Copyright and artificial intelligence: challenges and prospects

Artificial intelligence. AI. These two words sounded like science fiction yesterday, but today they are already invading all spheres of our lives – from medicine to art. And, of course, copyright could not stay away from this technological breakthrough. Every day artificial intelligence becomes more and more creative, generating texts, images, music, code – works that used to be the exclusive prerogative of humans. But who is the author of these new creations? Do they belong to anyone at all? And how should copyright respond to this new reality?

In other words, the rapid development of artificial intelligence has raised a number of difficult questions to which traditional copyright does not seem to have ready-made answers. Can AI be considered an author? How to protect works created with the help of AI? Is it legal to use copyrighted works to train AI? Who is responsible forcopyright infringement committed by artificial intelligence?

I. Works created by AI: subject to copyright or not?

Can works created by artificial intelligence be subject to copyright? This is a key question that has arisen with the rapid development of AI. Traditional copyright is based on the concept of an “author” as a human creator. But is this concept applicable to AI? In this section, we will look at the traditional understanding of author, the arguments for and against recognising AI as author, and international experience in this regard. We start with the traditional concept of an author.

Subsection 1.1: The concept of “author” in the traditional sense

So, let’s start with the basics, so to speak, with the foundation of copyright – let’s find out what is actually meant by the term “author” in its traditional, classical sense. After all, the entire system of intellectual property protection, at least as we know it, was built around the idea of the human creator. It is man, with his unique personality, his creativity and intellect, who has long been considered the source of creativity, the “author” whose rights need to be protected.

Let’s turn to the primary sources, i.e. the law. The Law of Ukraine “On Copyright and Related Rights“, for example, in Article 1 clearly definesthe “author” as an individual, whose work created the work. Please note: a natural person, i.e. a flesh-and-blood human being, not an impersonal machine or program. A similar approach can be found in international treaties, such as the Berne Convention for the Protection of Literary and Artistic Works. Although it may not contain a direct definition of “author“, the whole spirit and logic of the convention is based on the understanding of authorship as a creative activity of a person.

In other words, traditional copyright law considers the “author” as a being endowed with consciousness, intelligence, creative imagination, and the ability to express himself. This is someone who puts a piece of himself, his soul, his vision of the world into a work. This is a person whose creative work is the source of the work’s originality and uniqueness. And a work, accordingly, is seen as the result of this personal creative activity.

Of course, this understanding of “author” and “work” was developed historically, in an era when artificial intelligence could only be read in science fiction novels. Back then, the issue of machine authorship was simply not on the agenda. Legislation was focused on protecting the rights of human creators, and this was quite logical and justified in those historical conditions.

But as we know, time does not stand still. Technology is evolving rapidly, and artificial intelligence is already knocking on the door of copyright law, challenging the usual notions of “author” and “work“. And will the traditional understanding of the “author” be enough to meet the new challenges of the AI era ? We will try to answer this question in the following subsections.

Subsection 1.2: Arguments for and against recognising AI as an author

So, traditional copyright law is based on the understanding of the “author of” as a human being. But does this mean that artificial intelligence can never, under any circumstances, be recognised as an author? The question is not as clear-cut as it might seem at first glance. There are arguments both for and against recognising artificial intelligence as the author of the works it generates. Let us consider the main ones.

Arguments in favour of recognising AI as an author:

• AI shows creativity: Modern artificial intelligences are capable of generating works that show signs of originality, novelty and creativity. They can create texts of various genres, paint pictures in different styles, and compose music that sounds quite professional. Isn’t this creativity, albeit machine-based?
• Stimulating innovation: Recognising AI as an author could stimulate further development of artificial intelligence technologies in the creative field. Developers of AI would be more motivated to invest in creating even more advanced creative algorithms if the results of their work were protected by copyright.
• Legal certainty: Recognising AI as an author would create legal certainty regarding the legal status of works created by artificial intelligence. This would simplify the use of such works, avoiding legal uncertainty and possible disputes.
• Economic benefits: Works created by AI may have commercial value. Recognising them as objects of copyright would provide legal protection for these investments and stimulate the development of a new industry of creative AI works.

