The world where people commonly use artificial intelligence (AI) in their daily work has become a reality. Already more than 90%[1] of developers say they use AI tools such as ChatGPT, Microsoft Copilot or Github Copilot when developing software. In parallel, there are still discussions among lawyers about the copyright of the results created with AI systems. The conclusions of these considerations have real and crucial implications for economic turnover. The assumption that software elements created with AI tools (e.g. source code fragments) are (or are not) copyrighted works has extremely important consequences both for IT solution providers and purchasers of these solutions, but also for individual employees and contractors.

In practice, the above qualification translates into an assessment of, among other things, the possibility of granting a licence or transferring (proprietary) copyright to the buyer (the end customer or, previously, to an employer or principal) and, as a result, determines the scope of carrying out of the concluded agreements and contracts. Moreover, the above issue also has a direct impact on other areas in the IT industry, including the assessment of the eligibility of tax reliefs aimed at creators.

Below we present the current status and conclusions of the ongoing legal debate in Poland (as well as globally) on the fundamental copyright issues of using AI tools for software development.

Computer program as a work

In most jurisdictions, computer programs are one category of works protected under copyright law. They are most often subject to protection as a separate category of works. Thus, in order to assume that a computer program that has been created with an AI tool is subject to copyright protection, it is required to examine whether the prerequisites for its recognition as a work (as in the case of other types of works that are not computer programs) are met. The rationale may be worded differently in different jurisdictions but will most often be similar in nature. Under Polish copyright law, for a computer program to be considered a work, the program must:

  1. be a manifestation of human creative activity,
  2. have an individual character,
  3. be determined.

Although the discussion on this topic is still ongoing, most lawyers agree that in the current state of the law (including both Polish and European law), in order for copyright protection to be granted, a given result must be the result of the creative activity of a human being and not solely that of a machine[2]. Thus, in order for a work to come into being, it is necessary for a human being to have a decisive creative influence on the creation of the result in question. This leads to the conclusion that if a computer program was generated automatically, without any creative contribution by a human being, or if this contribution was minimal (trivial and not exceeding the limit of copyright protection), such a program would not be a work within the meaning of Polish copyright law[3].

On the other hand, this allows for the assumption that the mere use of AI tools in the process of creating a given result does not automatically exclude the possibility of considering the program as a work within the meaning of copyright law and granting it appropriate protection. The conditions must be met that a human being appears in the creative act and that their role in the process of producing the final result— a computer program—is creative. In view of this, determining the degree and extent of human creative contribution to a given result created using AI tools requires analysis on a case-by-case basis, taking into account the specifics of the tool in question and the manner and extent of its use by human beings.

When does a human being have a decisive influence?

It is assumed that the result of human labour, and thus the manifestation of creative activity, is only those works whose creation was decisively, creatively influenced by a human being. What does this mean and where should the line be drawn from which copyright protection ceases to apply?

The legal literature points out that a creation (work) is a manifestation of creative activity as a whole and it is not necessary that all its elements have a creative character, nor that the activities leading to its creation are creative[4]. As can be seen, this is also the case in practice when a developer develops software without AI tools. This is because many of the activities that make up the software development process are routine and repetitive in nature and therefore not creative, even if performed by human beings alone.

Thus, if a “non-human” factor was involved in the process of creating a computer program, i.e. the developer was partly “relieved” by a tool (including one based on AI), but the final result in the form of a computer program was nevertheless decisively influenced by a human being, then, provided the other prerequisites are met, this may still be a work within the meaning of copyright law.

This impact can be assessed by examining what (of what rank) creative decisions were made by a human in the development of a given program, and what proportion of these decisions was made in relation to those “made” by the machine. The answer to this question will depend on the specifics of the particular AI tool used in a particular process. This is because these tools work in different ways; they can, for example:

  • completely generate a virtually ready-to-use computer program with specific functions (at the moment, this is rarely used in practice);
  • generate smaller code snippets based on user instructions (these can be just snippets but also entire functions);
  • self-complete minor code elements during the writing process, e.g. based on the documentation of the programming language or program under development or on user-learned programming practice;
  • assess the quality of the code (identify weaknesses and bugs or security holes);
  • provide recommendations for optimisation and error correction, including suggesting replacements for code already created;
  • rewrite code from one programming language to another;
  • help solve a given problem by providing an idea (concept) for solving it, rather than generating the code itself (so-called brainstorming).

In each of these cases, the human creative contribution to the final result is slightly different. In general, however, it can be assumed that in most of them, the developer, despite using an AI-based tool, still makes more or less creative decisions that demonstrate their input, e.g. deciding whether or not to accept the solution offered by the tool[5], how they will possibly modify it, and when to decide to use the tool at all and when not to.

Furthermore, the result generated by the tool is usually the result of prior human input, as it is the human being who directs the AI tool’s operation by indicating appropriate instructions, e.g. determining the goals or requirements to be met by the program. It is also the human being who puts the program elements together, determines their order and how they are connected. In other words, in most cases the final program is the result of a creative combination of multiple elements: those created by a human being and those generated by a machine. Furthermore, it is important to bear in mind that at the current stage of development of AI tools creating functional software will almost always require appropriate modification of the parts suggested by the tool and adding many elements on our own.

It can therefore be assumed that in the current state of technological development, in most cases the AI tool will only be a means of supporting human work and externalising its intention. In contrast, it is the human being who plays the key role, making a number of creative decisions and guiding the entire creative process, being aware of the context, the objectives of the project, and the potential constraints and requirements. Therefore, working on software development, even with the help of AI tools, undoubtedly requires the application of knowledge and experience, as well as the intellectual engagement of a human being, who remains the main architect and decision-maker in the creative process.

