AI and the open sharing of ideas

AI and the open sharing of ideas _Chair's notes for the session corpus, 7 October 2026. Sections are labeled as fact or as argument._ The argument (the chair's) Measurement has always shared its ideas openly before they are finished: posters, talks, conference papers, slides behind a QR code. That openness worked partly because turning someone else's idea into a working system took real time and expertise. AI removes most of that cost. A public artifact can now be turned into a detailed derivative (architecture, design, cost estimates, research agenda) in minutes, by anyone with a phone. Ideas of value to the testing industry can be captured as easily as taking a picture and asking a few questions. This complicates the profession's norms of open sharing, attribution and credit. It may also change incentives: if sharing early means being scooped, people may share less. The evidence (fact) On the evening of 6 October 2026, the chair photographed the QR code on an AIME-Con 2026 poster. The poster was "Toward Self-Improving Automatic Item Generation: Validating LLM Evaluators with Production Feedback" (Han, Gopalakrishnan, Kollbocker, Goculu, Palaghita, Zheng; Khan Academy and Google.org; presented Tuesday 6 October). The code led to a PDF of the poster, and the chair asked the ChatGPT phone application a few questions about it. The result is a 5,000-word document in the corpus, "From Self-Critique to Closed-Loop Automatic Item Engineering". It contains: - a system architecture; - critic designs and a data model; - a measurement model for LLM raters; - inference and development cost estimates ($100,000 to $5 million); - a staged research and product agenda. Checked against the poster, every figure it attributes to the authors is accurate, and so is its author list. Everything beyond the poster's findings is the AI's extrapolation, not the authors'. The document is not posted publicly. Two observations: - The derivative is competent. That is the point: it is plausible enough to act on. - It credits the original authors, but only because it was asked to summarize their poster. Nothing in the process requires a derivative to cite its source.

Two observations: - The derivative is competent. That is the point: it is plausible enough to act on. - It credits the original authors, but only because it was asked to summarize their poster. Nothing in the process requires a derivative to cite its source. Questions this raises for the panel - **Attribution and credit.** When an AI-derived plan builds on a poster, who gets credit for the resulting system? Does citing the poster suffice? - **What "public" means.** A poster is public, but it was shared on the assumption that building on it would take time and expertise. Briggs's slides note that journals bar reviewers from putting confidential manuscripts into AI tools. Public conference material has no equivalent norm. Should it? - **Consent.** Should authors be asked before their shared work is fed to an AI to produce a derivative, as opposed to read by a person? - **Accountability.** If a derivative misstates the original, who answers: the person who prompted it, or the tool? (The session's own record faces the same question.) - **Incentives.** Will researchers, especially in industry, share less, or later, if sharing is the first step of someone else's product? - **Connections:** - guiding question 3 (what must be verified, documented and disclosed when AI contributes materially); - guiding question 5 (where accountability remains human); - the Leiden Declaration's emphasis on attribution and human authorship. A candidate norm in the session's form (for discussion) - **Activity:** using AI to analyze, extend or build on another researcher's publicly shared work (poster, talk, slides, preprint). - **Conditions:** - the original work is cited prominently in any derivative that is shared or used; - the derivative is labeled as AI-assisted, separating the original's findings from the extrapolation; - commercial use of a derivative of unpublished work is discussed with its authors first. - **Answerable:** the person who prompts the derivative and shares or uses it.