In the digital age, the production and dissemination of content have accelerated exponentially, providing countless benefits but also posing significant challenges. Among these challenges, academic integrity and the originality of content have become increasingly susceptible to practices of plagiarism. In this context, artificial intelligence, with its capacity to process large amounts of data and understand complex linguistic structures, emerges as a tool to address the problem efficiently.
Traditionally carried out by humans, plagiarism detection has become overwhelming due to the massive volume of information available online. Artificial intelligence, specifically models like ChatGPT-3 and 4, among other generative AI tools, present themselves as a potential solution by offering advanced natural language processing capabilities. This raises crucial questions about how these technologies can improve the effectiveness of plagiarism detection, but also about their limitations and ethical challenges.
In this article, we will explore the role of ChatGPT-3, developed by OpenAI, as a language model based on artificial intelligence for plagiarism detection. We will analyze whether the technological tools created for this chatbot contribute to the identification of non-original content, examining both its successes and limitations. Understanding these aspects is essential to assess the suitability of artificial intelligence in preserving academic integrity and originality in content production in the digital age.
Artificial Intelligence and Copyright
In recent years there has been interest among countries in defining what artificial intelligence (AI) is. UNESCO, in its publication on ethical recommendations in AI, seeks to approach some definitions:
Information processing technologies that integrate models and algorithms that produce learning capacity and information in cognitive tasks that lead to outcomes such as prediction and decision-making in virtual environments (UNESCO, 2021).
For its part, the European Union (European Union, 2018) defines AI through its current regulations:
A machine-based system that is designed to operate with different levels of autonomy and that can, for explicit and implicit objectives, generate outcomes such as predictions, recommendations, or decisions that influence physical or virtual environments.
Likewise, the OECD accelerated its AI working groups, which developed principles on AI. Its definition is as follows:
Machine-based systems that can, for a given set of objectives defined by humans, formulate predictions, recommendations, or decisions that influence real or virtual environments.
It is also important to note that several academics have long addressed this topic. A prominent example is the early definition of Artificial Intelligence (AI) proposed by Alan Turing, where he conceptualizes AI as:
The science and engineering of creating intelligent machines, especially intelligent computer programs.With the public release of ChatGPT in November 2022, marking a milestone in the field of artificial intelligence, some challenges arise concerning attribution and copyright. The complexity lies in discerning who deserves to be recognized as the legitimate author of these creations: Should rights be attributed to the AI itself? To those who feed it their data? To the creator of the algorithm? To the owner of the software program that runs it? To the people involved in its training? Or to the user who guides it in its application? These questions not only pose legal challenges but also explore the limits and ethical responsibilities in developing and using technologies like ChatGPT, requiring careful analysis to find a fair and equitable balance in the assignment of copyright in this emerging domain of algorithmic creativity. To analyze this case, the copyright regulations in force in our jurisdictions are highlighted. The most important regulation to protect copyright is the Berne Convention (Berne, 1886), which defines copyright as an exclusive and legal monopoly granted by the State to the person who creates a work. This right is granted to people who generate original works, giving them exclusive control over the reproduction, distribution, and display of such works for a defined time. The purpose of this right is to stimulate creativity and innovation by providing economic incentives to creators, allowing them to share their works with the public without concerns of unauthorized appropriation.
In the Paraguayan context, copyright and intellectual property are found in the National Constitution of 1992, in Chapter IX “Economic Rights and Agrarian Reform”, Section I “Economic Rights”, in Article 110 which states that:
All author, inventor, producer, or merchant shall enjoy exclusive ownership of their work, invention, trademark, or trade name, in accordance with the law.
Regarding laws specifically related to copyright, there is Law 1328/98 and its amendment 5247/15. In Article 3 of Law 1328/98, copyright is defined as follows:
The protection of copyright applies to all works of the intellect, of a creative nature, in the literary or artistic field, regardless of genre, form of expression, merit, or purpose, nationality or domicile of the author or right holder, or place of publication of the work.
