Who steals the scientific articles?

One of the world's largest natural science publishers is suing creators of artificial intelligence. While the giants fight among themselves, the authors of the articles themselves are left out of the picture.

By Tal Sokolov, courtesy of the Davidson Institute for Science Education, the educational arm of the Weizmann Institute of Science.

Elsevier, one of the largest and most influential corporations in the field of publishing scientific journals, Joining a class action lawsuit against Meta, the parent company of social networks Facebook and Instagram, which in recent years has also been involved in the development of artificial intelligence. The plaintiffs claim that Meta is violating copyright and using books, textbooks and scientific articles without permission to train its large language models, known as Llama models. Although there are other similar lawsuits underway at the same time, such as The lawsuit filed by the New York Times against OpenAI, the operator of GPT chat, but this is the first time that such a significant body in the field of scientific publishing has taken part in this type of lawsuit.

The desire of scientific and educational publishers to hold copyright over the content they publish is not a new phenomenon, and it existed long before the emergence of language models. Academic and educational content is usually behind high paywalls that prevent the general public, and sometimes even researchers from small and underfunded institutions, from consuming content that is important for research and science. There are several repositories on the Internet that bypass them, through pirate sites dedicated to this purpose or databases that researchers and other people have voluntarily shared. In this arena, the courts usually rule in favor of the publishers, but it is not certain that they give due weight to the public interest in doing so.

In an era where the distinction between quality, reliable information and sloppy, even false, content is becoming increasingly blurred, access to reliable sources of information and research is more important than ever. It may be that the balance between the financial interests of publishers and the public's need for reliable information needs to shift. The vision of open science It is a movement that strives to open up scientific knowledge to everyone, and to this end demands easier access to articles and even the complete abolition of paywalls. As part of the movement, large research organizations such as the CERN particle research center in Geneva, which operates the world's largest particle accelerator, conduct many important studies, and this gives them leverage over publishers. Some use their power to promote funding models that will develop articles that can be read for free. On a broader institutional level, in the United States it has been decided that the findings of all government-funded research will be available for free.

Now, with the rapid development of the artificial intelligence market, and within it the Meta Corporation, we must understand how the new player changes the balance of power and the status of copyright on scientific content. It is possible that in the specific context of scientific knowledge and content, the interests of artificial intelligence developers and the public converge.

Joins class action lawsuit against Meta. Elsevier Publishing | II.studio, Shutterstock

Access to databases

Large language models rely on their training process to Huge databases, to develop the ability to generate rich text in the context appropriate to the user's request. The information is collected in a variety of ways. Some content providers are already choosing to sign direct and profitable agreements with AI companies that allow them to use their content. Academic book publisher Taylor & Francis, for example, Signed an agreement One with the software giant Microsoft. within the framework of the agreement The publisher will receive more than ten million dollars in exchange for providing access to its content for training artificial intelligence systems.

Alongside the agreed collection of content from rights holders, AI companies also routinely collect Current information from the InternetHere, a host of challenging conflicts are already arising between the content creators and the companies that collect and use the open information exposed to the public.

The copyright issue is exacerbated when it comes to collecting pirated content available online. In Elsevier's lawsuit, the publishing company claimed that Meta Copy content it owns from shadow libraries Like Sci-Hub and LibGen. These libraries are huge repositories that provide access to academic content protected by paywalls and scan books that are not open to the public. There are legal battles going on around the libraries over their legitimacy, even apart from the question of artificial intelligence. 

Payment for academic articles and books can often reach several dozen dollars per article. Since regular reading of current articles and frequent updating on innovations is an essential component in the advancement of research and human knowledge, universities and research institutes usually pay publishers subscription fees to give their researchers access to them. Thus, the main victims of the existing policy are individuals and entities with limited financial resources. In this space, pirate libraries have emerged.

Elsevier and other publishers have so far filed a number of lawsuits against pirate sites, and courts in the United States They ruled that Sci-Hub must pay them damages.In India, where many universities have tight budgets, courts tend to be lenient on copyright infringement. As long as it is used for scientific or educational purposesHowever, even there the judges rejected the website operators' claims about the importance of available scientific content, so access to it from India is currently blocked. 

Payment for academic articles and books can often reach several dozen dollars per article. Selection of scientific journals in the university library | Vmenkov, Wikipedia

Fair use, towards whom?

The copyright struggle also has another side – the scientists themselves. Many researchers and academics support opening up access to their own and others' articles. Some even take an active part in piracy, even though it contradicts the financial interests of the publishers who publish their articles. Sci-Hub, for example, builds its databases using the usernames and passwords that researchers with subscriptions secretly donate to it; under the heading “The Price of Knowledge”Researchers have long been conducting open protest activities in which they declare that they will not publish articles through Elsevier, will not cite them, and will not assist them with peer reviews, in opposition to the high prices charged by the publisher.

Scientific publishers charge high fees to researchers and scientists who request to publish articles in them. In many cases, researchers are required to pay for the publication from their research budgets, in amounts that often reach thousands of dollars. And if that were not enough, the payment goes only one way and the publisher does not reward researchers for the services they provide. In order for an article to be published in a scientific journal worthy of the name, it must pass the hurdle of peer review, that is, a critical examination by other scientists who are experts in their field of research. The reviewers, who are researchers themselves, We do not accept payment for the audit.So while the major publishers enjoy their status, which makes them desirable platforms for publication, researchers and the general public are harmed by the process.

Some scientists participate in piracy activities and provide user information and passwords to sites like Sci-Hub. Sci-Hub website on a computer screen | mundissima, Shutterstock

Giant War

We live in an era where research and theoretical articles and other academic study materials are valuable content for human society as a whole, including for the training processes of artificial intelligence models. They hold a lot of high-quality information, and their reliability usually exceeds that of other texts written by humans. The more high-quality and reliable the information on which the models are trained, the higher the quality of the content they will produce is expected to be.

This situation places academic content high on the priority list of creative AI developers. In addition, Researchers are increasingly using bots based on language models During the writing of articles and the search for relevant research information. In this situation, the researchers themselves also have an interest in the models being trained on the highest quality data. In practice, it seems that no one really asks the researchers themselves what they want to happen to their copyrights on the articles they wrote. Even in the agreement in which Taylor and Francis sold Microsoft legal rights to use their content, it was reported that the authors of the content – ​​the scientists themselves – were not involved in the deal and would not be compensated for it. 

From a legal perspective, the issue brings us back to the question of whether the use of scientific content to train large language models can be considered fair use, and therefore legal. Meta's representatives in the lawsuit argue that its use of the materials is considered fair, Because it is transformative, that is, uses the content in a way that significantly changes it from its original form.

Alongside this lawsuit, there are currently a host of other lawsuits dealing with copyright infringement during the training of artificial intelligence models or the products produced from them. Scientific and educational content may deserve different treatment than other content – ​​for example, works of art or commercial content. What is clear is that the original creators, i.e. the researchers, appear not to be taken into account in ownership considerations.

In the arena of struggle, a struggle is being waged between the profits of huge commercial expenditures, versus those of wealthy technology companies. In a particularly optimistic scenario, one might hope that the conflict will ultimately lead to more open and accessible science. For this to happen, the decision will have to take into account the common good, and as we know, it is never taken for granted. 

For additional articles and content: Davidson Institute for Science Education

More on the subject on the science website

• "The Artists' Revenge" Against Artificial Intelligence
• How the coronavirus is changing science
• A new international venture in the field of informatics will spread Israeli research in the world
• The first comic created by artificial intelligence – and it’s a joke
• Artificial intelligence, painting - and the bright future of art