Revisión sistemática de taxonomías de riesgos asociados a la Inteligencia Artificial

This article performs a systematic review of thirty-six taxonomies of risks associated with Artificial Intelligence (AI) that have been conducted from 2010 to date, using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol as a methodology. The study is based on the importance of these to structure risk research and to distinguish and define threats. This makes it possible to identify the issues that are of greatest concern and therefore require better governance. Three conclusions can be drawn from the research. First, it is observed that most studies focus on threats such as privacy and disinformation, possibly due to their concreteness and existing empirical evidence. In contrast, threats such as cyberattacks and the development of strategic technologies are less cited, despite their increasing relevance. Second, we find that articles focused on the origin of risk tend to consider more frequently extreme risks compared to papers addressing consequences. This suggests that the literature has been able to identify the potential causes of a catastrophe, but not the concrete ways in which it may materialize in practice. Finally, there is some division between those articles that deal with present tangible damage and those that cover potential future damage. Nevertheless, several threats are addressed in the majority of articles across the spectrum indicating that there are commonalities between clusters.