Arguments against recognising AI as an author:

• Lack of consciousness and intention: Artificial intelligence, to date, has no consciousness or intention in the human sense. It operates according to algorithms created by humans and is not aware of the creative process in the same way as a human. Copyright traditionally protects the expression of the author’s personality and intellectual contribution, not just the mechanical reproduction of algorithms.
• Blurring of the concept of “author”: Recognising AI as an author may blur the traditional concept of “authorship“, devalue human creative work and undermine the foundations of copyright, which was created to protect human creativity.
• Liability issues: If AI is recognised as an author, the question arises as to its liability for copyright infringement or other illegal actions. Can a machine be legally liable? Who will be responsible for the actions of AI? The developer? The user? The AI itself?
• Ethical issues: The recognition of AI as an author may raise ethical issues related to the role of humans in the creative process, the value of human creativity compared to machine-generated content.

As you can see, there are plenty of arguments on both sides. And there is no unambiguous answer to the question of whether AI can be recognised as an author. The problem is complex and multifaceted, requiring consideration of not only legal, but also philosophical, ethical and technological aspects. And the search for an optimal legal solution in this area is still ongoing.

Section 1.3: International experience

The issue of copyright for works created by artificial intelligence is relatively new, and no single approach to its solution has been developed in the world yet. Different countries are experimenting with different approaches, trying to find a balance between protecting copyright and stimulating the development of AI technologies . Let us consider some examples of international experience.

• USA: In the United States , the Copyright Office holds the position that copyright can only protect works created by humans. Works generated by AI without human intervention are not subject to copyright protection. There have been several precedent-setting cases when the Bureau refused to register copyright in works created by AI. For example, there is a well-known case of refusal to register an image generated by Steven Thaler’s Creativity Machine. However, if a person has made a sufficient creative contribution to the creation of a work using AI, such a work may be protected by copyright, and the object of protection will be the human contribution.
• European Union: The EU does not yet have a single legislative act that would regulate the copyright of works created by AI. However, the European Commission is actively studying this issue and developing recommendations. The general tendency is to believe that copyright should only protect works created by humans. However, different approaches to the definition of “human contribution” to the creation of a work using AI are being discussed.
• United Kingdom: In the UK, since 1988, there has been a rule stipulating that in the case of computer-generated works, the author is the person who performed the actions necessary to create the work. That is, not the computer itself (or AI), but the person who programmed, configured or used it to create the work. This approach allows to grantcopyright protection to works created with the help of AI without recognising the AI as the author.
• Australia: A similar approach to the UK.
• China: There is an interesting trend in China. There are precedents when courts have recognised copyright in works generated by AI. For example, in 2019, a court in Shenzhen recognised Tencent’s copyright in an article written by the DreamwriterAI system. The court reasoned that the article had an original structure and expression, even though it was created by AI. This shows that China may be moving towards a more flexible approach to copyright in the AI era .
• Japan: Japan is considering various regulatory options, including the possibility of creating a special regime for works created by AI that would differ from traditional copyright.

As we can see, the international experience is quite diverse. Some countries follow the traditional approach and do not recognise AI as an author, while others are looking for new approaches, trying to adapt copyright to the challenges of the AI era . And this legal evolution will undoubtedly continue, as artificial intelligence technologies are developing very rapidly, posing new questions to lawyers.

II. Using copyrighted works for AI training

The issue of authorship of AI-generatedworks is a complex one, but there is another side to it: using existing works to train AI. In order for AI to learn to create, it needs data, often protected by copyright. Is such use legal?

In other words: does “machine learning” infringe the copyright of? In this section, we will look at how this “learning” works, what exceptions exist in copyright for such cases, and what to do about it in Ukraine. Let’s start with the principle of “machine learning”.

Subsection 2.1: “Machine learning” and copyright

To understand why using works to train AI can be problematic from a copyright perspective , you need to understand howmachine learning actually works.

Machine learning is a sub-field of artificial intelligence that allows computer systems to “learn” from data without being explicitly programmed to perform a specific task. In other words, instead of giving AI clear instructions on how to act in each situation, it is provided with a large amount of data and it independently identifies patterns, learns to recognise patterns, make predictions and, ultimately, generate something new.

