The use of artificial intelligence in tax auditaction and algorithmic opacity: the case of the Customs Selection System Through Machine Learning – SISAM

This article brings reflections on the topic of algorithmic opacity in artificial intelligence systems - AI used by the Brazilian Tax Administration through the analysis of the specific case of the Customs Selection System through Machine Learning – SISAM, in use by the Brazilian Federal Revenue Office since 2014. The central topics discussed are the concept of Tax Administration, its relationship with the Principle of Administrative Efficiency and the massive application of Tax Law; the basic concepts of AI and the functioning of SISAM; the problem of algorithmic opacity, on the part of the Tax Administration, in face of the Brazilian constitutional principles regarding Administrative Transparency and, finally, the verification, in the case chosen for analysis, of the respect for constitutional guarantees regarding the aforementioned Transparency, essential in the scope of the Democratic Rule of Law. For this purpose, the deductive method is used, with an applied, qualitative and exploratory research, having as a corpus the bibliographic and documentary research, with a case study. The hypothesis of this article is that the lack of absolute transparency in the functioning of the SISAM algorithms does not, a priori, offend the taxpayers’ guarantees, due to the need to protect the confidentiality of tax audit action criteria