Deciding Isn't the Same as Choosing: From Decision Automation to Human Review

DA AUTOMAÇÃO DA DECISÃO À REVISÃO HUMANA

Abstract

The growing implementation of artificial intelligence in the legal field, especially in decision-making, has raised fundamental concerns about the preservation of fundamental rights and the complexity of applying artificial intelligence in the decision-making process. The transition from human decisions to algorithmic decisions presents an intrinsic challenge: the inherent discretion of judges and the subjectivity involved in their work. Humans have the ability to make decisions influenced by their creativity, sensitivity, sensory perception, and, most importantly, their human experience, which goes beyond the mechanical application of rules and syllogisms. However, machines, no matter how sophisticated they may be, tend to simplify perceptual and rational information into numerical data. The "mathematization" of factual and jurisdictional aspects limits the judge's ability to consider nuances and specific circumstances that may be crucial to resolving cases, as the advent of artificial intelligence systems raises the possibility of replacing the human factor with robotic decision-making. Machines issue decisions based on programmed statistical variables, without the flexibility to adapt to individual circumstances, and despite promises of impartiality, these systems are still susceptible to cognitive biases, prejudices, and inclinations present in their initial programming, which can result in significant errors because the information is collected by humans and replicated by the system, potentially amplifying bias. All of these issues raise the critical question: Is it possible to transfer the administration of justice to a machine? As a potential solution to the problem, human review emerges as a concrete human intervention in the automation of the judiciary, a primary safeguard for fundamental rights, and the only realistic alternative to the adoption of technologies. The method used is analytical-descriptive, involving bibliographic research with an exploratory bias, aimed at understanding and discussing the state of the art related to the proposed topic.