How does this relate to copyright? It is very simple. For AI to learn to write texts, for example, it needs to “read” a huge number of texts. To learn how to draw, it needs to “see” a lot of images. To learn to recognise musical styles, you need to “listen” to a huge number of musical compositions. And these texts, images, and music are usually subject to copyright, i.e. they belong to someone and are protected by law.

Themachine learningprocess typically involves the following steps:

1. Data collection: Large amounts of data are collected, which may include copyrighted works .
2. Data copying: This data is usually copied to servers where the AI is trained .
3. Data analysis: AI analyses the data, identifies patterns, studies the structure, style, and content of the works.
4. Model creation: Based on the data analysis, AI creates a model that allows it to generate new works similar to those it has learned from.

And here the question arises: is such copying and analysis of copyrightedworks an infringement of the authors’ rights? After all, as a general rule, copyright grants the author the exclusive right to use his or her work, including the right to copy, distribute, process, etc.

On the one hand, AI does not create exact copies of works, but uses them for learning, for creating something new. On the other hand, the learning process itself is impossible without copying and analysing original works. And this copying takes place without the permission of the authors, without paying them remuneration.

Therefore, the question of whethermachine learning is an infringement of copyright is quite complex and controversial. And the answer to this question depends on how one interprets the provisions of copyright and whether there are any exceptions or limitations that may apply to such cases. This is what we will discuss in the next subsection.

Subsection 2.2: Exceptions and limitations to copyright for AI learning

We have found that AI“machine learning” has the potential to conflict with copyright, as it involves copying and analysing protected works. But does this mean that any use of works for training AI is illegal? Not quite.

In copyright, there are so-called exceptions and limitations, which allow the use of works without the author’s consent and without payment of remuneration in certain cases clearly defined by law. These exceptions and limitations are intended to strike a balance between the interests of authors and the public interest, such as education, science, culture, freedom of speech, etc.

Can these exceptions and limitations be applied to the use of works for AItraining ? This question is a subject of active debate in the world, and different countries approach it in different ways.

Let’s look at some examples:

• “Fair use” in the United States:In the United States, there is a doctrine of “fair use” that allows limited use of protected copyrighted materialmaterials without the permission of the copyright holder for purposes such as criticism, comment, news, education, research. Four factors are taken into account when determining whether a use is “fair”:
  • The purpose and nature of the use (commercial or non-commercial, transformative or not).
  • Nature of the work protected by copyright.
  • The extent and significance of the part of the work used.
  • The impact of the use on the potential market or value of the copyrightedwork .
• Some lawyers believe that the use of works for training AI may fall under “fair use”, especially if it is non-commercial and transformative (i.e., theAI creates something new, not just copies original works). However, there is no unambiguous court practice on this issue yet.
• Exemptions for text and data mining (TDM) in the EU:In 2019, the European Union adopted the Directive on copyrightDirective on Copyright in the Digital Single Market, which introduced two exceptions for “text and data mining” (TDM), i.e. automated text and data analysis.
  • The first exception allows research organisations and cultural heritage institutions to perform TDM of works to which they have lawful access for the purposes of scientific research.
  • The second exception is broader and allows TDM by anyone, provided that the rights holders have not explicitly prohibited such use (the so-called “opt-out” mechanism).
• These exceptions are aimed at stimulating the development of AI in the EU, but at the same time leave the right holders with the ability to control the use of their works.
• Ukraine: Ukrainian copyright law does not yet have any specific exceptions or limitations that would directly address the use of works for AI training . However, there are general exceptions and limitations, for example, for the purposes of citation, education, scientific research, which could theoretically be applied to some cases of using works forAI training , but this requires a careful analysis in each case.

Thus, the question of whether it is legal to use copyrightedworks for AI training does not have a clear answer. It all depends on the specific circumstances, the legislation of a particular country and the interpretation of this legislation by the courts. However, it is clear that there is a need to adapt copyright law to the challenges posed by the development of AI, and possibly to introduce special exceptions and limitations that would strike a balance between the interests of authors and the public interest in the development of new technologies. It may be worth considering introducing such exceptions into Ukrainian legislation, taking into account the experience of the EU and other countries where such exceptions are discussed and implemented.

III. Liability for copyright infringement committed with the use of AI

If AI has generated copyright infringing content, who is responsible for it ? Developer AI? The user? Or perhaps the AI itself?

