By: Ricardo Arturo Gómez Hernández and Daniel Ascencio Rodríguez
With the advent of artificial intelligence (“AI”), more substantial changes to the way we live in private and in society will occur, which itself implies that the legislation and institutions that govern us will need to fit the upcoming mutations. One of the areas in which changes will exist – and where changes, although still budding, have already happened – is in the production of works, for which in this article we will exclusively analyze copyrights in relation to the use of AI for the production of works.
Before addressing our central topic, it is essential to reference the Dark Forest Theory of the Internet. The main idea of this theory is that the Internet is plagued by bots and AI-generated content and, as a consequence, ‘real people’ are increasingly confined to enclosed spaces where there are more entry barriers that prevent access to non-human agents (Appleton, 2023). This concern will seep into other areas of human life, as it’s currently happening in the domain of production of works, particularly literary works, in which many years ago, a novel written with the aid of AI tools participated in a literary contest in Japan, effectively bypassing the first filter of the contest designed to detect artificial intelligence (Olewitz, 2016). This milestone demonstrates that it will become increasingly easier for AI to bypass filters that distinguish between human works from those created by AI, while raising many questions regarding copyright such as if said novel is even capable of being copyrighted and, if affirmative, then who should be granted said rights.
Naturally, the answer to the proposed question will depend on the analyzed jurisdictions, therefore, in this article we will focus on (i) displaying the in-force legislation in Mexico regarding the copyright of works produced with the aid of AI tools, and (ii) propose, generally speaking, a strengthening of the Mexican institutions in charge of the matter at hand.
I. In-force legislation in Mexico regarding copyright relative to works created with the aid of artificial intelligence
Currently, in Mexico, the Federal Law on Copyright (Ley Federal del Derecho de Autor or “LFDA”) establishes in its Article 12 that the quality of author (i.e., a creator of an artistic or literary work) is recognized exclusively to natural persons (and, interpreting it contrario sensu, not to legal entities). Since, as of today, Mexican law doesn’t assign legal capacity to Large Language Models (“LLMs”), nor to other types of artificial intelligence, these technologies can’t acquire the quality of author as per the cited Article. It may be even more important to note that even if legislation would recognize legal capacity to artificial intelligence, not even this would be sufficient to render them able to be considered authors as long as they are not strictly considered natural persons (unless the definition of natural persons changed to include AI). The reason behind this restriction is none other than to encourage creativity as an exclusive attribute of human beings as an interpretation of Article 28, paragraph 11, of the Political Constitution of the United Mexican States.
However, what happens when a natural person is assisted by LLMs or other types of AI to produce a work? To whom does the copyright belong? To the user of the AI tool, to its creator, or to the owners of the copyrights of the information used by the data base of the AI tool? What issues does this reveal and what is proposed to solve them? Before boarding the Mexican case, we offer a glimpse to some solutions or criteria offered by other jurisdictions.
As expressed by Spindler (October, 2024), the two main problems can be reduced to the following: a) how are copyrights assigned to works produced with the aid of AI tools?, and b) what copyright protection should be granted to AI? In this article we will ignore the second question since other questions are attached to it that Spindler identifies and that are difficult to resolve, such as the way in which we would assign responsibility to an AI software in case its legal capacity is recognized (particularly, in Mexico, in the status of a natural person) or the way in which the AI would acquire or collect funds to enforce its copyrights (assuming it’s even possible in the future), in addition to the fact that currently such possibility does not exist within the Mexican legal framework as per Article 12 of the LFDA mentioned above.
The key to answering the first question, at least in other jurisdictions, lies in determining the degree of intervention in the work by artificial intelligence and to distinguish it from human-craft. For example, in United States of America, legislation and jurisprudence require human intervention to assign copyrights so that the parts in which AI intervened are not subject to being protected by copyright law (Coulter et. al., 2024). Something similar occurs in the European Union, where, derived from the Infopaq International A/S v Danske Dagblades Forening, the Justice Court of the European Union determined that “copyrights can only subsist if there is originality flowing from the ‘author’s own intellectual creation’”, but the interpretation of “author’s own intellectual creation” has varied per member of the European Union (Coulter et. a., 2024). United Kingdom – in contraposition to the aforementioned criterion –, establishes in its legislation (Copyright Designs and Patents Act 1988) that the author of a work is whoever makes the “arrangements necessary for the creation of the work”, which was ratified via an open consultation by the UK Intellectual Property Office, specifically on its application in relation to generative AI (Coulter et. al., 2024). In China, the Beijing Internet Court determined that an image generated by Stable Diffusion was an “intellectual achievement” and had “originality”, for which it was determined that it was possible to assign copyrights to the Stable Diffusion user relative to the generated image, since the instructions, parameters and inputs reflected his particular expression. Coulter et. al. (January, 2024) also warned that aside from considering the applicable law to the agreement that regulates the relationship between the user that inputs instructions to the software and its creator, it’s important to analyze the terms and conditions of the AI software or platform, the jurisdiction where the interested party wishes to protect its work with copyrights, among other factors, to determine applicability and assignment of the related copyrights.
In Mexico, although it’s clear that an AI is not susceptible of acquiring the quality of author and by not having recognized the status of natural person (and in general by not having any legal capacity whatsoever) as per in force legislation, the LFDA doesn’t resolve on whether if the user that inputs instructions or parameters in an AI software can be considered as the creator of a work (assuming that the user is effectively a human and not another AI, which would bring about another series of legal issues that we will ignore here), nor does it resolve to what extent would said user have copyrights over the work produced. Thus, the LFDA omits the parameters of the required degree of user intervention on the AI software to assign copyrights relative to the work produced. It’s convenient to remember that the LFDA dates back to the year 1996, so this legislative omission is not surprising.
