PREDICTIVE ARTIFICIAL INTELLIGENCE AND PUBLIC POLICIES: HUMAN AGENT BETWEEN PERSUASION AND BINDNESS
Abstract
New developments in artificial intelligence (AI) are announced every day, heating the debate about its usage by Public Administration, especially in the public policies realm. Administrative Law, guided by the efficiency paradigm according to the Brazilian constitution, is called to explore risks presented by the intersection between predictive AI and public policies. The article reports a research developed according to the exploratory method, and it aims to evaluate if the predictive findings presented through AI might be found biding in the public policies realm. Artificial intelligence is initially defined, considering the vagueness of such expression. New tendencies in AI are presented: data driven policies, generalization of narrow AI and a quest for intelligibility. The article proceeds to a critical part, in which hidden risks in all those new tendencies are presented, showing how they can deviate or impoverish deliberation in public policies. As a result, the article concludes that predictions delivered by AI should be taken as a non-decisive component in formulating public policies. Human agent deliberation in those domains is mandatory as means to control correction and fairness in AI predictions. That “human approval seal” is also relevant considering the constitutional choice for a Democratic State of Law.
