Everything but the Signature: A Proposal for AI-Born Journals

A journal’s distinctive contribution to scholarly communication lies neither in dissemination nor in storage, nor even in certification considered alone, but in the act by which certification is bound to the durable trust accumulated by the journal and to the reward conferred by its imprimatur: an act grounded in the judgement of the journal’s referees and editorial board, for which named editors accept answerability under uncertainty. I call that binding the editorial signature , and argue that its irreducible core is not the cognitive judgement that goes into it, which artificial intelligence (AI) may match or surpass, but the answerability of those who sign it, which no advance in capability can assume on their behalf. What surrounds the signature is infrastructure , and it is costly: on my estimate, an author spends 17 to 18 h formatting and feeding submission systems for each published paper, which comes to tens of millions of researcher-hours a year, and an AI-born workflow could bring the per-paper figure to about an hour. Since 2001, the cumulative labour in my baseline scenario amounts to some 700 million hours, or around 350,000 person-years, comparable in magnitude to Battelle’s estimated direct job-years from US human genome sequencing and its associated industry. AI is the first technology, since the modern journal apparatus took shape, that can absorb this infrastructure while leaving the signature in human hands, because it can do the delegable labour but cannot bear the answerability. Therefore, I propose a new class of AI-born journals , defined by necessary architectural features and governance controls, and I set out what such journals would owe their authors, reviewers, and readers.