A. Introduction
ChatGPT (''GPT-3'') is a dialogue model introduced by OpenAI. It can follow complicated instructions, and research and provides extensive responses considering all previously asked questions1. To elaborate its abilities, we can emphasize that GPT-3 has the potential to write an academic paper itself2. Apart from its writing skills, it can advise in the field of several areas including finance, trading and even law3. Some authors go beyond and claim that GPT-3 can be significantly successful at the bar exam4. However, these abilities bring significant legal problems too.
First, can content produced by GPT-3 be seen as work and if so, who is the owner of it? Second, can this content be subject to IP protection? Third, does GPT-3 infringe IP law when collecting a massive amount of data to generate text? In our work, on one hand, we will analyze how GPT-3 can pose a threat to Intellectual Property law (''IP law''), on the other hand, we will examine, to what extent, content produced by GPT-3 be considered work within the ambit of Turkish Intellectual Property Code numbered 5846 (''Turkish IP Code''). Lastly, we will address potential data protection issues, particularly regarding the legitimacy (?) of the collection of massive amounts of data.
B. Does Content Produced by GPT-3 Can Be Protected Within the Meaning of Turkish IP Code?
Under Article 1/B of the Turkish IP Code, the work is all kinds of intellectual and artistic products that have the characteristics of their owner and are considered as works of science and literature, music, visual arts or cinema.
1. Who is the Owner?
First, we should assess if Artificial Intelligence (''AI'') can be an owner. According to Article 1/B of the Turkish IP Code, owner shall have personality5. However, similar to most jurisdictions, AI has no personality under Turkish law yet6. This simply means that AI cannot be an owner. Then who can be an owner? At this point, one can come up with three ideas:
- The person who created the software may be the owner or
- The person(s) who trained the AI may be the owner or
- The User(s) who share their creative inputs into the software may be the owner7.
Even though, all the above-mentioned actors are crucial to create work, from our perspective, the person(s) who trained the AI is likely to be the owner since they are uploading the informative capacity to the AI to generate unique products.
2. Does Work Have the Characteristics of its Owner?
This element of work implies that work shall be creative to some extent and it shall include intellectual effort8. In a very recent decision, the Turkish Court of Cassation held that if work can be formed only when owner makes it, then it has the characteristics of its owner. Court also noted that the original parts of the work are subject to IP law protection9.
GPT-3 is unlikely to create masterpieces like Shakespeare's Hamlet but still, content produced by GPT-3 can be considered original because it evaluates all the transferred sources to present better content and this research mechanism itself is original likewise its results. Therefore, we believe that works produced by GPT-3 can be protected under the Turkish IP Code if it is sufficiently creative.
C. Does GPT-3 Infringe Copyright When Collecting Human Works to Generate Text?
It is important to note that numerous human works are included in the training set of GPT-3, often, without the author's knowledge or permission10. Some of the works may be a public source but others are protected by copyright11. Since works are used without their authors' permission, it seems that GPT-3 poses a serious threat to IP law.
1. Scope of Copyright Protection
Between Articles 21 and 25 of the Turkish IP Code, rights of owners are explained deeply. These rights mentioned below are valid for copyright owners as well:
- Derivative works
- Make copies of the work
- Distribute copies of the work
- Right to benefit
These rights are exclusively granted to the copyright owners; thus, GPT-3 cannot create derivative works, copy them or distribute the work without the allowance of the author.
2. Fair Use Doctrine
Fair use doctrine is an exception that legalizes to use of copyrighted materials without the owner's consent for criticism, comment, news reporting, teaching, scholarship, or research12. Turkish IP Code also considers personal use13and quoting from written and musical works that have been made public within the meaning of fair use14. The purpose of the fair use doctrine is to strike a balance between the interest of the copyright owner and society.
To determine whether fair use doctrine is applicable, four criteria have been taken into account by courts15:
a. The purpose and character of the use, including whether it is commercial, transformative, and non-expressive;
b. The nature of the copyrighted work;
c. The amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
d. The effect of the use upon the potential market for or value of the copyrighted work.
Activities of GPT-3 can be seen in research, but merely this is not enough to apply the fair use doctrine because the above-mentioned criteria seem to not have been fulfilled. First, GPT-3 is used for commercial purposes. This increases the possibility of violation. Second, despite the fact that OpenAI claims otherwise16, GPT-3 is not sufficiently transformative yet. If work produced by GPT-3 includes alterations that bring new expression or meaning, then this will increase the possibility to benefit from fair use defense. Besides, it is important to analyze the impact of used work on new work. If new work was inspired fundamentally by copyrighted work, then it is likely to be seen as a violation of copyright. So, in brief, "the more transformative the new work, the less possibility of violation of copyright17.''