The traditional answer is “the person who committed the violation”. But with AI, it’s more complicated. Who is the “perpetrator”? Is the fact of using AI enough to be held liable?

In this section, we will look at different approaches to the allocation of liability in such cases, analyse court practice and try to understand how the law in this area may develop.

Subsection 3.1: Who is liable: the developer, the user or the AI itself?

The issue ofliability for copyright infringement committed with the use of artificial intelligence is one of the most complex and controversial in modern jurisprudence. The traditional model ofcopyright , focused on the human infringer, does not always work adequately in situations where the infringement involves AI.

Let us consider the main variants of liabilitydistribution :

1. Liability of the AI developer:
   • Pro: The developer has created an algorithm that has the potential to infringe copyright. The developer should have foreseen this possibility and taken measures to prevent it (e.g. filtering the content generated by AI for copyright infringement ).
   • Cons: The developer cannot control all the actions of the AI, especially if theAI learns and develops independently. Placing full responsibility on the developer may slow down the development of AI technologies .
   • Possible approaches: Limited liability of the developer, provided that he/she has taken reasonable measures to prevent infringement. Liability for “negligent” or “intentional” creation of an AI capable of infringing copyright.
2. Liability of the AI user:
   • Pro: The user gives the AI tasks, uses the content it generates, and therefore should be held liable for any infringements.
   • Cons: The user may not know and may not be able to verify whether the content generated by the AI infringes someone’s copyright. They may have a good faith misconception about the legitimacy of using such content.
   • Possible approaches: Liability of the user, provided that he or she knew or should have known that the AI-generated content infringes copyright. Liability for “negligent” or “intentional” use of AI for copyright infringement .
3. Liability of the AI itself (in theory):
   • Pro: If AI is recognised as a legal entity (which is currently unlikely, but is being discussed), it can be held liable for its actions like any other human being.
   • Arguments against: AI does not have consciousness, will, or property that could be used to hold it liable for the damage it causes. Bringing AI to justice is currently purely theoretical and has no practical meaning.
4. Rather, this option is currently purely theoretical and has no practical basis.

As we can see, there is no unambiguous answer to the question of liability for copyright infringement committed with the use of AI, yet. Each of the options has its pros and cons. And, most likely, the future law in this area will be developed by combining different approaches, taking into account the specific circumstances of each case and the degree of “autonomy” of AI. Perhaps, special rules will be developed to regulate liability in such cases, or the court practice will follow the analogy of law or analogy of law. In any case, this issue remains open and requires further research and discussion.

IV. Prospects for the Development of Copyright Law in the Age of AI

The development of artificial intelligence poses serious challenges tocopyright. Is the current legislation capable of effectively regulating relations related to the creation and use of works AI? Is it time to adapt it? In this section, we will discuss why changes to copyright are needed and what they might be.

Subsection 4.1: The need to adapt legislation

The question of whether copyrightlaw needs to be adapted to the challenges posed by the development of artificial intelligence is no longer purely theoretical. The current legislation, both in Ukraine and in most other countries, was created in an era when one could only dream of AI capable of generating content. It is focused on the human creator and does not take into account the peculiarities of the functioning and use of artificial intelligence.

Why is it necessaryto adapt the legislation ?

• Legal uncertainty: Current legislation does not provide clear answers to many questions related to copyright and AI. This creates legal uncertainty that hampers both the protection of authors’ rights and the development of technologies AI. For example, it is unclear who owns the rights to works generated by AI, how to determine liability for copyright infringement committed with the use of AI, whether it is legal to use works for training AI, etc.
• Protection of authors’ rights: Current legislation may not adequately protect the rights of authors when their works are used to train AI or when AI generates content that infringes their copyrights.
• Stimulating innovation: On the other hand, overly strict regulation may hinder the development of AI technologies , which have enormous potential for the public good. It is necessary to find a balance between protecting copyright and stimulating innovation.
• Alignment with international trends: Many countries are already actively discussing and implementing changes to copyright legislation in connection with the development of AI. Ukraine should also join this process to avoid being left behind in global development.
• Ensuring competitiveness: Adaptation of legislation can create favourable conditions for the development of the AI industry in Ukraine, attracting investment and creating new jobs.

