Desafíos actuales de la Inteligencia Artificial

Generative AI content misuse and the DSA 93 duction, the exact nature of this role needs careful assessment. The CJEU judgement in Papasavvas clarifies that a digital service provider falls outside the intermediary taxonomy of the DSA if they make all critical decisions about the hosted content 37 . In Papasavvas, this was evident, as the content was produced by journalists of a digital newspaper, the main defendant in a defamation case. However, Generative AI applications differ in that they are primarily tools used by individuals. These applications do not decide to create content nor influence its parameters, direction, tone, or substance; these decisions are made by the users 38 . Even if Generative AI applications significantly aid in synthesising output, content creation involves more stages than just producing the output. Thus, determining the actual creator of AI-generated content requires examining several parameters. Each use case might also need individual assessment, as the levels of human and AI involvement in content production can vary. It appears, therefore, impossible to prima facie classify all Generative AI applications as actual content creators. Third, the argument that Generative AI applications play an active role beyond what is permissible under art. 6 of the DSA, especially in light of the findings of the CJEU in the L’Oreal case, is not entirely convincing. To begin with, L’Oreal might not be the best precedent, particularly given subsequent CJEU case law offering a more nuanced approach to content-related applications. For example, the CJEU’s approach in YouTube and Cy- ando 39 differs from L’Oreal. In this case, the Court considered whether content storing and sharing applications could benefit from art. 14 of the Electronic Commerce Directive (the predecessor to art. 6 of the DSA). The Court did not rely on L’Oreal but instead assessed YouTube’s and Cyando’s business models independently. It ruled that features like automatic content indexing, search functions, and video recommendations based on user profiles did not imply that the operator had specific knowledge of illegal activities or information on the platform, something that would remove their status as passive inter- mediaries 40 . Had the Court relied on L’Oreal, it would have likely deemed YouTube’s and 37 Case C-291/13 Papasavvas v O Fileleftheros Dimosia Etairia Ltd and Others ECLI:EU:C:2014:2209, paras 40-46. 38 HASSOUN, Amelia; ABONIZIO, Ariel; OSBORNKaty; WU, Cameron; GOLDBERG, Beth, “The Influencer Next Door: How Misinformation Creators Use GenAI”, available at https://arxiv.org/abs/2405.13554#: ~:- text=Based%20on%20longitudinal%20ethnographic%20research,their%20personal%20needs%20and%20 desires (last access 30.07.2024) demonstrate with impressive insightfulness and precision that Generative AI applications are passive tools in the hands of misinformation creators. 39 Joined Cases C-682/18 and C-683/18 YouTube and Cyando ECLI:EU:C:2021:503. 40 Joined Cases C-682/18 and C-683/18 YouTube and Cyando ECLI:EU:C:2021:503, para 114: “In that re- gard, the fact that the operator of an online content-sharing platform automatically indexes content uploaded to that platform, that that platform has a search function and that it recommends videos on the basis of users’ profiles or preferences is not a sufficient ground for the conclusion that that operator has ‘specific’ knowledge of illegal activities carried out on that platform or of illegal information stored on it”. Despite content storing and sharing platforms performing more functions than just storing user-generated content, the Court did not remove their intermediary status. Notably, the Court also ruled that these platforms do not perform a ‘commu- nication to the public’ under art. 3(1) of the InfoSoc Directive by merely making their platform available (paras 76-102). Although this was a separate issue, the Court acknowledged (para 108) that whether these platforms