Abstract
All scientific medical journals face similar challenges: the number of manuscript submissions vastly outnumbers the capacity of capable and knowledgable reviewers. The workload on top reviewers for any journal is always too high and their inbox is virtually never empty. Finished tasks are almost instantaneously replaced with the next, leaving a sense of frustration and overwork on the part of the reviewers. The strategies to overcome this situation are all quite alike: special motivational scores (special acknowledgements, prizes and awards), pre-screening protocols to involve reviewers only with higher quality submissions, leadership pathways to the editorial boards of journals for the top reviewers and multiple tools to facilitate their important task.
Nevertheless, long review-times to first decisions and the quality of reviews are key metrics for any journal and play an important role where potential authors submit their work.
With the seemingly exploding use of artificial intelligence in many aspects of medicine, the field of medical publishing is no different. The environment is changing rapidly, the algorithms have evolved in a way (and at an evolutionary lightspeed) unknown until just recently. The creative adaptation of workflows and tasks attributed to AI is impressive.
When reviewing a manuscript, the legal checkbox not to use AI is mandatory. And it is for a good reason. Review means „peer-review“. An honest evaluation of a peer’s scientific work. This adds tremendous credibility to the process, trust and impact. We have elaborated on this topic in this very column before. But as artificial intelligence capabilities are increasingly used in more and more important and consequential fields of medicine, will they stay out of scientific peer reviews? It might seem naive to think so. Authorship of a well composed piece of text can hardly be differentiated between human or artificial intelligence, a sad and problematic but common situation.
So what are the arguments in the discussion on AI-based manuscript reviews in medical publishing: • The speed and capacity and unbeatable. An algorithm can summarize, fact check, plagiarism-check, grammar-check any manuscript within seconds. 24/7/365. No breaks needed. • The standard evaluation of a manuscript content against the existing body of literature on the subject is also no challenge to any existing modern AI. • So the formalities of any review are facilitated greatly by AI, presumably without any loss of quality.
The true problems become clear with 2 key aspects of a scientific review of a submitted manuscript: Novelty and scientific value. AI systems are trained on existing knowledge. Publishing, however, means distribution of previously unknown knowledge against the context of relevance, plausibility and potential scientific impact. The tremendously important recommendations to reject, revise or accept a paper cannot be based on whats out there already (read: learnable by AI) but must be based on what should be published. What has the potential for improving care, understanding of previously unconnected facts, future research. This fact ensures scientific credibility and trust in medical publishing and ensures the quality of a journal.
An AI-edited journal publishing AI-generated manuscripts which have been reviewed by AI-algorithms does sound absurd for a reason. The potential and scientific value of a journal will inherently lie beyond any AI because it cannot be learned. It can only be projected by the authors´ strict but appreciative human peers in the review- and editing-departments.