Science publishes AI and science editorial

Science published an editorial on May 21 that asks whether modern research can remain a “trustworthy, evolutionary system” if artificial intelligence begins to “do” science in materially different ways. The piece, titled “Progression without progress,” appears in the journal’s May 21, 2026 issue and is credited to J. M. Ottino and B. Uzzi. Science Magazine also circulated the editorial on its X account on May 21. ### Which editorial is this, exactly? The May 21 table of contents for *Science* lists “Progression without progress” as an editorial on page 791 of Volume 392, Issue 6800. The same issue PDF identifies J. M. Ottino and B. Uzzi as the authors. Science’s issue materials also note that signed editorials reflect the authors’ views rather than official positions of AAAS or their institutions. (science.org) Science Magazine’s X post on May 21 linked to the editorial and described it as a piece about whether science can remain a trustworthy, evolutionary system if AI “does” science differently, matching the framing in the article itself. ### What question are the authors putting on the table? The editorial centers on a question about scientific process rather than a single laboratory result. (science.org) In the text surfaced from the *Science* PDF, the authors describe science as an evolutionary system in which independent lineages explore different directions and produce conceptual breaks that may be less likely to emerge from systems trained on the past. (science.org) The same passage says “independence is engineered away” in AI-shaped pipelines and argues that fully autonomous systems may occupy only part of the scientific landscape while still representing a broader reshaping of knowledge production. That framing places the emphasis on how discoveries are generated, checked and trusted, not on a specific regulatory proposal. ### Why do reproducibility and trust show up in this debate? (science.org) Science has already published separate guidance on AI use in its journals, and that document ties acceptable use to disclosure and confidentiality rules. The guidance says some uses of AI are allowed with conditions, including naming the tool and explaining how it was used, while entering all or part of a manuscript into an LLM is listed as not allowed in some circumstances. (science.org) A separate *Science* policy-oriented article published earlier this year argued that AI can be a “true partner in science” only if results are verifiable and if data, methods, code and outputs are available for public scrutiny. That article is distinct from the May 21 editorial, but it shows the same journal has been framing AI in research around verification, openness and reproducibility. (science.org) ### Did the editorial tell researchers or policymakers what to do next? The May 21 editorial does not appear to set out a checklist, policy package or enforcement mechanism in the material available from *Science*. Instead, it poses a structural question about whether scientific knowledge remains trustworthy if the system generating that knowledge changes in ways that reduce independence and alter how novelty emerges. (science.org) That makes the piece an argument about the character of scientific inquiry rather than a new journal rule. Science’s own issue materials distinguish signed editorials from formal institutional positions, underscoring that the article is a viewpoint by its named authors. ### Who are the named participants here? The editorial is attributed to J. (science.org) M. Ottino and B. Uzzi in the May 21 issue materials. The journal published it under the editorial label, and Science Magazine amplified it through its official social-media account the same day. Science has also published broader AI-related material in recent months, including guidance for authors and an article on “trustworthy development of AI,” showing that the journal has been treating trust as a recurring theme rather than a one-off concern. (science.org) ### Where can readers find the next relevant documents? The May 21, 2026 issue of *Science*, Volume 392, Issue 6800, contains the editorial on page 791. (science.org) Science’s AI-use guidelines for its journals remain available as a separate document, and future editorials or policy updates would typically appear through the journal’s issue pages and official account postings. (science.org)