Generative AI: strategies and use cases for astronomy

Le résumé fourni par la source

Generative AI (GenAI) promises a disruptive impact across multiple research and operational activities, including the domain of astrophysics and the development of astronomical instrumentation. This rapidly evolving tech nology requires a cross-disciplinary, systemic approach to ensure effective integration into astronomical research workflows. Its adoption also requires strict governance to ensure scientific reproducibility while addressing ethical and regulatory concerns. Gen.IA, a dedicated study group within the Italian National Institute of Astrophysics (INAF), proposes a structured strategy to bridge the gap between experimental AI and operational deployment. As a first step, the study group launched an internal survey1 to assess current adoption levels, expectations, and ethical sentiments about AI within the Institute; we present results from this survey. We also outline a strategic roadmap involving: knowledge-sharing frameworks (internal meetings, web resources, newsletters) to disseminate best practices and emerging standards; upskilling for both technical experts and general staff to foster responsible adoption; seminars with external experts to address ethical and regulatory concerns. Additionally, we propose specific use cases for investigation: Retrieval-Augmented Generation (RAG) systems for mining complex technical documentation and driving requirement management, improving traceability and compliance in instrument development and system engineering workflows; Agentic AI for orchestrating automated data analysis, software development, and optimizing research pipelines; AI-assisted literature review to accelerate knowledge discovery in astrophysical domains. This work proposes a strategic framework for using GenAI to enhance astronomical research activities and instrument development, emphasizing the balance between innovation, best practices, and scientific integrity.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.