The overarching goal of the EOSC-ARENA project is to develop a multi-agent GenAI system as a virtual scientific assistant, covering the entire end-to-end research-data lifecycle.
The rapid advancement and utilization of Generative AI (GenAI) models (such as Large Language Models (LLMs)) has brought significant transformative changes across various sectors, including scientific research. However, this widespread adoption has brought to the forefront issues of misutilization, quality, ethics, and integrity concerns. As AI becomes increasingly part of scientific discovery there is a clear need to align its development and use with scientific integrity, transparency, quality, and European sovereignty, so that users (i.e., researchers and innovators) can operate in an informed manner. Moreover, current AI systems, particularly those developed by industrial parties, often pose additional ethical and privacy concerns.
EOSC-ARENA (aligned with the EU AI Act) recognises that the foundation of trustworthy and ethical AI systems lies in placing human users at the center, ensuring that their needs, values, and rights are integral to system design and implementation. Beyond involving humans as supervisors (e.g., human-in-the-loop), we argue that users and developers must be taken into consideration from an holistic, sociotechnical perspective, and therefore trustworthy and ethical values must ground the new generation of researchers’ skillset.
EOSC-ARENA posits that understanding key design choices and ethical implications of AI technologies is crucial for European responsible innovation.
The underpinning concept of EOSC-ARENA is to deliver a sovereign open science environment to effectively exploit GenAI assets in the context of the EOSC, with a strong focus on interoperability at the technical and semantic level, and collaborative human-AI work in science.
EOSC-ARENA is structured in three different foundational areas that serve as the basic layer to further develop this concept.
- A set of infrastructure services and platforms, compatible with the existing AI4EOSC platform and derivatives, focused on filling the gap of GenAI application needs (e.g., federated inference services, secure information retrieval and enrichment services.)
- A GenAI agentic system and a set of specific GenAI models that can be exploited to assist researchers across their research activity, also providing access to agent and model marketplaces within the EOSC.
- Community development and skills building, focused on ethical, responsible and trustworthy usage of GenAI in scientific research, aligned with EU values.
EOSC-ARENA Foundation Areas
All together, these areas serve as enablers in order to deliver the final sovereign AI ecosystem to enable open science and collaborative Human-AI work.
Project results will be validated through 12 cross-disciplinary use cases covering life sciences, medical imaging and oncology, climate science, cosmology, materials science, food science, and research data management.
A sophisticated GenAI agentic system, with a strong focus on scientific robustness
A community building and training program focused on responsible usage, enabling researchers to perform informed decisions, aligned with EU values from an ethical and human-centric perspective;