However, it should be borne in mind that the use of AI tools tends to result in relatively fewer of these creative elements than in the case of software development without such support. In the extreme case, if the human input required to achieve the final result is no longer significant enough to consider the whole as the result of human labour, it will be deprived of copyright protection. Overstepping this boundary in the case of computer programs may prove to be relatively straightforward, especially as copyright protection for computer programs has itself always raised some questions. These include the fact that the shape of the program is largely dictated by technological considerations and a desire for optimisation, or is the result of learned, schematic skills. It can often be a bit of a challenge in the case of simple computer programs just to demonstrate their creative originality and the elements that would individualise the creator. Meanwhile, the use of an AI tool can further undermine these characteristics.

On the other hand, it can be concluded that, since the use of AI tools primarily eliminates the need to do the most routine work, which is not creative anyway, even if a human being would do it themselves, the degree to which a program is a manifestation of creative activity changes only minimally. However, it should be borne in mind that this will always depend on the specific case and each situation should be viewed individually.

It is also possible that with the technological development of AI tools, which are, after all, constantly being improved, the role of human beings in the creation process will be increasingly diminished and the protection of such works will be increasingly difficult to defend under copyright law.

And that is not all yet

In addition to the fundamental issue of the qualification of the results generated, there are a number of additional, collateral concerns under copyright law associated with the use of AI tools. You can find out more about them in our articles here and here.

Why is this important?

The conclusions of the above considerations have a real impact on business transactions. Determining that software developed using AI tools does not constitute a copyrighted work raises extremely significant implications for the entire IT industry, both for solution providers and customers, as well as for individual employees and contractors.

As an example, it is worth recalling the tax benefit commonly used in the IT industry in Poland in the form of accounting for 50% of deductible costs by employees. According to the tax law, one of the basic conditions for receiving this benefit is the transfer to the employer, in exchange for a royalty, of the proprietary copyright in the works produced in the performance of one’s official duties. In the IT industry, this is done, for example, by transferring the rights to computer programs made by the developers employed by the organisation. In turn, with this benefit, employees can significantly reduce the amount of income tax they pay, which provides them with a great advantage without additional payments on the part of the employer.

Therefore, if we consider that the fact that an employee uses AI tools in the course of creating works (e.g. software) means that the result produced will not constitute a work within the meaning of copyright law, such an employee cannot dispose of these rights and therefore cannot benefit from this tax benefit (cannot account for 50% of deductible costs), and would therefore be obliged to pay higher tax (not to mention the consequences of erroneous accounting in case of an audit by the tax authorities). In short, a programmer who does not produce works within the meaning of PrAut in the course of his professional duties cannot account for 50% of the costs (for more on tax benefits in the IT industry, check here). Therefore, the use of AI tools may also affect the applicability of tax credits targeted at creators.

In the above case, an employer who supplies its clients with a solution created by its employees would also not, for example, be able to transfer the copyright of the created elements to its client or grant a licence (since these elements are not works), and thus would often be unable to properly carry out the concluded contract or agreement. Worse still, elements of a computer program not covered by copyright protection could be freely copied and modified by others, including competitors (of course, we are leaving aside here the issue of protection under unfair competition law, including as a business secret), who could profit from this.

What to do?

In the new reality of widespread use of AI tools, the issues described above are crucial. Importantly, while we primarily analyse them in the context of the IT and software development industry, the issues presented are also applicable to other industries where content is or can be created using AI tools, including in particular the marketing and creative industries. So, firstly, it is important to be aware of these legal concerns and not overlook them when implementing and using AI-based solutions.

Secondly, when implementing AI tools, it is worthwhile to carry out a legal analysis of the scope and use of such solutions in the software (or other content) development process and to define this scope accordingly, and ultimately to set specific rules for the use of AI tools to be communicated to the staff. In particular, it is worth defining who can use the AI tools being implemented and for what purpose, and then transferring the new rules into internal regulations (e.g. rules for the use of AI tools) or into contracts and agreements with employees and contractors.

When applying tax benefits (e.g. a 50% deductible accounting system), one should analyse the creative work process and the extent to which employees are using AI tools, and then adjust the 50% deductible accounting rules accordingly. This will avoid unpleasant surprises in the future in terms of possible inspections by the tax authorities.

Thirdly, the principles of using AI tools need to be taken into account in business relationships with contractors and when entering into contracts and agreements. Above all, it is worth discussing and agreeing on these issues with the contractor in a businesslike manner first, and then transferring the arrangements into contracts and agreements by introducing appropriate provisions on the use of AI tools.

The above actions will reduce the most important legal risks arising from doubts about the qualification of the results generated by AI tools that arise under copyright law. This, in turn, will enable the informed and safer use of this revolutionary technology in the operations of the organisation.

[1] https://github.blog/2023-06-13-survey-reveals-ais-impact-on-the-developer-experience/

[2] R. Markiewicz, “ChatGPT i prawo autorskie Unii Europejskiej” [“ChatGPT and the European Union Copyright Law”], https://assets.contenthub.wolterskluwer.com/api/public/content/57183ddaf8274d279a4f0b0eee4695f8?v=fb63f500.

[3] A. Niewęgłowski [in:] Copyright Law. Commentary, Warsaw 2021, Article 1.

[4] A. Niewęgłowski [in:] Copyright Law. Commentary, Warsaw 2021, Article 1.

[5] Statistics for GitHub Copilot show that only 30% of the suggestions are immediately accepted, see: https://www.itpro.com/technology/artificial-intelligence/github-30-of-copilot-coding-suggestions-are-accepted.