The rights recognized in this law are independent of the ownership of the material object in which the work is incorporated, independent of the method of initial or subsequent fixation, and their enjoyment or exercise shall not be subject to the requirement of registration or compliance with any other formality.
Additionally, the same law protects the work throughout the author’s life and up to 70 years after their death. With the emergence of AI in the processes of creating texts through technologies like ChatGPT-3 and 4, once again the originality of authors is called into question. Thus, plagiarism reappears and becomes increasingly complex to detect unauthorized appropriations without proper attribution or permission from the original creator.
Biases When Detecting Plagiarism
The dialogue surrounding copyright and artificial intelligence raises complex questions that challenge not only the legal sphere but also the ethical. The appearance of increasingly sophisticated works using these technologies demands a review of current paradigms in intellectual property. The coexistence between human creativity and the ability of AIs to generate similar works invites reconsideration of the need to establish regulatory frameworks that protect the rights of creators without hindering technological innovation or creative development. Despite significant advances in AI, there are limitations that must be addressed with caution. One key limitation is the lack of deep and contextual understanding by language models like the one used by ChatGPT. Although these models can generate text coherently, they do not possess underlying knowledge or true understanding of content. This can lead to the production of incorrect or biased information: even AI tools aimed at detecting possible plagiarism can yield false positives.
Another important limitation is the dependence on training data. AI models learn from specific data sets, which means they can reflect and, in some cases, amplify biases present in those data. This raises ethical challenges, as AI can reproduce and perpetuate existing societal prejudices. For the author of The Atlas of Artificial Intelligence, Kate Crawford:
Rather, it exists in a tangible form, materializing as something composed of natural resources, fuel, labor, infrastructure, logistics, as well as historical influences and classifications. AI systems do not possess autonomy, rationality, or the ability to discern without extensive and intensive computational training, which involves the use of enormous data sets or rules, and predefined rewards. In reality, artificial intelligence, as we know it, depends entirely on a much broader set of political and social structures. Due to the capital required to build large-scale artificial intelligence and the perspectives it optimizes, ultimately, AI systems are designed to serve existing dominant interests. In this sense, AI is a certificate of power.
Artificial intelligence is neither artificial nor intelligent.
Artificial intelligence is neither artificial nor intelligent.
Rather, it exists in a tangible form, materializing as something composed of natural resources, fuel, labor, infrastructure, logistics, as well as historical influences and classifications. AI systems do not possess autonomy, rationality, or the ability to discern without extensive and intensive computational training, which involves the use of enormous data sets or rules, and predefined rewards. In reality, artificial intelligence, as we know it, depends entirely on a much broader set of political and social structures. Due to the capital required to build large-scale artificial intelligence and the perspectives it optimizes, ultimately, AI systems are designed to serve existing dominant interests. In this sense, AI is a certificate of power.
In her words:
AI is fundamentally political; we must go beyond neural networks and statistical pattern recognition and instead ask what is being optimized, for whom, and who makes those decisions. Then, we can trace its implications.
According to AI expert Meredith Broussard, we must focus on positioning AI as an algorithm programmed by a human:
To understand that a computer does not do, we need to begin to understand what the computer does well and how it works.
The implementation of a technological solution involves several technical phases based on mathematical principles. These phases contain numerous processes that must be followed systematically. These steps include data acquisition, database cleaning, evaluation of missing data presence, and integration. It is also necessary to homogenize the data, reduce relevant variables, eliminate possible dimensions of the database, and merge related variables. Once data preparation is complete, the next step is to select the appropriate algorithm for the task at hand. Choosing the algorithm involves careful consideration to ensure its effective fit to the model. After this choice, training data are used to refine the model. However, solution validation goes beyond this initial process: it is essential to subject the model to rigorous testing using different data to verify the solution’s effectiveness and generalization. In Cathy O’Neil’s words, these technical phases are essential and form part of an integral methodology that seeks to ensure the robustness and reliability of technological solutions.
