In the previous parts of this series of blog posts, we have discussed the use of copyrighted material as training data for an AI system and the question of the extent to which the results generated by an AI system can be copyrighted.
Are AI models copyrightable – or protectable by any IPR?
The final area that we are covering is the valuable parts within an AI system itself. In many cases, as discussed earlier, a machine learning system takes large amount of data as an input and using sophisticated computer programs produces a model from it. This model includes all the essential characteristics of the original data, but is much smaller, portable and effective. Therefore, the model can have a lot of business value. However, it is likely that no intellectual property right covers it.
As noted in the previous sections, to qualify for copyright protection, a work must be original, the author’s own intellectual creation. The author must have made creative choices in creating the work. Since computer programs produce the model automatically, the process does not usually involve creative choices. Therefore, the model is unlikely to be protectable by copyright. Only if a human being contributes to the development of the model by making creative choices can the end result be copyrightable.
As discussed in the previous sections, it is also possible that a copyrighted work included in the training data is copied into the model. In this case, the copyright holder of the work may also become a copyright holder of the model. The same applies to other IP rights, like trademarks, for example: if a protected trademark is copied into an AI model, it is possible that the trademark owner (given the scope of the trademark) has a say in the use of the model.
Usually, an AI model is not covered by database sui generis right either. That is because the sui generis right according to the EU directive 96/9/EC on the legal protection of databases requires a substantial investment in terms of quality and/or quantity in the obtaining, verification or presentation of the contents of the database. It is usually not the case with AI models. Although it may require significant investments to develop them, the investment is not targeted to the obtaining, verification or presentation of the contents of a database, but rather in the process of automatically creating the model, which according to the Court of Justice of the European Union (C-203/02 “Horseracing”) does not qualify for protection: “The expression ‘investment in … the obtaining … of the contents’ of a database […] must be understood to refer to the resources used to seek out existing independent materials and collect them in the database. It does not cover the resources used for the creation of materials which make up the contents of a database.”
In addition, the Nordic countries still have what is known as a ‘catalogue right’. If a large number of information items is compiled into a catalogue, it is protected by this catalogue right. It should be noted that this right does not require the same kind of investment as the database sui generis right. Thus, if an AI model is considered as a compilation of a large number of information items, it is possible that the model is protected in the Nordic countries.
Could there be something patentable in an AI model? To be eligible for a patent, an invention needs to be new, involve an inventive step and have industrial application. An invention is held to involve an inventive step if it is not obvious to the skilled person in the light of the state of the art. As an AI model is produced automatically by a deterministic computer program, it hardly includes inventive or non-obvious steps, but rather a large amount of non-inventive, obvious steps.
Therefore, it seems that the most valuable part of an AI system can remain outside the intellectual property rights regime.
Conclusions
Depending on the algorithm, machine learning process may be relevant from a copyright point of view. If the training involves copying the creative choices made by the original author of a copyrighted work in the training data, it could violate the author’s exclusive rights. On the other hand, if the machine learning process can be considered as data mining, it can be within the limitation or exception defined in the DSM directive and therefore lawful within the EU. Yet, if the output of a generative AI system includes copies of the works in the training data, that cannot be justified by that limitation or exception.
Both the results generated by an AI system and the most valuable parts of the system itself can arguably remain outside the copyrightable subject matter. There will be business incentives to lobby lawmakers for additional protection. This creates a risk to have yet another specific set of rules in the copyright regime, which already is complex, detailed and casuistic. To lawmakers, we wish the resilience to resist the temptations of lobbyists, and not to add complexity to the system anymore.
1001 Lakes’ experts are happy to discuss these topics with you if you have concerns of AI and copyright or how to develop and use AI in compliance with the IPR laws.
Olli Pitkänen, 
Waltter Roslin, 
Sami Jokela, 