Generative artificial intelligence

Scope

The EU Artificial Intelligence Regulation contains no definition under the name “generative artificial intelligence.” The Regulation defines models that can be trained with large amounts of data, display significant generality and can perform a wide range of different tasks as a general-purpose AI model (Art. 3/63), and systems based on such a model that can serve a variety of purposes as a general-purpose AI system (Art. 3/66); its recitals regard large generative models as a typical example of this. Providers of systems that generate synthetic audio, image, video or text content are obliged to ensure that the outputs are marked in a machine-readable format and that it can be detected that they have been artificially generated, while those who use systems that generate deepfake content depicting existing persons, places or events as if real are obliged to disclose that the content has been artificially generated (Art. 3/60, 50/2, 50/4). In Türkiye, the Personal Data Protection Authority (Kişisel Verileri Koruma Kurumu) has published the “Guide on Generative Artificial Intelligence and the Protection of Personal Data.”

Practical significance

Prompts entered into generative artificial intelligence tools and documents uploaded may contain personal data, trade secrets or information subject to a duty of confidentiality; if personal data is involved, the principles and processing conditions of Law No. 6698 and, if the service is provided from abroad, the rules on transfers abroad (Art. 9) may also come into play. As regards the content produced, Law No. 5846 on Intellectual and Artistic Works defines a work as an intellectual and artistic product bearing the distinctive character of its author, and the author as the person who creates the work (Art. 1/B); because there is no provision specific to AI outputs, whether such outputs are protected as works and who holds the rights is disputed. Because these tools can also produce content that is false but convincing, outputs must be verified in legal texts and citations of sources.

Commonly confused concepts

Generative artificial intelligence is a subtype of artificial intelligence; systems that perform classification, prediction or scoring may also be artificial intelligence but do not generate new content. Nor does the “general-purpose AI model” in the EU Regulation overlap exactly with generative artificial intelligence: a model is a component of a system, and not every general-purpose model may be used to generate content. Deepfake content, in turn, is a specific type of output of generative artificial intelligence; it is not the concept itself.