The possibilities brought by artificial intelligence are being discussed everywhere. Equally, the uncertainty and risks it brings are sparking a lot of debate. Can a company's trade secrets escape into a competitor's knowledge? Who owns the copyright to material produced by artificial intelligence?
Artificial intelligence (AI) does not yet have a universally accepted definition. Generally, it refers to a computer or computer program's ability to utilize human-like attributes such as creativity, planning, reasoning, and learning. AI can help create entirely new products and services or improve existing ones. On the other hand, AI that utilizes large amounts of data raises many threats and ethical questions.
There are already numerous cases worldwide where confidential information has ended up in an AI application and thus become widely accessible. For example, Samsung's source code has leaked into the ChatGPT application. Legal disputes have been settled regarding copyrights to creative content such as images and text.
The rapid growth in the use of AI raises many questions, and the European Union is currently enacting AI regulations. Kolster's experts Hannes Kankaanpää and Jussi Ilvonen respond to current questions about AI.
Jussi: There have been questions worldwide about whether AI software can be considered the inventor of an invention or the creator of a literary or artistic work in a legal sense. So far, in most cases, the answer has been negative, but some of the related legal and administrative processes are still ongoing.
If patents and copyrights were to be more widely granted to inventions and works created by AI, i.e., by someone other than a human creator, it has been speculated that this would dilute the protection opportunities for existing rights holders. The capacity of modern generative AI applications to create content compared to humans is enormous. In terms of patents, it can be thought that the relatively high costs of the patent registration process limit this flood of protection, whereas copyright registration is not generally required at all.
In trademarks, the situation appears relatively clear. The registration of a trademark does not require a "creator" of the trademark, so a trademark created by AI can be registered by a company or individual who, for example, owns the AI application that created the trademark or commissioned the design of the mark from a company that, in turn, gave the task to its owned AI application. Instead of the owner, the user of the AI application can also act similarly in principle, although in this case, it is important to consider any restrictions arising from the terms of use of the AI application.
Jussi: It is somewhat unclear in which situations, when training AI, permission from the copyright holders must be sought for the use of copyrighted material. For example, the current EU copyright regulation allows the use of works obtained from legal sources for the purpose of data mining and data extraction. However, it is not clear in all cases when AI training involves data mining and extraction as intended in EU legislation.
Many questions do not yet have clear answers, and therefore, various legal disputes are expected on the subject until the legal situation begins to clarify.
Jussi: Yes, if we consider that by feeding the company's information into the system, it is no longer considered or can be considered secret. Of course, the terms of use of the AI application and specific provisions and case law in different countries may have an impact.
Hannes: Trade secret law and the directive require that the legal holder of the trade secret has taken reasonable measures to protect it. If a large group of people has access to information, for example, through an AI application, and/or the trade secret is not subject to confidentiality, it can be difficult to prove that the information has been adequately protected.
Hannes: It depends on the contractual terms, i.e., what is stated in the terms of use of that specific application or service regarding the AI outputs and the rights to them. The significance of contractual terms is emphasized in AI-produced content that does not generally enjoy copyright protection.
Jussi: Various situations are possible, and all sorts of demands may be made without the legal basis being in place. In such cases, it is advisable to consult a lawyer in advance. The claimant may have a relatively strong position, for example, when it concerns a registered trademark created with the help of AI, and another company later seeks registration for an identical or confusingly similar mark. A more ambiguous situation arises if the AI provider invokes copyrights in a situation where AI has played some creative role.
Jussi: The starting point is that the code created by AI cannot be protected by copyright because a human has not been behind the creative solutions of this "work."
Hannes: We do not know how, for example, ChatGPT has been coded. Does the application prevent it from copying other people's code in any way, or does it take into account, for example, the terms of open-source licenses so that the code can also be used for the intended purpose? It is advisable to be cautious with coding intended for commercial use. The output always requires scrutiny from the perspective of quality and compliance, i.e., compliance with laws and regulations. There have been cases where employees of companies have fed ChatGPT their own code considered a trade secret for review. Thus, it has become part of the AI system's data bank, and the trade secret has been lost.
Jussi: It depends on the evidence. Proving borrowing is possible if, for example, copyright-relevant parts of the code that are exactly the same as those in the licensor's code can be shown. In cases of infringement, it must be separately assessed whether the use has occurred based on permission or a restriction or whether the borrowed code should be considered a work protected by copyright.
Hannes: In the case of open-source codes, proving "borrowing" can be relatively straightforward in some cases. The original code is available online, and there are various automated tools for scanning and identifying code. An AI application can inherently operate in such a way that it utilizes existing material with which it has been trained and/or which it mines from the web.