When examining the legal status of outputs produced by artificial intelligence in Türkiye, it is necessary to address the issue from two different perspectives: the license agreement of the artificial intelligence producing the content and the Law on Intellectual and Artistic Works No. 5846 (“FSEK”).
The use of artificial intelligence (“AI”) is a novel issue from a legal standpoint, therefore, there are no established rules or practices in our law accordingly. FSEK is designed to protect works created by humans; how outputs produced by AI will be evaluated and protected within the existing legal framework remains uncertain.
In this context, the licenses of AI models will play a significant role in determining the ownership of outputs produced by AI. The licenses of AI models are agreements that determine under which conditions the companies publishing the models offer them to users. These licenses specify how the model itself can be used, modified, and distributed, and most importantly, who owns the outputs produced using the model and how they can be used. Therefore, to assess the ownership of an output produced by AI, it will not be sufficient to examine only the intellectual property law; it will also be necessary to examine the license conditions of the AI model used to produce that output.
Different AI models are offered to users under open-source licenses or with their own unique license agreements. While open-source licenses generally allow the model to be freely used, modified, and distributed, other licenses offer different conditions unique to them and should be examined specifically for each case. Examining licenses offered, especially for commercial use, is extremely important.
The purpose of this article is to examine the status of AI-generated content in Türkiye in terms of intellectual property law. In particular, AI model licenses and how fundamental concepts such as authorship, originality, and types of works under FSEK should be interpreted for AI-sourced content is an issue that needs to be evaluated. Additionally, important cases in the international arena will also be mentioned.
- Definition of Work in Terms of Intellectual Property Law in Türkiye
Intellectual property law in Türkiye aims to encourage creativity and innovation by protecting the rights of authors. The main regulation in this field is the Law on Intellectual and Artistic Works No. 5846. FSEK protects works in four main categories: literary and scientific works, musical works, fine art works, and cinematographic works.
According to FSEK, for a work to be protected, it must “bear the characteristics of its owner,” that is, it must be original. In addition, the work must be the result of the creative activity of the “author,” that is, the person or persons who created the work. The author has financial and moral rights over the work.
- Status of AI-Generated Outputs in Türkiye in Terms of Intellectual Property Law
How AI-generated outputs will be evaluated or protected within the scope of intellectual property law in Türkiye is a matter that has not yet been fully clarified. Even who owns the elements to be protected is a matter of debate, and no legal solution is currently available. Since the existing legal regulations focus on human creativity and originality, there are uncertainties and debates about how AI outputs will fit into this framework.
Before addressing the issues regarding licenses and international examples, it would be beneficial to determine the current legal situation in Türkiye.
a. Authorship Problem
According to FSEK, the author is the person who created the work. Although the term ‘real person’ was removed from the definition of work within the scope of FSEK in 2004, ‘person’ is still included in the definition. This indicates that the producer must be a person for an output to be considered a work. This will be the first problem that needs to be solved for AI-produced outputs. If AI productions cannot be considered a work, they cannot be subject to intellectual property law, and the validity of the license agreements regarding them will be questionable. In this case, who will be the owner of a work produced by AI?
Different views are put forward on this issue:
- The Person Who Developed the AI: According to this view, the legal or natural person who developed the AI should also be the owner of the outputs produced by the AI. The basis of this idea is that AI model is the AI developer’s creation and the main contribution to the emergence of the outputs is this. However, there are also technical problems here: such as the data sets used when developing the AI does not primarily belong to the developer. If this view is accepted, license agreements for AI users will play an important role in determining the ownership of content.
- The Person Who Uses the AI: According to this view, the person who produces the output by using the AI should be the owner of the output. The reason for this is that AI is only used as a tool and the user’s guidance (prompt) and choices (sampling) are decisive in the final form of the output. The problem here is the removal of artificial intelligence, which has very high production costs, from its producer.
- The Work Being Ownerless (Public Property): According to a different view, outputs produced by AI cannot belong to any person and should be accepted directly as public property. The justification for this is that artificial intelligence is generally produced from complex data sets that do not belong to a single person and neither the producer nor the user of the model has a sufficient degree of contribution through the output of the model, only guidance.
- Artificial Intelligence Itself: As a final view, there are those who argue that artificial intelligence itself should be the owner of the outputs, but today this is clearly a view that does not need to be discussed in the scope of this article.
There is no legislation or precedent on this issue in Türkiye. However, it can be said that this is the first problem that needs to be solved, as it may even prevent AI outputs from being the subject of intellectual property law. It can be said that the closest view to being accepted is that AI outputs basically belong to the person who developed the AI and the continuing ownership is determined by license agreements. The AI models that continue to be released and the general practice also support this, and especially companies using AI generally act according to the licenses of the models as a preventive practice for possible legal problems in the future. It is likely that the application will progress ahead of the legislation and influence future legal regulations in this direction.
b. Originality Problem
According to FSEK, a work must “bear the characteristics of its owner,” that is, it must be original. We can say that originality means that the work is a reflection of creativity. How the originality criterion will be applied for AI-produced outputs is debatable.
AI models generally generate new outputs by being trained on large data sets. In this case, the question arises whether the outputs produced by AI are completely original or whether they are a derivative or copy of existing works. If the AI produces a new output by significantly changing the existing works in the data set and adding its own originality, it can be accepted that it is original. However, the originality criterion will not be met if existing works are only copied or very small changes are made.
