Forget Plagiarism: It's Originality, Utility, and Contribution That Matter in Academia Now

Recent technological advances in AI text generation have arguably made influential concepts in academic integrity guidance and policy documents obsolete. Plagiarism is probably the most high-profile victim of AI text generation. Before we get there, let's take a step back and ask why there are prohibitions—in academic research publishing—against duplicate or redundant publications, of which plagiarized content is only one such variety. In times gone by, when academic journal print real estate1 was limited and precious, duplicate publications would have taken valuable space from other researchers, whose actual original contribution may well not have seen the light of the day because of a duplicate publication academics' conduct. This thwarted the advance of knowledge. However, this isn't the case today—digital-only journals have no real-world space limitations. Databases ensure that only people actually interested in a particular output will find it, if they choose their search terms wisely. Of course, these researchers will then also quickly discover that some content from an author, spread across multiple journal outputs, is repetitive. The odds are that they will be a bit disappointed, because their time has been wasted, and they will have to delete some of those outputs. We are not futurologists, but let us boldly place a bet that soon LLM's will assist us in eliminating duplicate content in our database searches. We don't think that these historical rationales against duplicate publications constitute a major ethical problem today. Arguably, the minor cost incurred by academic researchers is outweighed by the additional readers these outputs would have found by appearing in multiple places in some form or shape. And, if you have something truly important to say, it bears repeating, or does it not? However, at the time of writing, redundant or duplicate publications remain a problem because they waste a different kind of scarce resource, namely highly sought-after reviewer time. In the empirical sciences, they can also lead to skewed results of systematic reviews, as the International Council of Medical Journal Editors notes in its guidance on the subject 1. This latter issue will undoubtedly eventually be resolved by AI sorting, if that is considered of sufficient importance. Let us now turn to a particularly pernicious form of duplicate or redundant publication, one involving theft and plagiarism. While there is a surprising variety of definitions of plagiarism definitions, they all seem to be driven to some extent by ethical disapproval of researcher A passing off researcher B's intellectual content as their own. Historically this occurred in the research publishing context by A copy-pasting B's intellectual content into their own text without proper attribution and acknowledgment and hoping that nobody would notice. What exactly made this a wrong? A researcher who makes a discovery deserves credit for that discovery, as opposed to someone who stole that discovery from them and pretended that they furthered knowledge. In some ways this constitutes theft of intellectual property. Theft is typically a bad thing. It's also the case that, unlike the discovering researcher, their plagiarizing counterpart is not willing to be or cannot be held accountable for their work, for the apparent reason that they didn't undertake it in the first place. They are fraudsters. This is the paradigmatic case of plagiarism. This is why plagiarism prohibitions are featured prominently in institutional policies. This kind of plagiarism occurs relatively rarely today. A different kind of plagiarism has largely superseded it. The reason for this is that ChatGPT and other LLM apps permit the plagiarizing, thieving academic to easily re-write the content they wish to steal, to the point that it is impossible to demonstrate conclusively that the theft even occurred. There exists a multibillion plagiarism detection software industry, but this arms race bounces forth and back between temporary reprieve, and loss, and ultimately it will likely be lost by the plagiarism detective. The competent plagiarizer can today run their stolen content sufficiently often through LLM re-writing exercises until the detection software greenlights their efforts. This much certainly is true for research outputs not relying on empirical research. Halaweh and Refae, for instance, report what happened when they investigated the widely used Turnitin software, and similar tools. They conducted an experiment where they used texts generated by ChatGPT, and they had it evaluated by Turnitin and various other AI text detection tools. Here is their result: “Through multiple iterations and interventions, the text was paraphrased by ChatGPT until it appeared original and could not be detected as AI-generated by Turnitin's AI detection tool. The findings revealed that all the AI detection software tools that were examined failed to detect the AI-generated text by ChatGPT in the final iteration. ” 2 What does this mean for the eager plagiarizer? Well, just that, run your stolen text sufficiently often through ChatGPT and, for good measure, some additional LLMs, and no plagiarism detection software will be able to demonstrate your theft conclusively. Incidentally, and while this is probably for another Editorial, this also suggests policies prohibiting the use of LLMs in academic text productions amount to little more than virtue signaling activities, given the lack of enforcement abilities. Perhaps it is wiser to simply revert to asking authors to take responsibility for their outputs, as is the case anyway. Little is served by demanding an acknowledgment that AI had a hand in writing some output if you know you have no reliable tools to demonstrate and enforce that. It seems not too far-fetched to suggest that those who still get caught plagiarizing get caught because they have been, with respect, unusually incompetent in terms of covering their tracks. This is not to say that all academic use of LLMs is wrong. Far from it. LLMs can be used to enhance and expedite academic writing just like computers and internet searching have done when used in the right way (we will say more about that in the third part of this Editorial). It is to say that the straight plagiarizing of content from LLMs can itself be covered up. Indeed, there are different varieties of duplicate publishing. As humans, we are