Algorithmic transparency - a regulatory or a technological problem?

Abstract

As technology evolves, it will become increasingly frequent for decisions such as personnel selection, credit granting or partner selection, to be adopted by machine learning algorithms. Many of these decisions may contain biases that result in discriminatory decisions that affect the subject on whom the decision falls. In order to reveal how such decisions have been made, the idea of transparency in terms of algorithms has been used, implementing it through regulations. Such a normative or regulatory approach ignores a basic premise: transparency is not only a legal problem, but also a technological problem, which becomes relevant when we talk about machine learning algorithms. Machine learning algorithms are the basis for the development of a more sophisticated technology in automating processes than the classic programming approach, since it involves access to more powerful and flexible rules, which are automatically adjusted to the environment as the sample data evolve and recognize variations depending on many factors. However, the degree of sophistication of these algorithms leads to a complexity of their models, which generates a difficulty in their interpretation. The regulatory approach does not take into account the complexity mentioned above, what generates an insufficient and rigid regulatory framework. This article proposes to exceed this vision through flexible proposals, understanding the role of technology, recognizing its value and stimulating its use, this means proposals that promote and strengthen innovation in Chile.