The Collaborative Exoskeleton of AI Science (asimovaddendum.substack.com)

A recent exploration highlights the urgent need for AI systems in scientific research to integrate existing scholarly infrastructure to mitigate issues like hallucinated citations and reliance on retracted papers. As AI increasingly assists in generating academic content, researchers have found that AI tools often produce fabricated references—studies indicate only 25% of AI-generated citations are accurate. Furthermore, many AI tools lack the necessary integration with databases that track retractions, leading to the propagation of flawed literature. This has raised alarms within the AI and scientific communities that AI's current deployment may undermine the integrity of scientific research. To address these concerns, initiatives like MIT's VRAIX project are proposing a reconfiguration of infrastructure akin to how GitHub supports software development. By leveraging existing resources such as DOIs for document identification, ORCID for researcher verification, and OpenAlex for scholarly knowledge graphs, the goal is to create a collaborative exoskeleton for scientific publishing. This approach seeks to ensure that AI-generated work adheres to established norms of integrity and reliability by situating papers within the broader web of knowledge. As the "AI for science" market continues to grow, there is a critical call for AI firms to not only utilize these infrastructures but also contribute to their sustainability, fostering a more trustworthy relationship between AI technology and scientific integrity.