Collaborative and Autonomous AI for Science and Innovation: Practices, Challenges, and Future Directions

Abstract

Science and innovation constitute a tightly coupled process in which scientific discovery and technological advancement continuously reinforce each other, yet contemporary progress is increasingly constrained by information overload, fragmented knowledge, slow experimentation, and difficulties in recognizing genuine innovation. Recent advances in artificial intelligence, particularly large language models, have created new opportunities to address these challenges across the science and innovation pipeline. This survey organizes existing research into two complementary paradigms: a collaborative AI paradigm that augments human reasoning, creativity, and contextual decision-making, and an autonomous AI paradigm that emphasizes scalability through large-scale knowledge mining, hypothesis generation, and integrated experimental workflows. We systematically review AI techniques and applications across key stages of the science and innovation pipeline, including knowledge acquisition and problem formulation, idea and hypothesis generation, experiment design and execution, and scientific communication and presentation, as well as integrated systems that support end-to-end science and innovation. Distinctively, this work explicitly examines the interdependence of science and innovation and highlights the role of collaboration throughout the pipeline. We further discuss key challenges in interpretability, evaluation, creativity, and the identification of substantive innovation, and outline open research directions for AI-enabled science and innovation. A curated list of related papers is publicly available at https://github.com/TJUNLP-xxy/Awesome-AI-Science-and-Innovation.

Ähnliche Arbeiten