So, Who Owns the Text Produced by ChatGPT?
Answering this question is not so easy, and there are still more questions and doubts than certainties. Advocates of artificial intelligence argue that these technologies are collaborative tools that expand human creativity, and that true authorship lies with those who design and train the algorithms. On the other hand, critics argue that works generated by AI should not automatically be considered the property of those who created the algorithms, because the machine has an active role in the creative process.
For example, at the end of 2023, the New York Times filed a lawsuit against Microsoft and OpenAI, accusing the companies of infringing copyright and abusing the newspaper’s intellectual property to train large language models. Microsoft invests in OpenAI and provides it with Azure cloud computing technology. According to the NYT:
These tools were built with independent journalistic content that is only available because we and our colleagues report, edit, and verify it at high cost and with considerable expertise.
The Times said in its statement that it recognizes the power and potential of generative AI for the public and for journalism but added that journalistic material must be used with the permission of the original rights holder to obtain commercial profit.
This case could set a major precedent in this dispute. If the court rules in favor of the newspaper, clearer limits could be established on the use of AI in content creation. OpenAI has argued that avoiding copyright violation is complex in this context.
Philosopher and linguist Noam Chomsky criticized ChatGPT, labeling its operation as “high-tech plagiarism” and “a way to avoid education.” That is, artificial intelligence lacks a criterion to select the information it outputs as results, so it really does not contribute anything to the advancement of education or science.
Therefore, when evaluating the effectiveness of artificial intelligence tools promoted as solutions for plagiarism detection, it is crucial to recognize the inherent limitations of those designed for language models such as ChatGPT. These restrictions, mainly associated with their ability to identify plagiarism, are influenced by biases introduced during their training phase.
This is the case of technological solutions such as OpenAI’s Text Classifier, GPTZero, AI Detector, Originality.ai, Corrector.app, Copyleaks, Writer, and others that seek to identify plagiarism in texts produced by artificial intelligence, especially with the use of ChatGPT.
An illustrative case of this is shown by a result from GPTZero, where it is claimed that 96% of the United States Declaration of Independence was written by an artificial intelligence. In August 2023 a Paraguayan case was documented in which a lawyer filed a complaint against a judge for supposedly using ChatGPT to draft a judicial resolution. The way this was detected was by analyzing the rejection of a constitutional petition as a guarantee of due process. What caught high attention was its unconstitutionality and that led him to conduct queries with ChatGPT. These queries matched between 80% and 90% with the judge’s resolution, which led the lawyer and his team to analyze filing a formal complaint against the judge for poor performance before the Jury of Prosecution of Magistrates (JEM).
As can be seen, both plagiarism detection technologies and the curious methodology carried out by Paraguayan lawyers can regularly produce false positives, asserting that certain text created by a human was actually created by an AI.
An important example in the educational field is the case at Stanford University, where a professor employed artificial intelligence tools to detect plagiarism. Interestingly, plagiarism matches only appeared when students did not have English as their native or main language. The simplicity of these students’ English, less sophisticated compared to a native speaker, led to incorrect plagiarism identification. This tends to misclassify more than 50% of writing samples from non-native English speakers as AI-generated, while maintaining almost perfect accuracy in native speaker samples.
This finding highlights the importance of addressing biases in ChatGPT detectors, as ignoring these biases could result in the marginalization of non-native speakers in educational or evaluative contexts. This pioneering study advocates further research to correct these biases and improve detection methods, thus ensuring a fairer and safer digital environment for all users. And lastly, this finding not only demonstrates that this is an error of the AI tool in terms of false positives, but also an evident bias that affects vulnerable people, such as migrants. This further deepens inequalities and exclusions of certain population groups, as AI experts expose.
Conclusion
Answering the question of who owns the text generated by ChatGPT is not trivial; that is, it is quite complex to define who and how the text offered by this chatbot was produced and, therefore, how plagiarism applies in this context.
