On February 27, 2023, the Finnish Parliament approved amendments to the Finnish Copyright Act (404/1961) and the Finnish Electronic Communications Services Act (917/2014). This is the most significant copyright law change in Finland in the last 20 years. The new copyright law replaces the much-debated directive on copyright and related rights in the Digital Single Market (the so-called DSM Directive, 2019/790/EU) and the copyright and related rights applicable to certain online transmissions of broadcasts. enforce the Directive on exercising the right to Organizations and retransmissions of television and radio programs ((EU) 2019/789).
The main objective of Finnish copyright law and the DSM Directive is to modernize copyright law. Its goal is to make copyright law reflect the various phenomena of today’s modern society, arguably the most important and challenging phenomenon being artificial intelligence (AI), which has made great strides in recent years. included. Will the new copyright law provide new answers to copyright issues related to AI applications? As AI technology will be essential for businesses in the future, companies should consider how legislative changes could change the rules around AI. It is necessary to consider whether there is a
In this article, we take a legal look at how the new copyright law will affect the evaluation of AI training. Copyright issues related to AI applications have been discussed at a general level in previous articles.
New Copyright Exception for Text and Data Mining in Section 13b of the Copyright Act
The most significant legal challenges associated with AI-assisted content generation include the risk of copyright infringement that may arise during the AI application training process. AI content generation is often based on machine learning. During machine learning, algorithms are trained to look for patterns and correlations within large datasets. This is also how ChatGPT, which is very popular and available on the internet, works. With the help of machine learning, AI applications make decisions and predictions and learn how to separate relevant information from irrelevant information.
Regarding copyright, issues arise when training data contains material that is protected by copyright or related rights. For example, the open-source code of the image generation tool Stable Diffusion has been observed to train a generative model using a large set of copyrighted images without the consent of the owner.
The question of whether it is permissible to use copyrighted works for machine learning purposes has long been unclear in our legal system. For example, Finland does not enforce copyright restrictions that specifically allow copying of works for the purpose of training AI applications. The new copyright law does not change the situation, as it does not contain explicit restrictions. However, the new copyright law includes provisions for text and data mining in Article 13b, which implements the text and data mining provisions of the DSM Directive, which may clarify the legal assessment of machine learning to some extent. I have.
According to Section 13b(1) of the new Copyright Act, anyone with legal access to a work may make copies of it for the purposes of text and data mining and keep such copies solely for that purpose. is allowed. expressly and properly reserves this right. In other words, text and data mining are now generally allowed unless explicitly prohibited by the author.
Application of AI to learning is unclear
At first glance, the new provisions appear to significantly clarify the legal situation. However, it is unclear whether the new Section 13b applies to training AI applications, so it does not yet provide a shortcut to success. The DSM Directive does not mention AI or machine learning, so this issue is not addressed. Similarly, the government’s proposal for a new copyright law does not offer an opinion on whether its provisions apply to the operation of AI.
A narrow interpretation of the definitions of text mining and data mining can be thought of as applying only to activities aimed at performing analysis. Traditionally, text and data mining have been defined as “automated analytical techniques designed to generate information, trends, or correlations.” In this case, the operation of AI is considered outside the scope of data mining and text mining regulations. After all, the goal of training AI applications is often to create content, not just analysis.
However, text mining and data mining may be given a broader interpretation in legal practice, in which case Article 13b may also apply to artificial intelligence. As the situation is unclear at this time, we will have to wait for a clear court ruling or more detailed guidance from the authorities.
However, if the new provisions on text and data mining are expected to apply to the AI training process as well, the reform could be seen as positive for AI developers. In the future, data mining will be allowed unless explicitly prohibited. This provision promotes free competition and freedom of trade in situations where the work is used only indirectly, i.e. where copies are made solely for the purpose of analyzing information using technical methods. can be regarded as
According to the DSM Directive, the exercise of rights should be explicit and proper. This could be, for example, terms of a subscription to a content service, or terms in machine-readable form. It will be interesting to see how this right is exercised in practice. Some international artist portfolio platforms already use the metadata tag “NoAi” that can be added to the metadata of your work. In the future, the range of different techniques may expand.
future outlook
As noted above, the new copyright law does not significantly clarify copyright issues related to artificial intelligence. The new copyright law contains no special restrictions on AI training, but the provisions on text and data mining are fairly close. Whether or not this provision applies to AI is still uncertain, so the question of whether AI is allowed to study on copyrighted material remains unclear.
