AIgiarism is plagiarism: artificial intelligence can (be perceived to) plagiarize and can also be plagiarized

Article information

Sci Ed. 2024;.kcse.346
Publication date (electronic) : 2024 September 13
doi : https://doi.org/10.6087/kcse.346
Department of Biochemistry, Yong Loo Lin School of Medicine, National University Health System, National University of Singapore, Singapore
Correspondence to Bor Luen Tang bchtbl@nus.edu.sg
Received 2024 August 1; Accepted 2024 August 8.

A recent news feature in Nature discussed how plagiarism has become an even more complex issue due to the emergence of artificial intelligence (AI), particularly freely available generative AI tools based on large language models (LLMs) such as OpenAI’s GPT 3.5/4 and Google’s Gemini [1]. The article highlighted a particularly interesting point of confusion or contention—namely, whether “[…] using unattributed content written entirely by a machine—rather than by a human—counts as plagiarism.” Although some view AI-assisted plagiarism, for which the term “AIgiarism” has been coined, as involving acts of misconduct, it appears that others might not think so. For instance, the European Network for Academic Integrity (ENAI) has defined the prohibited or undeclared use of AI tools for writing as “unauthorized content generation” rather than as plagiarism as such [2], and the researcher Deborah Weber-Wulff was quoted as opining that “plagiarism, for me, would have things that are attributable to another, identifiable person” [3].

A typical scenario involves a human providing short prompts for a generative AI to generate academic content (with proper credit attribution [i.e., quotes and reference citations] to original authors or otherwise) and using the output in reports and publications (either verbatim or with only minor/cosmetic changes). The question posed above could potentially be differentiated into two parts. The first is whether AI could in fact plagiarize the work of others, and the second is whether work by an AI entity could itself be plagiarized. It could be argued that the answer to the former question would appear to be negative, because a machine or an algorithm would not act autonomously without any prompts. The latter possibility remains somewhat more plausible, as a human author could easily pass off AI-generated work as his or her own, and AI chatbots are in any case not permitted to assume authorship as a community-wide consensus. However, arguments against characterizing the above act(s) as plagiarism center on the perception that an AI platform cannot be considered a person, and as such cannot expected to seek credit or benefit from citations of its work. Thus, if an AI system cannot initiate content generation unless prompted by a human, it cannot commit plagiarism. Furthermore, since AI is not human, its work cannot be plagiarized because it would suffer no loss of credit or benefits as a result. Following this logic, verbatim usage of AI-generated content would not be considered plagiarism.

I argue that such reasoning is flawed and offer the following two counterarguments. Firstly, although an AI tool cannot generate content unless prompted, it could nonetheless be viewed or recognized as a plagiarizer under certain conditions, such as when it is named as an author, or when its contribution to content generation is made known. If an AI system is identified as an author or contributor to an article, it would need to have the corresponding accountability for the authenticity and veracity of the content generated. This is a good reason why AI entities should not be named as authors, but their use should be explicitly declared. As such, an AI system can potentially plagiarize, or be perceived to have done so, if its involvement in content generation is known.

My second counterargument, that an AI tool can itself be plagiarized, rests upon the “generative” nature of AI entities based on LLMs, and their capability to quickly produce complex, voluminous, as well as grammatically and semantically accurate contents with only simple, short prompts. Such AI entities are definitely not human, but nonetheless have humanlike attributes (and superior computational prowess), at least in terms of academic content generation. In fact, GPT-4 has passed the Turing test [4] and will likely also pass tests on selfawareness [5]. While AIs are not capable of true conscious experiences, or emotions and feelings, advanced AI could potentially approach the cognitive status of a philosophical zombie [6] equivalent (i.e., physically or functionally indistinguishable from humans, but not conscious) when it comes to writing papers. Although an AI system would likely be devoid of any credit- or benefit-seeking mindset and would not be emotionally or materially compromised when its work is stolen by humans, the very fact that a generative AI could produce highquality content of apparent academic value (which no previous applications or algorithms could) makes its product desirable by interested human parties. It is this capacity that makes the generative AI “plagiarizable” compared to other literaturesearching or language-editing algorithms that are not contentgenerative. Grindrod has recently argued that while AI-generated content is not associated with mental intentionality, it could nonetheless have linguistic intentionality, and in the latter regard could be considered potentially meaningful [7]. For the human to then use the AI entity’s work without attribution would be wrong, not because credit or benefits would thus be lost on the part of the AI entity, but rather because the human’s gain of credit and benefits was neither owned nor deserved. Much like acts of plagiarizing materials produced by other humans, a lack of true understanding and intellectual ownership of the materials thus acquired by AIgiarism would have repercussions, particularly if there are errors and inaccuracies within the content, which would likely occur due to “hallucinations” by AI.

I have previously noted elsewhere that AIgiarism is a form of bypass plagiarism [8], and that “[…] the wrong of AIgiarism is not so much of inappropriate credit allocation, but the potential propagation of factual and interpretive errors as well as biases, which undermines knowledge acquisition and understanding” [9]. It should be clear from the above discussion that AIgiarism is not only justifiably wrong but is indeed a legitimate form of plagiarism. It should also be clear that for both AIgiarism and other more conventional acts of plagiarism, the ultimate fault lies with the human plagiarizer whose desire to cut corners eventually leads to misconduct. If the linguistically proficient generative AI still lacks a mind, the human plagiarizer would then be, at least metaphorically, the proverbial “ghost in the machine” [10] who instigates and eventually completes the act of plagiarism.

Notes

Conflict of Interest

No potential conflict of interest relevant to this article was reported.

Funding

The author received no financial support for this article.

Data Availability

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Supplementary Materials

The author did not provide any supplementary materials for this article.

References

1. Kwon D. AI is complicating plagiarism. How should scientists respond? Nature 2024. Jul. 30. [Epub]. https://doi.org/10.1038/d41586-024-02371-z.
2. Foltynek T, Bjelobaba S, Glendinning I, et al. ENAI recommendations on the ethical use of artificial intelligence in education. Int J Educ Integr 2023;19:12. https://doi.org/10.1007/s40979-023-00133-4.
3. Weber-Wulff D, Anohina-Naumeca A, Bjelobaba S, et al. Testing of detection tools for AI-generated text. Int J Educ Integr 2023;19:26. https://doi.org/10.1007/s40979-023-00146-z.
4. Bergen BK. People cannot distinguish GPT-4 from a human in a Turing test [Preprint]. arXiv [posted 2024 May 9]. https://doi.org/10.48550/arXiv.2405.08007.
5. Berglund L, Stickland AC, Balesni M, et al. Taken out of context: on measuring situational awareness in LLMs [Preprint]. arXiv [posted 2023 Sep 1]. https://doi.org/10.48550/arXiv.2309.00667.
6. Kirk R. Zombies [Internet]. Stanford Encyclopedia of Philosophy 2023. [cited 2024 Aug 1]. Available from: https://plato.stanford.edu/archives/fall2023/entries/zombies/.
7. Grindrod J. Large language models and linguistic intentionality. Synthese 2024;204:71. https://doi.org/10.1007/s11229-024-04723-8.
8. Taylor JS. Reassessing academic plagiarism. J Acad Ethics 2024;22:211–30. https://doi.org/10.1007/s10805-023-09478-4.
9. Tang BL. The underappreciated wrong of Aigiarism: bypass plagiarism that risks propagation of erroneous and bias content. EXCLI J 2023;22:907–10. https://doi.org/10.17179/excli2023-6435.
10. Koestler A. The ghost in the machine Reprint edth ed. Penguin Random House; 1990.

Article information Continued