Furthermore, the Public Registry of Copyright (the “Author’s Registry”) is governed by a principle of good faith registrations by which the facts and acts are presumed to be true, which is reflected on Article 168 of the LFDA and Article 59 of its Regulations. This brings about certain issues that we will proceed to expose.
II. II. Proposal to strengthen the Mexican institutions in charge of the registration of works subject to copyright
Alan Turing first proposed a test designed to have an observer determine – purely based upon writs – if those being interrogated were people or machines (Oppy & Dowe, 2021); if a machine managed to fool the observer into believing that it is human, then the test would have been effectively bypassed. Currently, due to their increased capacity, AI and particularly LLMs, have been bypassing the Turing test with increasingly more ease, which has prompted the invention and design of other tools to distinguish AI products from human ones. One of these inventions is the Reverse Turing Test or “RTT” (Alizadeh et. al., 2021) in which a machine is assigned the task of distinguishing between a human and another machine (inversely to the classical Turing Test). An example of the application of an RTT is the Captcha system that uses text and images to discern if the user of some web page is human or not, but even this system is becoming increasingly inadequate to differentiate between human and non-human users of social networks, having a relevant influx of bots (Alizadeh et. al., 2021).
AI tools may be implemented in such a way that they prove no major obstacle to the registration of works in Mexico – RTTs such as Captcha take no longer than 5 minutes in their analysis, and other more advanced RTTs even run on the background while they monitor user behavior in real time –, which may enable us to integrate a system that allows us to distinguish between AI and human works. Perhaps such a filter could be integrated to the Author’s Registry without undermining its good faith nature and without overloading the authority by imposing it with the obligation of analyzing if there is any work whose copyrights might be getting infringed.
By virtue of the above, its important to recognize the protectionist quality of the Mexican copyrights system – particularly regarding the rights of recognition of authors. We consider it fundamental to recognize that the answers that have been presented to this new set of problems can be found by regressing to the origins of the Mexican copyrights system; thus, by recognizing that one of the main practical problems of AI is not the challenge of designing rules for the assignment of copyrights to a non-human entity but the easiness that such technology provides to human users to plagiarize or to steal artistic or intellectual works of other authors, we may get closer to a real solution.
In this sense, our central proposal is to reinforce the institutions in charge of organizing the structure and guarantees of the Mexican copyrights system, therefore giving them tools to be better equipped against the waves of potential issues that loom and derive from new AI technologies. Thus, our specific proposals are two: a) include a filter with a viable TTR as requirement to register a work in the Author’s Registry, and use that same filter to implement it and have it running in the background, which would effectively avoid the issue of overloading the authorities with an herculean task as well as respect the good faith character of said Registry; and b) to propose an external consultation service to retrieve information pertaining to protected and registered works at the INDAUTOR to facilitate the monitoring of the interests of the copyright owners or to benefit the people seeking out to register a work. These proposals seek to diminish the number of cases that are brought to Court, which, as the production of works increases due to the apparition of LLMs and other AI tools, will likely increase.
In conclusion, this article intends to highlight the need to implement measures aimed at recognizing that AI tools are here to stay, and that their existence implies certain challenges relative to the assignment and protection of copyrights. At the same time, it’s important to open a dialogue with experts to debate the proposals made, so that eventually we may obtain their cleaner, more efficient version to better anticipate the upcoming issues.
Notes:
- McCarthy (2004) define “inteligencia artificial” como “la ciencia e ingeniería para crear máquinas inteligentes”, y define a su vez “inteligencia” como “el componente computacional de la habilidad para realizar objetivos en el mundo” (la traducción es nuestra).
We chose McCarthy’s definition over other alternatives because we believe it sheds a better light on the need to attend current copyright issues in Mexico derived from AI tools. Nevertheless, and to our knowledge, there is no single definition of artificial intelligence that is unanimously accepted by all scholars. Please see Abbass’ (2021) article on the subject to find an alternative definition of artificial intelligence for editorial purposes, but there are many more.
B. Appleton (2023) provides WhatsApp, Slack and Discord as examples of a closed space.
References:
Abbass, H. (Abril, 2021). Editorial: What is Artificial Intelligence? IEEE Transactions on Artificial Intelligence, Vol. 2, No. 2.
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9523786
Alizadeh, F., Mniestri, A., & Stevens, G. (2022). The Reverse Turing Test: Being Human (is) enough in the Age of AI. Proceedings http://ceur-ws.org ISSN, 1613(2022), 0073.
Appleton, M. (2023). The Expanding Dark Forest and Generative AI. Maggie Appleton.
https://maggieappleton.com/ai-dark-forest
Coulter, C., Tracy, R., Bartholomäus, R., Sangaré-Vayssac, L., & Yu, Z. (Enero 29, 2024). Copyright ownership of Generative AI Outputs Varies Around the World. Cooley LLP.
Keane, I. (Enero 18, 2024). Prestigious literary prize awarded to novel written with help from AI. New York Post.
McCarthy, J. (Noviembre 24, 2004). What is Artificial Intelligence? Computer Science Department, Stanford University.
https://cse.unl.edu/~choueiry/S09-476-876/Documents/whatisai.pdf
Olewitz, C. (Marzo 23, 2016). A Japanese A.I. program just wrote a short novel, and it almost won a literary prize. Digital Trends.
Oppy, G. & Dowe, D. (Octubre 4, 2021). The Turing Test. The Stanford Encyclopedia of Philosophy (Winter 2021 Edition), Zalta, E. (ed.).
https://plato.stanford.edu/entries/turing-test
Spindler, G. (Octubre 15, 2019). Copyright Law and Artificial Intelligence. Max Planck Institute for Innovation and Competition, Munich 2019.
