Desafíos actuales de la Inteligencia Artificial

The interplay between AI act and GDPR 45 special categories of personal data to prevent biases, the extent to which the right to data protec- tion should be limited to ensure the most accurate output of the system remains unclear. 67 Under the EU AI Act, the concept of accuracy refers to AI systems ensuring their out- puts are free from bias and do not result in discriminatory outcomes. 68 Therefore, in high-risk AI applications, accuracy includes not only the system’s ability to generate correct results but also its compliance with non-discrimination and equality principles, ensuring that all individ- uals are treated regardless of their personal characteristics. Article 10 of the AI Act addresses the issue of bias in training data, but it is specifically aimed at developers of predictive AI systems in high-risk areas, not developers of generative AI. 69 If generative AI is developed for a high-risk area, developers are required to comply with these rules. However, the effectiveness of Article 10 in addressing generative discrim- ination will largely depend on how ‘bias’ is interpreted within the AI Act. 70 Currently, it is uncertain whether the harms discussed in this chapter would be classified under the AI Act’s concept of ‘bias.’ 71 Regulators might interpret ‘bias’ in a technical sense (regarding the di- versity of training data) rather than in a social and ethical sense (relating to demeaning and abusive content). 72 The Right to Explanation might not be useful for our purposes either. This provision focuses on explaining decisions made by predictive AI in high-risk areas like employment, criminal justice, or education. 73 Since the AI Act does not categorize generative AI as a high-risk application, the right to explanation generally does not apply to generative AI. 74 The AI Act does not include generally applicable provisions for developers of generative AI systems to mitigate bias. 75 The provisions concerning generative AI focus on transparency, 67 Ibid. 68 M. WINAU, ‘Areas of Tension in the Application of AI and Data Protection Law’, 130. 69 European Parliament legislative resolution of 13 March 2024 on the proposal for a regulation of the European Parliament and of the Council on laying down harmonised rules on Artificial Intelligence (Artificial Intelligence Act) and amending certain Union Legislative Acts (COM(2021)0206 – C9-0146/2021 – 2021/0106(COD)) (Ordinary legislative procedure: first reading).References to articles and recitals introduced or changed by the Parliament will be labelled in this paper as “the EU AI Act”. 70 P. HACKER, F. ZUIDERVEEN BORGESIUS, B. MITTELSTADT, and S. WACHTER, ‘Generative Dis- crimination: What Happens When Generative AI Exhibits Bias, and What Can Be Done About It’ (2024) p. 39. 71 HACKER, ZUIDERVEEN BORGESIUS, MITTELSTADT, and WACHTER, ‘Generative Discrimination’, p. 39. 72 Ibid. 73 Article 86 European Parliament legislative resolution of 13 March 2024 on the proposal for a regulation of the European Parliament and of the Council on laying down harmonised rules on Artificial Intelligence (Ar- tificial Intelligence Act) and amending certain Union Legislative Acts (COM(2021)0206 – C9-0146/2021 – 2021/0106(COD))(Ordinary legislative procedure: first reading).References to articles and recitals introduced or changed by the Parliament will be labelled in this paper as “the EU AI Act”. 74 N. HELBERGER and N. DIAKOPOULOS, ‘ChatGPT and the AI Act’ (2023) 12 Internet Policy Review. 75 HACKER, ZUIDERVEEN BORGESIUS, MITTELSTADT, and WACHTER, ‘Generative Discrimination’, p. 40.