Our next guest on A Little Privacy, Please!® is Jakub Pietrasik, who leads the Intellectual Property and Technology Media and Telecom practice at Wolf Theiss in Poland. Jakub will speak about generative AI both generally and from the perspective of an attorney practicing in Poland.
First, the basics — What is generative AI?
Generative artificial intelligence (AI) is a pretty new technology gaining popularity due to its ability to generate realistic content like images, audio, or text. It is used in business, for example, in marketing agencies, or even in IT companies, to create marketing slogans, marketing campaigns, or even software code. In practice, AI-generated content looks like human-made content, which is why it is so interesting and useful in real life.
Does generative AI have the potential to impact businesses in their regular course of business?
It creates many risks and threats. It is unclear who the author and owner of the AI-generated content is. In some countries, like the UK, there is a concept of computer-generated works, but in Poland, the author can only be a human.
This generates certain questions. If the AI creates a marketing slogan or campaign, can I use it for commercial purposes? From a practical angle, there is an answer in the general terms and conditions of some generated AI content providers. According to these terms, usually, the user has full rights to use the output generated by the system or at least has a very broad license. It is always risky when it comes to money and big business.
Are there risks to people or businesses that are associated with the use of generative AI?
Apart from the authorship issue mentioned, another aspect relates to the professional infringements of copyrights. As we know, the AI system is able to create content thanks to the huge amount of data content provided to it in the development stage. When we talk about images, it is unclear what the relation is between the original images and new AI-generated images and whether the new images should be regarded as derivative works. Right now, it is not clear whether you need permission from the original author to use the AI generated content, and whether the AI generated works would qualify as derivative works.
Are there privacy issues implicated by the use of generative AI?
This is another very important layer. It should be specified that AI tools can be dangerous when we feed them confidential information. We should remember that all data that’s provided to the system is retained on some server and then used to train AI models. In other words, everything we give as input data can later be used to create output for other users.
This means that if an employee provides software code to the system or the description of a patent or invention, which is often a business’s most valuable asset, the same or similar content can be generated for other users.
The same applies to personal data. When users give personal data to the AI system, special data protection issues arise, including the information obligation toward data subjects, scope of processing, retention periods, or international data transfers. This last issue, transferring data outside the European economic area, is very sensitive for EU-based entities.
Has Poland enacted any laws restricting private use of generative AI technology in a company environment?
On the level of the European Union, there is a pretty advanced legislative process aiming to regulate AI in general, including general purpose AI systems. There is a draft AI Act, which is a regulation providing a general legal framework applying to AI systems with a special focus on high-risk AI systems, that will impose specific regulatory obligations and requirements regarding risk management, data governance applications, transparency issues, and many more.
Additionally, two proposed acts regulate liability issues of the new product liability directive and AI liability directive. The first deals with liability for AI systems that will be qualified as products, and the second provides procedural rules relating directly to AI.