Desafíos actuales de la Inteligencia Artificial

A review of high-risk artificial intelligence (AI) systems that assess social security eligibility 257 Although the result for Sweden is optimistic, it is necessary to consider the context in which the Swedish government and recipients of social protection operate in order to make comparisons. According to the latest Digital Economy and Society Index published by the European Commission, which has been monitoring member states in the digital field since 2014 and analysing their level of digitalisation and progress in this area, Sweden ranks fourth out of 27 member states and continues to excel in connectivity and the use of digital tech- nologies (European Commission, 2024). “Digital technologies, both existing and emerging, are more widely used and integrated in the developed countries of the EU, which has a direct impact on the results obtained through the application of AI systems in the field of social protection” (European Commission, 2024). However, the reality in Brazil is quite different. According to the United Nations Devel- opment Programme (UNDP) Human Development Report 2023-2024, the country ranks 89th out of 193 nations, indicating significant digital inequality (UNDP, 2024). According to the IBGE, 25 per cent of the Brazilian population does not have access to the Internet, which equates to around 54 million people (IBGE, 2024). Despite this, the body responsible for social protection benefits and services, the INSS, has automated its services. Through its platform, access to social services and benefits from the Brazilian government. As a result, social protection is now only available via the internet and the vast majority of benefits are automatically analysed for eligibility (INSS, 2024). Against this backdrop, the agency itself has demonstrated the increasing rejection of applications. While in 2018 there were 3,889,600 rejections for 5,123,777 grants, in 2022, the last year recorded in the report, there were 5,113,354 rejections for 5,212,631 grants (INSS, 2024). This is highly problematic given the need to care for the vulnerable and meet the specific needs of the most marginalised and excluded groups in an unequal society, such as low-income earners, the elderly, the disabled and the sick. Australia also has a negative experience. The country has an agency (Centrelink) that is central to the delivery of government services and benefits to the Australian community (Ser- vices Australia, 2024). In detail, the agency is responsible not only for the social protection of pre-defined vulnerable groups (unemployment, old age, housing problems due to natural dis- asters or domestic violence, alimony), but also for the recovery of allegedly wrongly received payments (Services Australia, 2024). In the Australian welfare system, social security entitlements are based on individual in- come, so recipients declare their income to the government every two weeks, and the govern- ment uses this information to calculate their eligibility and the amount of payment they are entitled to receive. Within the agency, Robodebt has been developed, an automated system to check that recipients have correctly reported their income (Services Australia, 2024). Australia’s automated system has proved problematic for incorrect income and social security payments. “The system overlooked a critical issue: averaged income is a poor guide to actual income in any particular fortnight. Averaging obscures variations in a person’s actual earnings from fortnight to fortnight, which are essential for an accurate assessment of their entitlements.