D. Potential Data Protection Law Concerns Regarding Activities of GPT-3
GPT-3 is designed to process massive amounts of data including personal data within the meaning of the Turkish Data Protection Code numbered 6698 (''Data Protection Code''). One may accurately ask these questions: What is the legal ground(s) to process personal data? Who is a data controller? Does GPT-3 comply with general principles of data protection law such as data minimization, transparency and purpose limitation?
1. Legal Grounds of Collection of Personal Data by GPT-3
For the Users who get service from GPT-3, consent may be a primary legal ground for the processing. The more controversial point is personal data that is obtained without the consent of data subjects by GPT-3 during the training process. In this case, it can be argued that these data were publicly available and therefore, it is legal under Article 5 (2) (d) of the Data Protection Code and General Data Protection Regulation (''GDPR'')18. However, it is likely to be seen as a violation of the purpose limitation principle because data subjects probably did not make publicly available their personal data for AI activities of GPT-3. Hence, without consent of data subjects, activities of GPT-3 may be seen as a violation of data protection law.
2. Who Is A Controller?
Data controller is the real or legal person who determines the purposes and means of the processing of personal data. Here, OpenAI is the data controller because it determines the aim of processing as well as in which way data will be processed.
3. General Principles of Data Protection Law and GPT-3
GPT-3 seems to infringe on at least three fundamental principles of data protection law. First, activities of GPT-3 directly contradict with the principle of purpose limitation. The principle states that personal data shall be collected for ''specified, explicit and legitimate purposes and not further processed in a manner that is incompatible with those purposes". On the other hand, GPT-3 ignores the real purpose of why personal data is made available and uses any personal data for its commercial purposes.
Second, activities of GPT-3 are likely to against the principle of data minimization. This principle entails to process data that is adequate, relevant and limited to what is necessary for relation to the purposes. However, GPT-3 processes plenty of data without making difference between what is necessary and what is not.
Third, activities of GPT-3 may violate the principle of transparency. The reason behind this is the lack of information on the process of decision-making. We have limited knowledge of to structure and logic of the decision-making system. GPT-3 needs more transparency like any other AI solution.
Apart from this, the principle of lawfulness, data accuracy and storage limitation may be argued. The fact that current nature of GPT-3 is not data protection-friendly AI and it may be subject to data infringement investigation by Data Protection Authorities in the future.
Footnotes
1. Fei Sun, ''ChatGPT, the Start of a New Era: A Bright and Gloomy Future'', 2022, p.2
2. GPT Generative Petrained Transformer, Almira Osmanovic Thunström, Steinn Steingrimsson, ''Can GPT-3 Write an Academic Paper on Itself, with Minimal Human Input?'', 2022, p.6
3. Michael Dowling, Brian Lucey, ''ChatGPT for (Finance) Research: The Banarama Conjecture'', 2023, p.2
4. Michael Bommarito, Daniel Martin Katz, ''GPT Takes the Bar Exam'', 2022, p.9
5. Even though Turkish IP Code seems to include both real and legal person, some authors claim that only real persons can be the owner. For the discussion see: Mustafa Zorluel, ''Artificial Intelligence and the Copyright'', 2019, p.318
6. For US See: Simon Chesterman, ''Artificial Intelligence and the Limits of Legal Personality'', 2020, p.19, For France See: Clemence Lapotre, Celine Bey, ''Artificial Intelligence in France'', 2022, p.3
7. Arthur Roberts, Alain Godements, ''Who Owns the Copyright in AI-generated Art?'', 2022, p.2
8. Osman Gazi Güçlütürk, Rifat Cankat, ''Yapay Zekâ ile Olusturulan Ürünlerin Eser Niteligi ve Eser Sahipligi Meselesi'', 2021, p. 208
9. Turkish Court of Cassation 11th Circuit, E.2021/8676, K.2022/3718, T.11.5.2022
10. Carol Hayes, ''Did a Robot Write This Title? Creativity, Ownership, Justice and Copyright Law'', 2022, p.2
11. Ibid, 2
12. Diana Bikbaeva, ''AI Trained on Copyrighted Works: When Is It Fair Use?'', 2023, p.3
13. See article 38 of the Turkish IP Code
14. See article 35 of the Turkish IP Code
15. Benjamin L. W. Sobel, ''Artificial Intelligence's Fair Use Crisis'', 2020, p. 5
16. Comment of OpenAI LP, Comment Regarding Request for Comments on Intellectual Property Protection for Artificial Intelligence Innovation Docket No. PTO-C-2019-0038, p.5
17. Sobel, ibid, p.5
18. See Article 9 (2) (e) and Article 86 of the GDPR
Originally published 26 January 2023
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
