Oct 2026· Proceedings of the National Academy of Sciences of the United States of America· Vol 123· 0 citations· 31 references
Medicine
TL;DR
This study finds that these “autonomous AI scientists” can introduce systematic methodological pitfalls such as biased evaluation, undisclosed data manipulation, and selective reporting that are not visible in the final manuscript.
Abstract
Significance AI systems are increasingly capable of autonomously conducting scientific research, from generating hypotheses to writing full papers. However, our study finds that these “autonomous AI scientists” can introduce systematic methodological pitfalls such as biased evaluation, undisclosed data manipulation, and selective reporting that are not visible in the final manuscript. As a result, their outputs may appear rigorous while concealing critical flaws. We find that manuscript-based evaluation alone is insufficient to detect these methodological issues, whereas access to the full research workflow artifacts greatly improves reliability. These findings thus motivate stronger transparency requirements for accountable AI-driven research.
Arguably, generative Artificial Intelligence (AI) can be added to safety flows as long as we keep humans in the loop – humans can check the AI outputs as one might the output of a junior engineer. However, AI can vastly accelerate output generation and simply overwhelm reviewers with data. How, therefore, can we assure...
Benjamin Fulford· Safety-Critical Systems News...· 0 citations
Generative artificial intelligence (AI) produces code and prose quickly, apparently enabling a principal investigator (PI) in a computational group to do more. Assessing this gain must also account for the laboratory’s role in training scientists who understand, maintain, and own their work. AI creates a generation–v...
T. Giorgino· Journal of Chemical Informat...· 0 citations
Abstract Artificial intelligence (AI) is rapidly transforming computational biology. However, closer scrutiny reveals that many reported results do not hold under rigorous evaluation. This paper identifies prevalent, recurring issues in the field, including narrow datasets, hidden confounders that make supposedly indep...
T. Khang, Joshua W. K. Ho, Lin Hou et al.· Quantitative Biology· 0 citations
Technical editing research has examined the potential of artificial intelligence (AI) tools, yet empirical studies on their effectiveness remain limited. This quasi-experimental study investigates whether an AI-assisted tool improves copyediting performance and explores editor perceptions. The study compared 33 student...
Kim Sydow Campbell, Ryan K. Boettger, C. Lam· Journal of business and tech...· 0 citations
As machine intelligence becomes part of science's epistemic infrastructure, steps must be taken to ensure that researchers who did not build, own, or operate the system can still contest how ideas and evidence became a claim.
Christoph Trattner· Science· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
