Inteligência artificial: definição dinâmica, base axiológica e abordagem regulatória antropocêntrica baseada em riscos

Abstract

In light of constitutional civil methodology, this work aims to present a proposal for defining an Artificial Intelligence system that adjusts to technological advances and the effects of time, allowing the legal operator to glimpse the limits of norm improvement to the concrete case, especially when Artificial Intelligence applications threat to injury or injury human rights. Under the premise of presenting specialized regulations in the current national legal framework to guarantee the stakeholders of such technologies full protection when entitled to, demonstrating the importance of an anthropogenic regulatory approach based on risks and attentive to diversity of types of Artificial Intelligence. In the light of a systematic-unitary reading, without shying away from suggesting solid answers, the main risk matrices of Artificial Intelligence and the harmful potentialities generated by these systems are faced, illustrating and relating them to the ethical principles of Artificial Intelligence. It also strives to build a technical and coherent proposal that deals with new forms of damages caused by algorithms of Artificial Intelligence, embodied in the field of civil liability, facing interpretative divergences of the current legal apparatus and the truly innovative. Recognizing impossible to exhaust the subject, Artificial Intelligence and a myriad of related notions are comparatively analysed, in addition to classifying Artificial Intelligence models according to typology, attributes and levels of development and autonomy – focusing on the framework by the Organisation for Economic Co-operation and Development –, categorizing the agents of these systems as suppliers and operators. It highlights the main dilemmas in establishing regulatory frameworks in national and foreign experiences and the conflicts between concreteness and legal security, in addition to an axiological basis and recommendations for good practices and governance, creating enforcement mechanisms. A critical view of the bills that tried or intend to settle the Legal Framework of Artificial Intelligence in Brazil is presented, focusing on questions about the regime, objective or subjective, of accountability of the agents in the Artificial Intelligence value chain. The possibility and appropriateness of the incidence of theories of risk are verified, according to the state of the art, monitoring the development of end-to-end system and different levels of dangerousness generated by the insertion of new technologies in the informational society. It also considers the application of precautionary principle as a compass for research, development and use of Artificial Intelligence, given periodic algorithmic audits and regulatory sandboxes as some of the means of controlling inputs and outputs of data that operationalize the most diverse Artificial Intelligence models in the era of data-driven economy. Ultimately, it seeks to functionalize civil liability, with a view to preventing and repairing damage caused by intelligent computational projects, with the evaluation of algorithmic impacts from the design, attention to the duties of information, explanation and algorithmic transparency – among others – , confronted with business secrets and bearing in mind diversity as the guide for multisectoral governance and institutional arrangements.