When evaluating originality, the role and contribution of AI in the creation of the content should be taken into account. If AI is considered only a tool that fulfills the user’s instructions, it should be accepted that the originality belongs to the user. If it is accepted that the AI’s own characteristics and creativity are at the forefront, it should be accepted that the originality belongs to the person who developed the AI or the AI itself. As long as there is an original work, considering that human creativity is also based on people’s past experiences, this problem should be resolved within the scope of the problem of ownership of the work, rather than the content of the data sets, and originality should be attributed to the ownership of the work.
The main issue will be the existence of originality. In other words, it should be examined whether the output is just a copy or has been changed very little from its similar works in the data set.
- Licenses of AI Models and the Effects of Licenses on Intellectual Property Rights
The development of AI models is an intensive and costly process that requires large amounts of data and computation. In this context, developers of AI models resort to various licensing methods to control the use of their models while offering their models to users. As we have stated, the basic tendency in practice is currently that the outputs of AI models basically belong to the developer of the AI and the ownership is determined by licensing. We will make the reviews under this section with the assumption that the ownership of the work is resolved in this direction (which is also the case in the current situation).
a. License Types and Their Effects on AI-Generated Content
AI models are generally offered to users under three main license types:
Open Source Licenses:
Open source licenses are licenses that are widely used in the software world and allow the software to be freely used, modified, and distributed. These licenses are also used for AI models. An open source licensed AI model can generally be freely used, modified, and distributed by anyone. However, open source licenses also have different types and conditions within themselves.
- Copyleft Licenses: These types of licenses require that the contents produced using AI are also distributed under the same license. That is, if an AI model is distributed under a copyleft license, an image, text, or music piece produced using this model must also be shared as open source under the same copyleft license. This situation may restrict the commercial use of AI-generated content.
- Licenses That Allow Commercial Use: These types of licenses allow the AI model and the outputs produced to be used for commercial purposes. These types of licenses will allow the outputs produced by AI to be used widely.
- Licenses That Do Not Allow Commercial Use: Unlike licenses that allow commercial use, these types of licenses prevent the outputs produced by AI or the AI models themselves from being used commercially. These types of licenses are generally used in AI models for research purposes.
Closed Source Licenses:
In this regard, it will not be possible to make general distinctions as in open source licenses. Each developer uses its own unique license agreement. If the AI model is not published with a known open source license, the license must be examined in detail and a legal evaluation must be made specifically for the model.
For example; The license of LG’s (“LG Management Development Institute Co., Ltd.”) EXAONE model clearly states that all content produced with the EXAONE model will belong to LG under all circumstances. Meta’s (“Meta Inc.”) Llama models, on the other hand, protect them with different original licenses, and even if the outputs belong to the user depending on the version of the Llama model, there may be different conditions for commercial use.
In short, each license must be examined separately and a decision must be made with a legal investigation according to the current situation.
b. Effects of API Usage on Licensing and Intellectual Property Rights
Access to and use of AI models can generally be divided into two main categories:
- Access via API (Application Programming Interface): In this method, instead of running the AI model themselves, users access and use the model remotely through an API offered by the model’s developer or provider.
- Running the Model Directly: In this method, users run the AI model directly on their own systems. This method is generally possible with open source licensed models or in cases where a downloadable version of the model is offered.
The use of API has significant effects on the intellectual property rights and licenses of outputs produced by AI. If you are accessing an AI model through a third-party service (e.g., an API provider other than the developer), the ownership of the output will also be subject to the terms of service of that provider. In this case, the terms of service may vary from service provider to service provider, so the terms of service will need to be examined together with the licenses and applicable legislation. Some providers retain ownership of the output, while others may grant ownership to the user.
- Judicial Decisions and Examples in Practice
There is no precedent regarding the intellectual property rights of outputs produced by AI in Türkiye. However, considering the universality of intellectual property law, existing court decisions abroad can be guiding on AI outputs and situations we may encounter in the future.
- DABUS Case
DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) is an artificial intelligence system developed by Stephen Thaler. DABUS has made two independent inventions without human intervention and has been shown as the inventor of these inventions in the patent application.
However, patent applications made in the USA and the United Kingdom were rejected on the grounds that “The inventor can only be a human”. The Republic of South Africa, on the other hand, became the first country to officially accept an artificial intelligence system as an inventor.
- Naruto Selfie Case
In 2011, a macaque monkey (“Naruto”) in Indonesia took his own selfie with a photographer’s camera. This photo caused controversy. The animal rights organization PETA argued that the monkey should be recognized as an individual and therefore the copyright should belong to Naruto. However, American courts ruled that copyright can only be claimed by humans.
- Possible Solutions and Outlook for the Future
a. Updating Current Legal Regulations
The most important step is to update the existing legal regulations to cover the outputs of AI models as well. With the changes to be made in FSEK, issues such as the definition of works and ownership of works should be clarified.
Another option is to create a special protection system for AI models’ outputs instead of including them in the existing types of works. This system may include special rules on issues such as ownership of works, protection period and scope, taking into account the unique qualities of AI models’ outputs.
b. Contractual Solutions
As long as the uncertainties in legal regulations continue, contractual solutions, especially licenses, are gaining importance. Judicial decisions can shabe Agreements and licenses, and new rules can be derived in the light of judicial decisions.
c. Co-Evolution of Technology and Law
While artificial intelligence technology continues to develop rapidly, it is imperative that the law keeps up with these developments. In the future, the role of artificial intelligence in the creative process will increase even more. In this case, intellectual property law must also adapt to this new reality and evolve to protect works that are the product of human-artificial intelligence cooperation.