Accountability in Research-Policies and Quality Assurance最新文献

Background: In this paper, we explore the question "Why can't AI be a coauthor?" and reveal a rarely discussed reason.

Methods and results: First, allowing AI to be a coauthor disregards the uniquely human experience of writing texts. This means that human authors are seen as mere producers of texts rather than rational beings engaged in a value-added and humanized learning process expressed through the paper. The relationship between the human author and the thesis is reduced to a mere result of generation rather than a result of individual human critical thinking. Second, allowing AI to be a coauthor leads to self-delusion about one's own rationality and thus violates the responsibility to understand the world correctly. In this process of self-deception, it is not as if those who grant AI coauthor status do not realize that AI is not the same as humans; however, they self-deceivingly assume that AI has the same internal state as humans. This means that the relationship between the author and the work is no longer seen as a position to be respected, but as something probabilistic and gamified.

Conclusions: Finally, we discuss the potential consequences of these rationales, concluding that including AI as a coauthor implies a disregard for humanization.

Background: Universities are increasingly offering training in research integrity (RI) to enhance research quality and foster RI. Despite the importance of integrating scientific virtues into such training, there is a lack of assessment of virtue ethics-based RI training.

Methods: This was a randomised controlled study assessing the impact of a virtue-based training for RI, performed at the University of Split School of Medicine in 2020-2021. We included first-year medical students who were randomly assigned to a control group, receiving a RI lecture, or the interventional group, receiving the same lecture plus a virtue-based training for RI. We measured changes in Ethical Climate Questionnaire (ECQ) scores as the primary outcome and the difference between perceived and desired ECQ scores as the secondary outcome.

Results: Of 181 participants, 105 (55 control, 50 experimental) completed the study. The virtue-based training did not significantly change ethical climate perceptions between groups. Dominant climates were Company rules and procedures and Laws and professional codes. Overall, the preferred climates emphasized Team interest and Social responsibility.

Conclusions: This study demonstrates the value of using ECQ to assess RI training and highlights the need for further research into the long-term effects of virtue-based training.

Research is conducted in workplaces that can present safety hazards. Where researchers work in laboratories, safety hazards can arise through the need to operate complex equipment that can become unsafe if faulty or broken. The research literature also represents a workplace for millions of scientists and scholars, where publications can be considered as key research equipment. This article compares our current capacity to flag and repair faulty equipment in research laboratories versus the literature. Whereas laboratory researchers can place written notices on faulty and broken equipment to flag problems and the need for repairs, researchers have limited capacity to flag faulty research publications to other users. We argue that our current inability to flag erroneous publications quickly and at scale, combined with the lack of real-world incentives for journals and publishers to direct adequate resources toward post-publication corrections, results in the research literature representing an increasingly unsafe workplace. We describe possible solutions, such as the capacity to transfer signed PubPeer notices describing verifiable errors to relevant publications, and the reactivation of PubMed Commons.

Objective: This study aimed to compare the accuracy of metadata for retracted articles in Retraction Watch Database (RWD), PubMed, and Web of Science (WoS).

Methods: Twenty general internal medicine journals with an impact factor > 2 were randomly selected. RWD, PubMed, and WoS were used to retrieve all retracted articles published in these journals. Eight metadata variables were examined: journal, title, type of article, author(s), country/countries of affiliation, year of publication, year of retraction, and reason(s) for retraction (assessed only for RWD, as this information was unavailable in PubMed and WoS). Descriptive analyses were conducted to document errors across databases.

Results: Thirty-five retractions were identified, and 280 metadata entries (35 × 8) were analyzed. RWD contained the most metadata errors, affecting 16 articles and 20 metadata entries, including seven errors in year of publication, six in article type, six in author names (five misspellings, one missing two authors), and one in country of affiliation. WoS had one error (a missing author), and PubMed had none.

Conclusion: The relatively high error rate in RWD suggests that researchers should cross-check metadata across multiple databases. Given the preliminary nature of this study, larger-scale research is needed to confirm these findings and improve metadata reliability in retraction databases.

Background: We consider a research model for manuscript evaluation using a two-stage process. In the first stage, the current submission reminds reviewers of previous reviewing experiences, and then, reviewers aggregate these past review experiences into a kind of norm for assessing the scientific contribution and clarity of writing required for a manuscript. In the second stage, the reviewer's norms are imposed on the manuscript under review, and the reviewer's attention is drawn to discrepancies between the norm retrieved from previous similar peer review experiences and the reality for this submission.Methods: Five research hypotheses were integrated into this research model. In our study, we tested these five research hypotheses for statistical differences among reviewers by gender, experience, and academic rank using an online survey. There were 573 respondents.Results: We did not find significant differences among reviewers in their basic behavioral patterns. The only exception was that the low-rank reviewers agreed with the first hypothesis "H1: Selective norm" to a greater extent than the high-rank reviewers.Conclusions: The interaction between a reviewer's past review experiences and the actual scientific contribution and writing clarity of the manuscript under review can explain the lack of consistency among different reviews for the same manuscript.

Retraction, as a post-publication quality control measure increasingly adopted by mainstream journals, has been observed in a few potential predatory journals (PPJs), but the extent and handling of retractions by PPJs in general remain unclear. This study investigated retraction practices among the 1,511 standalone PPJs on the updated Beall's List. Data from the Retraction Watch Database revealed that only 46 of the PPJs, including 18 indexed by the Web of Science (WoS) Core Collection, had retracted a total of 645 publications as of 2022. The retraction handling performance of these PPJs was evaluated in terms of publicity of retraction policies, availability of retraction documents, visibility of retractions, and informativeness of retraction notices. Overall, the retracting PPJs performed poorly against these criteria and showed a trend of inadequate documentation of retraction policies and documents over time. A positive correlation was found between WoS inclusion and retraction handling performance of the PPJs except for the publicity of retraction policies. These findings suggest that retraction handling performance could serve as an additional important criterion of journal editorial practices and highlight the desirability of evaluating journal legitimacy in terms of post-publication quality control through retraction.