Furthermore, the use of generative AI, which includes language models like ChatGPT, is not exempt from ethical responsibility. Although these models can be valuable tools, it is essential that users adhere to ethical practices and avoid plagiarism by generating original content and providing appropriate copyright attribution when necessary.
However, despite the creation of detectors specific to language models like ChatGPT-3 and 4 with the aim of reducing risks linked to content produced by artificial intelligence, certainty regarding their accuracy, reliability, and effectiveness remains in doubt due to limited evaluation. This deficit of knowledge is cause for concern, as it could result in harmful consequences by incorrectly labeling innocent people’s work as plagiarism when they produce a text.
Nonetheless, it is fundamental to recognize the inherent limitations in identifying plagiarism and copyright violations through artificial intelligence. Despite efforts to develop effective detectors, challenges persist in terms of precision and reliability. The risk of incorrectly identifying content as infringing can have harmful consequences, especially in educational contexts. Therefore, continuous reflection on ethics and constant improvement of detection tools is needed to ensure a proper balance between protecting intellectual property and preserving fairness.
References
Regulations of international organizations
UNESCO, 2021. Recomendaciones sobre la etica en la Inteligencia artificial. https://unesdoc.unesco.org/ark:/48223/pf0000381137
OECD, 2022. OECD AI Principles overview : https://oecd.ai/en/ai-principles
Inteligencia Artificial para Europa. 2018: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=COM%3A2018%3A237%3AFIN
European Commission. (2020). White Paper. On Artificial Intelligence – A European approach to excellence and trust.
Convención de Berna, 1886: https://www.wipo.int/copyright/es/
Constitucional Nacional Paraguaya (1992).
Ley 1328/98 Derecho de Autor y derechos Conexos. https://www.bacn.gov.py/leyes-paraguayas/908/derecho-de-autor-y-derechos-conexos
Academic literature
Broussard, Meredith. 2018. Artificial Unintelligence. How computers Misunderstand the World. The MIT Press.
Cathy O’Neil. 2016. Weapons of Math Destruction.
Kate Crawford, 2022. Atlas de inteligencia artificial. Poder, política de inteligencia y costos planetarios.
Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, James Zou. GPT detectors are biased against non-native English writers, Patterns. 2023. https://www.sciencedirect.com/science/article/pii/S2666389923001307
News blogs
Aristiegui. 2023. Chat GPT ‘es plagio de alta tecnología’ y ‘una forma de evitar el aprendizaje’: Chomsky https://aristeguinoticias.com/2302/libros/chat-gpt-es-plagio-de-alta-tecnologia-y-una-forma-de-evitar-el-aprendizaje-chomsky/
Chequeado. 2024. Arte, inteligencia artificial, derechos de autor y copyright: ¿a quién le pertenecen las obras? https://chequeado.com/investigaciones/arte-inteligencia-artificial-derechos-de-autor-y-copyright-a-quien-le-pertenecen-las-obras/
Ferrante, Enzo. 2021. Inteligencia artificial y sesgos algorítmicos. https://nuso.org/articulo/inteligencia-artificial-y-sesgos-algoritmicos/
FasterCapital. El papel de la IA en la detección de plagio de contenidos. 2023. https://fastercapital.com/es/contenido/Papel-de-la-IA-en-la-deteccion-de-plagio-de-contenidos.html
Andrés Guadamuz. 2017. La inteligencia artificial y el derecho de autor https://www.wipo.int/wipo_magazine/es/2017/05/article_0003.html
Suarez, Alex. 2023. La IA hace más difícil detectar el plagio: cómo ChatGPT coló un artículo a una revista educativa. https://www.lavanguardia.com/tecnologia/20230411/8837929/ia-mas-dificil-detectar-plagio-chatgpt-cuela-articulo-revista-educativa.html
[Launch] Fairwork Paraguay Report 2025 
Use of electronic voting machines in the 2023 general elections in Paraguay