Fair Use of Augmented Intelligence and Artificial Intelligence in the Preparation and Review of Submissions to the Society of Critical Care Medicine Journals: Critical Care Medicine, Pediatric Critical Care Medicine, and Critical Care Explorations

Abstract

PREAMBLE The first official journal of the Society of Critical Care Medicine (SCCM), Critical Care Medicine, launched in 1973. At that time, all submissions were typescripts. During the next decade, standalone word processors predominated as tools for the preparation of submissions. During the 1990s, personal computers became the preferred tool for the preparation of most scientific prose, including submissions for review in professional journals. In addition to word processing, personal computers expedited data analysis and the preparation of tables and illustration. The software enabling these common clerical tasks soon acquired features that guided and even enhanced preparation of submission. Such early features included spell checkers and grammar checkers, as well as “wizards” that would recommend the use of specific statistics. At the turn of the century, it was standard practice to cite in the Methods section the source and version of common office software (i.e., word processors, spreadsheets, and simple statistical programs) used in preparation of submissions. These tools augmented the intelligence of authors and are now in routine use during the preparation and refinement of manuscripts. Of note, these tools functioned (and continue to function) on personal computers even when isolated from the internet. During the 2010s, there was great progress in the development and use of artificial intelligence. Some of that intelligence is created as scholarly work, for example the creation of models using the principles and tools of machine learning and deep learning that search for relationships among data toward forecasting an outcome. Some of that intelligence is used to design and generate a research product, such as (computer) code, implantable device, or targeting drug. Use of artificial intelligence in the performance of the research itself is fair, and authors are already generally aware that their use of artificial intelligence in performing their studies must be reported as methodology. However, there is a large and growing body of artificial intelligence that is increasingly applied to the creation and review of research/literature submissions. The purpose of the following fair use statements is to clarify the position of the SCCM journals regarding the fair use of this new artificial intelligence in the creation and assessment of the prose, tables, and figures that become a submission to the SCCM journal family. STATEMENTS The tools of augmented intelligence and of artificial intelligence are human creations. They are neither independent nor sentient. They cannot be held accountable for their products. Only humans can be accountable, and therefore only humans can be named as authors or contributors to submissions. Authors are always responsible for reporting to coauthors, editors, reviewers, and readers their uses of augmented intelligence and of artificial intelligence not only in the research itself but also in the creation of their submissions. At a minimum, the reporting should include: the purpose of its use, the Why; the augmented intelligence used, the What; the way in which it was used, the How; and, which author evaluated the result, the Who? For example, “Dr. Smith submitted the text to the Generative Pretrained Transformer version 4.0 (GPT-4.0) to analyze the spelling, grammar, syntax and usage of the prose and to propose edits that conform to standard scientific English.” Such reporting is always appropriately included in the Methods section of the submission. The use of certain augmented intelligence after writing and before submission of the report is encouraged as standard practice. These augmented intelligence applications are generally used to improve the quality, completeness, and consistency of the author-written prose and of its presentation. Such applications include: embedded spell checkers and grammar checkers (e.g., as embedded in Microsoft WordTM),standalone purpose-specific electronic prose editors aimed at improving scientific English(e.g., GrammarlyTM), the use of search engines to identify relevant references based on keywords/titles/author (e.g., Google ScholarTM and PubMedTM), reference managers (e.g., EndNoteTM), reference organizers (e.g., software used to manage systematic reviews and meta-analyses), citation checkers, and plagiarism checkers (e.g., iThenticateTM). Collectively such tools aim to improve what the author(s) have written and need not be specifically disclosed. If there is uncertainty about the appropriate use or disclosure of a particular augmented intelligence tool, authors should contact the journal office at [email protected] for guidance. Certain uses of artificial intelligence during the creation of the report are strongly discouraged. These discouraged uses include creation of prose, tables, or figures; creation of reference lists to accompany prose that is either already written or being generated by computers; and creation of adjacent materials including cover letters and supplemental digital content. Electronic tools that create content often diminish the critical thinking required of authors, which is a component of responsible scientific reporting. Moreover, content created by artificial intelligence can be inaccurate (e.g., “AI hallucination”). Should any author use artificial intelligence to create content—whether that created content includes prose, displays (e.g., tables, figures, videos), reference lists —each co-author will be held jointly accountable for the accuracy and reliability of the submission. Authors must exercise good scientific judgement in the selection of citations supporting their work. Citations of unreviewed sources (i.e., “gray literature”) can increase the risk of perpetuating inaccurate content that may have been generated by artificial intelligence. Repeating here for emphasis, should artificial intelligence be used for content creation, its use must always be reported in the Methods section of the submission. Concealing the use of artificial intelligence in content creation violates scientific integrity and is regarded as scientific misconduct; disclosure is always required. Authors of a submission collectively assume all responsibility for artificial intelligence created content and of its verification. Failure to verify the creation of artificial intelligence created content before submission is scientific misconduct. 5. Reviewers and editors acknowledge that submissions under review are privileged communications. Communicating any component of that submission to online services, including Large Language Models (LLM, which currently includes such tools as GPT-3.5, GPT-4, Google Bard and Pathways Language Model [PaLM], Cohere, Falcon, and Claude v1) violates the confidentiality privilege, as these and similar LLM can incorporate all user interactions and materials into their data stores. While reviewers constructing reviews are permitted to use augmented intelligence to improve readability of their reviews just as authors are encouraged to use augmented intelligence tools to improve the quality of their submissions, only authors may use LLM in the preparation of their submissions. Authors should exercise care with any use of LLMs during submission preparation because of the risk to legal copyright. 6. Once submissions are provisionally accepted, the managing editor and the editor-in-chief and their designates have a shared responsibility to verify the quality of papers that will reach publication. Prior to publication, fair use of augmented intelligence and of artificial intelligence includes at least the following: plagiarism checking, reference verification, and assessment to determine whether any part of the submission is an unacknowledged use of augmented intelligence. SUMMARY Computers are indispensable tools in clinical care, in research, and in preparation of submissions for review and publication. Use of computers to enhance insight, to foster hypothesis generation, to facilitate analyses, to illuminate associations, to identify relevant references, and to improve the quality of prose and of presentations are collectively fair. Use of electronic tools as substitutes for critical thinking and writing is strongly discouraged. Failure to acknowledge the specifics of use of augmented intelligence and artificial intelligence in the creation of submissions to the SCCM journal family will be considered a violation of scientific integrity. Authors are referred to the statement on Error, Fraud, and Responsibility previously published (Buchman TG, et al. Crit Care Med 2023 Sep 1;51(9):1105−1107). COMMITMENT The editors of the SCCM journals acknowledge that the evolution of augmented intelligence and of artificial intelligence and of their application in all aspects of research, reporting, and publication is accelerating. The editors of the SCCM family of journals are monitoring that evolution and are committed to revising this policy statement when change is warranted. Authors, reviewers, and readers are invited to recommend consideration of changes at any time by written communication to [email protected]. Please insert “Fair Use of AI” in the subject line for all electronic communications about this policy. Communication of a recommended revision does not constitute authorization to deviate from this policy.

Ähnliche Arbeiten