Desafíos actuales de la Inteligencia Artificial

92 Desafíos actuales de la Inteligencia Artificial widely, with some commentators believing the DSA’s scope is too narrow to cover Generative AI tools, while others explore more flexible models for their inclusion. However, a common theme among commentators is that Generative AI applications raise significant content moderation issues, enabling average users to create potentially illegal, harmful, malicious, or misleading content. Another point of agreement is that AI-generated content can be regulated by the DSA at a secondary level when it appears on traditional con- tent-sharing platforms. There is also consensus that incorporating Generative AI elements into traditional content storage and sharing applications should not be prima facie excluded from the DSA’s scope. Nevertheless, the various approaches regarding the applicability of the DSA to Genera- tive AI applications merit individual assessment. At one end of the spectrum, the opinion that dismisses the applicability of the DSA to Generative AI applications appears reasonable. It is fair to say that Generative AI was not the primary regulatory target of the DSA, posing challenges in finding the appropriate regulatory framework to address the risks associated with their content-related capabilities. However, this opinion may not be entirely convincing for several reasons. First, even if one accepts the rather reasonable assumption that the DSA did not pri- marily target Generative AI applications, this is not a decisive argument for their prima facie exclusion from the scope of the DSA. The realm of emerging technologies, which the DSA primarily regulates, is fast-paced and constantly evolving. This structural characteristic of digital environments justifies a flexible, creative, and dynamic interpretative approach that extends beyond the strict confines of a grammatical interpretation of the DSA’s provisions 35 . This is especially true for those referring to intermediary services that might benefit from liability immunities but also face accountability obligations regarding content moderation 36 . The DSA includes mechanisms that support such a flexible approach. Recital 28 of the DSA acknowledges the evolving nature of digital applications, while Recital 29 further suggests that the classification of a specific service as a “mere conduit”, “caching”, or “hosting” service should depend solely on its technical functionalities, which may evolve over time and should be assessed on a case-by-case basis. In this context, it might not be unreasonable to propose that only after thoroughly exploring the outer teleological limits of the DSA in relation to Generative AI applications can a conclusion be drawn regarding their inclusion or exclusion from the regulatory framework of the DSA. Second, the argument that Generative AI applications do not fit the notion of “host- ing” providers because they produce content rather than merely storing it, is debatable and requires nuance. While it is true that Generative AI applications play a role in content pro- 35 In the same spirit, Stalla-Bourdillon, supra note 28. 36 For a more general argument on the flexibility of the intermediary phenotypes of the DSA see REVOLIDIS, Ioannis: “Internet Intermediaries and Copyright Enforcement in the EU: In Search of a Balanced Approach”, in: Corrales, Marcelo; Fenwick, Mark; Forgó, Nikolaus (eds.): New Technology, Big Data and the Law, Singapo- re: Springer Nature, 2017, pp. 223-246, especially p. 229-232.