Implications of algorithmic fairness in artificial intelligence

Abstract

The use of Artificial Intelligence algorithms is not limited, as is sometimes assumed, to effective procedures; the use of this vocabulary raises several conceptions, interpretations and problems. In order not to get distracted in this linguistic labyrinth, we have taken a position on a very common characterization in the psychological sense that consists of conceiving it as a capacity possessed by certain organisms/mechanisms to adapt to new situations using for this purpose the knowledge acquired in the course. . from previous adaptation processes. The emergence of artificial intelligence (AI) is increasingly integrated into society and is generally used to make timely decisions that affect society and therefore people, in different areas. In the development of AI algorithms, systemic and repeatable errors can occur in a computer system that create unfair results, such as privileging an arbitrary group of users over others. These so-called biased algorithms are generally characterized by the existence of biases or distortions in the training data. The scientific community and government institutions have launched proposals to combat these risks that seek to reduce their negative impact on society.  It is imperative to resolve aspects that go against Ethics, justice, Transparency and Equity of data, algorithms and their predictions.  The present work aims to raise awareness that the development of algorithms to make decisions must meet three requirements: first, guarantee the balance between the set of data used and the programming of the algorithm with fairness that avoids discrimination and bias, second, guarantee conditions of transparency in the results, that is, the result obtained must be explainable to any user in a clear and simple way. The regulation of requirements for the development and use of AI should not be ignored, it must be aligned with the non-affecting of fundamental human rights