In recent months the topic onartificial intelligence has held and continues to hold court, thanks to AI systems capable of learning thousands of documents such as texts or images, then producing creative contents, but can they be covered by copyright?
Artificial intelligence and copyright: where are we?
NordVPN, Best VPN service in Italy
This has led many to question the possibility of protecting works created by AI with copyright, but is it really possible? After all, isn’t it the human hand that lays the first piece, then involving artificial intelligence in the construction process?
Summing up, who is the author of the work generated by an AI? Can copyrighted content be used to power AI systems?
Second Lydia Mendolapartner of the Portolano Cavallo law firm to date “there is no law that directly and unequivocally answers this question”.
“The conclusion reached – continues Mendola – is that, at present, the work generated by an AI model cannot find protection in copyright due to the lack of human input in the creative act. Both the EU and most of the national ones in the world have taken the position according to which AI models cannot be qualified as authors of a work, and therefore the content produced by a generative AI model cannot be considered a work protected by copyright, unless an appreciable creative contribution of human beings can be identified.
Thus, in the absence of creative input from a human being, there is a possibility that AI-generated works fall into the public domain. The author will thus have to demonstrate how the artificial intelligence model has represented a moment or a tool within a more complex creative process.
For example, recently in the USA, the US Office of Copyright granted the registration of a cartoon generated with the help of the “text to image” artificial intelligence Midjourney, but only because it was properly exploited l human contribution to the creative moment. In any case, the decision is still being examined by the Office which has reserved the right to further investigate precisely the point concerning the creative contribution of the user of the AI model used, i.e. the artist who created the work.”
However, in recent weeks the famous Springer Nature had prohibited the enumeration of Chat GPT as author within the articles. So it seems even more difficult to predict or understand how an artificial intelligence can reserve the right to copyright.
One more question
Can content produced by third parties and protected by copyright be used to “train” artificial intelligence models?
“About this – he replies again Mendola – there are many doubts. Most AI systems are trained using enormous quantities of content collected from the web (texts, codes, images, etc.), through scraping, but it is not possible to give a single answer on the legitimacy of using this technique.
In the United States the doctrine of fair use it is invoked to legitimize the mass collection of digital content, or to legitimize the use of any material protected by the copyright of others in the absence of authorization from the copyright holder. However, this doctrine does not find immediate and direct application in Italian or European law.
This does not mean that the European and Italian legislators have not addressed the problem of balancing the rights and interests of authors and other right holders on the one hand, and of users on the other, with respect to some new types of use of digital works.
With the introduction of the exception “Test and Data mining” contained in the Directive on copyright and rights connected in the digital single market, mass mining of digital data and their reproduction is freely permitted provided that the use of the extracted works and other materials has not been expressly “reserved” by the right holders in an appropriate manner. In other words, it is the owners of exclusive rights who must take steps, with appropriate means, to protect their works and ensure that they are not subject to massive data mining.
As for the ways in which this reservation must be expressed and brought to the attention of third parties, the Italian legislator does not give specific indications. Instead, for example, the German or Dutch legislators were more precise.”
Lydia Mendola also answers one last fundamental question, concerning the acquisition of copyrighted data.
If the acquisition of copyrighted data is legitimate, then the work generated by the AI model would also be legal?
“Even where an artificial intelligence model is trained using content protected by the copyright of others legitimately, this does not exclude that the output of the AI model can still integrate a copyright infringement” notes Mendola, who concludes:
“Let’s think of a “text to image” model of AI. If the model is trained on many millions of images and used to generate new images, this is extremely unlikely to constitute copyright infringement as the end result will be very different from the original works.
But if images of a specific artist were used as a model, with the aim of generating works that reproduce his style, his technique, and therefore be confused with his original work, then the artist in question could oppose the circulation and exploitation of the new work generated by the AI model, even if it has not expressed any reservations upstream with respect to the scraping of its contents. For example, it could complain about the existence of a evolutionary plagiarismwhich occurs when the original (plagiarized) work is still recognizable in the new work.
And at this point the question becomes: who is responsible for the possible illegal activity? Who is responsible for plagiarism? The generative AI model, its programmer, the company that owns the related platform, or the user who interrogated the AI model to obtain the plagiaristic work? Also in this case, the answer is not unique and it will be necessary to investigate the steps of the creative process that led to the production of a certain content.”
Faced with these answers, we cannot help but think of the complexity of the issue and the difficulties that the regulatory framework will have to unravel, especially in the face of the awareness that the future will increasingly focus on artificial intelligence.