error-prone in our activities, and so there are also innocent varieties of redundant publications that are a result of an academic having been sloppy in their work, as opposed to being bent on stealing someone else's intellectual contribution. Quite rightly, guidelines writers in universities and research funding agencies have developed elaborate scales against which they measure the severity of the wrongdoing. If in your doctoral thesis you have two or three minor instances of unacknowledged duplicate content, you will have less of a problem, if you get caught, than if you have 50 or 100 such instances. While you may not survive your tenure as president of a leading global research university even in the former scenario, such distinctions make a lot of sense. In the former case you probably have genuinely worked sloppily, which isn't great, but to make mistakes is human, while in the latter case you are likely to have committed intentional fraud. Some prominent academics' careers have prematurely ended, because improved detection methods made it possible to discover their plagiarism from the 1970s and 1980s 3. To this point in this Editorial then, we have argued that technological advances in AI are of such a kind that successfully policing most instances of plagiarism, and other forms of duplicate or redundant publishing, has become all but impossible. Let us turn briefly to “self-plagiarism 4”. It remains popular with university administrators. However, its widespread use notwithstanding, there is no such thing. It is pretty difficult to steal your own intellectual content, even if you tried. “Self-plagiarism” is a contradiction in terms. Typically we come across “self-plagiarism” related policies in two distinct contexts. One is, unsurprisingly, in university courses. Educators want their students to produce new content for their evaluation, in each course, and so there is a prohibition on recycling content from courses they have already taken. That is a perfectly reasonable approach to education and training. It's unclear why that isn't said, and why students are misled into thinking they could steal their own intellectual property. It does not reflect well on universities that so little thought is given to terminology in their student-directed policies. The other context in which “self-plagiarism” arises is—and that is more relevant to this publication—in journal publications. This occurs when an academic uses some (sometimes a lot of) content they have published already, and presents it in a submission to a journal as if it was original content. Of course, this also doesn't constitute self-plagiarism, for the reason given already, but certainly it constitutes redundant or duplicate publication. And papers still do get retracted for that reason. It is a form of academic misconduct. We suggest then, in this part of the Editorial, that not only has plagiarism outlived its usefulness, but so has its younger sibling, “self-plagiarism. ” The latter never had a reasonable claim to existence to begin with. It is a form of redundant or duplicate publication, but the academic integrity violation isn't caused by the miraculous feat of stealing one's own intellectual content. Thinking about the ethics of this practice, it is also less obvious where the ethical problem lies. After all, the content was produced by the author of the redundant piece. There is no theft of intellectual property here. One reasonable explanation is that academic journals promise to publish original peer reviewed content, so quite likely the author would deliberately have misrepresented to the editorial team and the publisher what the nature of the output is that they submitted to a journal. Uncontroversially, this constitutes a form of academic misconduct (not theft of intellectual property). There have been carve-outs in relevant guidance documents, when journal audiences are distinctly different, and when the duplicate publication is acknowledged. An example of this is this paper 5, which was re-published in another peer-reviewed journal 6, because the editors of the latter journal were keen on having a discussion of the paper in their pages. They acknowledged the duplicate publication. Going forward, we should eliminate “self-plagiarism” in policy and regulatory documents, because prohibitions on duplicate or redundant publications fully cover the problem for academic journal purposes. Besides historically hampering the advance of knowledge, there is a second reason why plagiarism matters. We live in a “publish or perish” academic environment. Publication number and quality are key benchmarks in academic competitions for promotion, honors, and research funding. In our view this is more of a sign that the criteria chosen to establish academic excellence, in so far as they are of a simplistic quantitative nature, are not fit for purpose. They also lead to the publication of false or misleading results. Journal editors and learned societies have long since caught on to this, as documents like the Declaration on Research Assessment demonstrate 7. Universities and funders are called upon to do their homework and address this issue in a meaningful way, and—to be fair—some efforts to that account have been made. For these reasons, we suggest that it is time to put at a minimum the concepts of plagiarism and “self-plagiarism”, for regulatory and policy purposes, to rest. The latter concept never made any sense and should be abandoned. However, it would be naive to suggest that plagiarism is a thing of the past, we merely acknowledge that technological advancements are such that policing plagiarism proper today has become nigh impossible, because not only would one have to demonstrate duplicate or redundant publishing, but also intent to steal that content. Technological progress means today that detection and enforcement will be limited to some very few cases of egregious wrongdoing that were executed incompetently in recent years by the guilty party, or by researchers active during the last century, where the risks involved in trying to evade the detection of one's duplicate publications were more limited due to non-existent automated detection tools. That can still be covered by prohibitions against duplicate publications and academic misconduct, without the need to resort to plagiarism. Sadly today's tools are unlikely to reliably catch misconduct executed competently by technology-savvy academics—and students. We may well need new concepts and tools to address this