In other words, the need to adaptcopyrightlegislation to the challenges of the AI era is dictated by both practical needs (elimination of legal uncertainty, protection of authors’ rights, stimulation of innovation) and strategic considerations (compliance with international trends, ensuring the country’s competitiveness).
This does not mean that the legislation needs to be completely rewritten. But, of course, there is a need for point changes that would take into account the peculiarities of the functioning of AI and ensure a balance of interests of all stakeholders. We will discuss what these changes might be in the next section.

Sub-section 4.2: Possible areas of regulation

The question of how to adapt copyrightlaw to the era of artificial intelligence is a complex and multifaceted one. There is no single universal solution that would satisfy everyone. However, there are several possible areas of regulation that are being discussed in the legal community and implemented in some countries:

1. Clarification of the concept of “author”:
   • Retain the traditional approach: Retain the definition of “author” as a natural person, emphasising that AI cannot be an author. In this case, works generated by AI may either not be protected by copyright at all or protected under a special regime (see below).
   • Expanding the concept of “author”: Include legal entities that use AI to create works in the concept of “author” . This approach could be applied, for example, in cases where a company invests heavily in the development and training of AI and then uses it to generate content.
   • Recognition of “authorship” by the person who made the contribution: As in the UK, the person who ensured the creation of the work is considered the author.
2. Special legal regime for works generated by AI:
   • Introduce a “sui generis” right: Create a special type of right, distinct from traditional copyright, to protect AI-generated works . This right could be shorter in duration, have different restrictions than copyright, and be owned, for example, by the developer or user of the AI.
   • Application of the “works created in the course of employment” regime: If AI is developed and used by a company, the rights to the works generated by it may belong to the company, as in the case of works created by employees in the course of their employment.
3. Clarification of exceptions and limitations to copyright:
   • Introduction of exceptions for “text and data mining”: Similar to the EU, exceptions could be introduced to allow the use of copyrighted works for training AI (text and data mining) under certain conditions (e.g. for non-commercial research or with a mandatory “opt-out” for copyright holders).
   • Expanding existing exceptions: Existing exceptions and limitations tocopyright ( e.g. for quotation, parody, caricature ) could be expanded to cover some uses of AI-generated works .
4. Regulation of liability:
   • Allocation of liability: Clearly define who is liable for copyright infringements committed with the use of AI: the developer, the user, or, in some cases, both.
   • Introduce liability insurance: Consider introducing mandatory or voluntary liability insurance for AI developers and/or users .
5. Encourage transparency

• Content labelling: Encouraging or even requiring labelling of AI-generated content.

These are just some of the possible areas of regulation. Of course, each of them has its advantages and disadvantages, and the choice of a specific approach requires careful analysis and discussion, taking into account the interests of all stakeholders: authors, developers of AI, users, and society as a whole. It is important to find a balance between protecting copyright and stimulating the development of AI technologies that would contribute to progress, innovation and cultural development.

Conclusions.

Copyright and artificial intelligence… Not long ago, these concepts existed in parallel realities, but today they have collided, giving rise to a number of complex but fascinating questions. We have come a long way from the traditional understanding of the “author” as a human creator to the realisation that a machine, although not endowed with consciousness, can generate content that, at least outwardly, looks like the result of creative activity.

We found out that:

• The issue of authorship of works generated by AI remains open and controversial. The world has not yet developed a unified approach.
• The use of copyrightedworks for AI training is not a black-and-white situation, but a zone of finding a balance between the interests of authors and progress.
• The problem of liability for violations committed with the use of AI requires innovative solutions.
• Current copyrightlaw needs to be adapted to the new realities associated with artificial intelligence.

What will be the future of copyright in the AI era ? There is no clear answer. Perhaps there will be new, “hybrid” forms of authorship, where humans and machines will cooperate. Perhaps special legal regimes will be created for works generated by AI. Perhaps copyright will evolve into something completely different, adapted to digital reality.

But one thing is certain: Artificial intelligence has already changed the world, and copyright cannot stay away from these changes. We, lawyers, legislators, and society as a whole, will have to find answers to new challenges, balance between rights protection and innovation, and create a new legal reality that will be in line with the spirit of the times. And this work is already underway. And we, Polikarpov Law Firm, are closely monitoring the development of events and are ready to provide qualified legal assistance in this exciting and dynamic area.