflows via EOSC:
  • promote a sound training programme to facilitate the uptake and the use of Generative AI as a means to facilitate the FAIRification of data and data curation;
  • demonstrate how Generative AI can facilitate quality assessment of FAIR data;
  • advance the realization of machine-actionable (MA) research data and services, including AI-based systems;
  • propose protocols and policies to govern automatic data workflows within the network of repositories and services.

The proposals are expected to deliver on one or more of the following:

• Develop, promote and support real-life use cases for Generative AI models in scientific research domains, in line with the GenAI4EU initiative and other key EU initiatives, like the Apply AI strategy, such as:
  • augment datasets in scientific fields that rely on image analysis, such as biology, astronomy, and materials science: by generating synthetic images that closely resemble real data, researchers can expand their datasets, improve model robustness, share anonymized version of sensitive data and generalize better to unseen scenarios;
  • learn the underlying patterns of complex time-series data, such as sensor readings in environmental monitoring or physiological signals in healthcare: by generating data samples that match the learned distribution, these models can detect anomalies or deviations from normal behaviour;
  • accelerate materials design and discovery by predicting the properties of new materials without the need for extensive experimental testing: these models can generate novel material structures with desired properties, such as strength, conductivity, or catalytic activity, based on learned relationships between material compositions and properties;
  • advance drug design and molecular modelling by generating novel molecular structures with desired pharmacological properties: these models can explore vast chemical spaces, predict the interactions between molecules and biological targets, and optimize drug candidates for efficacy and safety;
  • simulate complex systems and phenomena in various scientific domains, such as physics, chemistry, and ecology: by capturing the underlying dynamics and interactions of the system, these models can generate realistic simulations that mimic observed behaviour or predict future outcomes under different conditions.