misconduct in academic publishing, or—in the case of students—new performance metrics. It's time to let go of “plagiarism”. It has become functionally obsolete. While some concerns about duplicate and redundant publishing are reasonable, not least the waste of reviewers' time, and enforcement may succeed to some extent vis-à-vis empirical research outputs, that may realistically not be the case in the humanities, law and similar text-based research endeavours, for the reasons highlighted in this Editorial. We have so far argued that it is time to reject the concept of plagiarism largely due to the advent of large language models (LLMs). We argued that other existing concepts of academic misconduct are sufficient to cover the inappropriate use of LLMs. But what constitutes inappropriate use? One solution that has been used to address plagiarism is attribution: the author is required to attribute to another author or their self the redundant text. This has been extended to the use of LLMs. Statements of attribution and responsibility have been proposed: “Any use of generative AI in this manuscript adheres to ethical guidelines for use and acknowledgment of generative AI in research. Each author has made a substantial contribution to the work, which has been thoroughly vetted for accuracy, and assumes responsibility for the integrity of their contributions. ” 8 This is helpful, but does not fully address the problem. Firstly, authors can lie, as they sometimes do with human content. Secondly, just as there is vagueness about what constitutes sufficient contribution to warrant authorship (despite influential guidance documents such as those issued by the International Council of Medical Journal Editors), it is unclear how and to what extent LLMs can be legitimately used to augment academic work. This is partly due to the fact or concern that LLMs continue to hallucinate to such an extent that at the time of writing major global insurance companies have started refusing to insure companies against claims that may arise as a result of their deployment of AI 9. Academic work has three principal values: promotion of knowledge and human achievement. From the perspective of promotion of knowledge, it would not matter whether the output was from a LLM, a human or some combination. What matters is the nature and extent of the increase in useful knowledge. Indeed, since the promotion of knowledge is the primary goal of academic research, the use of LLMs might be morally required if used in an ethical manner. There is no need for limits on the use of LLMs from the point of view of increasing knowledge. Whatever amount produces the best improvements is desirable. What constitutes “best improvements” remains an open question. Let's call this the utility of knowledge. However, there is a second value that academic work reflects: human achievement. It is important to be able to evaluate and rank humans with respect to their production of knowledge. When it comes to rewarding human beings in relation to the production of knowledge, humans could be rewarded either for their production of knowledge itself or for their use of AI to produce that knowledge. Let's call this their personal contribution. Personal contribution is necessary for reward through bonuses or salary, for future investment through grants or other support and for distributing professional and social status. It is related to praiseworthiness. Praiseworthiness is a product of the value of the product produced (utility) and one's personal responsibility for that outcome, rather than luck. Responsibility in relation to praise for producing knowledge is a function of the control one exerts over the outcome (typically effort but one of us has argued that costly commitment is a better measure 10, and how original the aims were. It is not praiseworthy to make one's task unnecessarily difficult, for example, forgoing word processing to handwrite. Indeed, excellent academics use all tools available to them to pursue original research analytically. Originality—how original were the aims of the human being and how original is the product. Contribution—what did the human being do in relation to developing or executing the aims, in gathering evidence or engaging in logical reasoning or otherwise executing a method to deliver knowledge 3. Utility - how much value does the knowledge contribution add. Utility – how much value does the knowledge contribution add. We propose, while acknowledging serious concerns about current shortcoming in terms of policing such statements, that in addition to a statement outlining whether LLMs were used and taking responsibility for the overall output, authors also provide detailed statements of originality and nature of their contribution. This won't eliminate academic misconduct, but it will force the misbehaving academic to lie more aggressively about their work. This should translate into some deterrence effect. Like taking an oath, it has symbolic value, though, of course, people can lie. And, there may be capabilities developed to evaluate the veracity of such claims in a post hoc manner in the future. Given current enforcement shortcomings, it would simply mean that an author needs to pro-actively lie about their work, and in greater detail, if they wish to proceed with their academic misconduct. This increases the stakes for, and the effort required by, the researcher engaging in academic misconduct. Utility and personal contribution can come apart. We do not address how these values should be weighed, though plausibly utility is more important. There are many possibilities for personal contribution in relationship to praiseworthiness: sufficientarian thresholds, relative contribution to use of AI, etc. But that is, a topic of another editorial. To the best of our knowledge, this may be the first article to argue for a rejection of the concept of plagiarism in academia and replace it with assessments of originality, utility and contribution to provide evaluations of the value of academic outputs and assessment of the praiseworthiness of authors. U. S. conceived of the idea of rejection of plagiarism and drafted the first and second part of this Editorial. J. S. conceived of the ideas of requirements for originality and contribution, value and praiseworthiness and drafted the second part of this Editorial. Both authors edited and improved each other's draft contributions. Both authors take responsibility for the content of this output. LLMs or other AI were not used.