Desafíos actuales de la Inteligencia Artificial

A review of high-risk artificial intelligence (AI) systems that assess social security eligibility 251 cause it promotes a healthy society, without condemning to fate individuals who, for whatever reason – illness, old age, death, pregnancy – are unable to carry out an activity that guaran- tees their own subsistence and that of their family. Although still in its infancy, Wagner Balera (2010) points out that the Act for the Relief of the Poor , passed in England in 1601, was the first model of a permanent social protection programme, in which the state would be responsible for providing assistance to the neediest in situations of illness, disability and unemployment. After a slow and progressive evolution, the Treaty of Versailles, which sealed the peace at the end of the First World War, created the world’s first specialised social relations body, the International Labour Organisation (ILO), in 1919. The treaty stated that “[...] the building of peace should be based on the cornerstone of social justice” (Balera, 2010, p. 72), high- lighting the need for a social security programme, which was later addressed in conventions and recommendations. ILO Convention 102/1952 established minimum standards for social security, contain- ing detailed and complete provisions for the payment of specific benefits to cover certain social risks defined therein. From then on, the right to social protection, rooted in the concept of human dignity, became part of humanity’s legal heritage The fundamental right to social security is recognised by the Charter of Fundamental Rights of the European Union (Article 34, “the Charter”), and Regulation 883/2004 and its implementing Regulation 987/2009 are the main documents in the European Union dealing with social security. However, while on the one hand there has been regulatory progress and a real improve- ment in social protection formulas, on the other hand this new cyber-law has not yet been fully implemented. The basic premises of protection and the minimum social risks to be covered by the state were established; in other words, awareness of the indispensability of social protection was established globally. However, while the introduction of AI systems in this field can extend the range of protection, so that new subjects and new needs can be met in a timely manner, it can also exclude people who, despite their need for social protection, are unable to satisfy the virtualised process. People who, it should be emphasised, are in a moment of vulnerability and are looking for support from the state, and who may encounter technological obstacles to the realisation of their right, as will be demonstrated. The recently published European regulation on AI is based on the differentiation of risk levels of AI systems and imposes interventions proportional to them. Thus, the regulation moves from an absolute ban on systems whose risks are unacceptable, to strict regulation of high-risk AI systems, and finally to almost deregulation of AI systems considered low or no risk, for which only transparency requirements and some incentives to adopt codes of con- duct are established. Thus, by prohibiting the use of unacceptable AI systems, mitigating the risks of applica- tions of high concern and adopting a laissez-faire approach in areas of low concern, European