The integration of generative AI into scientific research accelerates code generation and data analysis, leading to a rapid increase in publication volumes. Between 2020 and 2025, the annual number of scientific papers doubled, a trend that intensified following the public release of ChatGPT. This surge has extended the peer-review process, with the time between submission and publication increasing from several months to a year.
A 2025 study by researchers from the University of Michigan, the University of Tennessee, and Sandia National Laboratories surveyed 527 scientists regarding their use of AI. The participants, who had a median of six years of programming experience, primarily used AI for data processing (23.9%), visualization (19.8%), debugging (17.4%), scientific computing (11.7%), and statistical analysis (10.9%).
Verification methods for AI-generated results remain limited. The survey found that 52.8% of participants simply ran the code, while only 2.8% used automated tests and 2% relied on peer verification. Researchers noted that less experienced programmers tend to trust AI outputs more, despite having less capability to detect potential errors in the generated logic or methodology.
The reliance on AI in research poses risks to the integrity of scientific findings, as peer reviewers face an overwhelming volume of submissions. Experts warn that automating the review process requires extreme caution to avoid creating agents that merely validate flawed AI-generated content. Without rigorous independent verification, the scientific community risks an increase in published but unverified or incorrect results.
