Interpretação jurídica e Inteligência Artificial: raciocínio e linguagem entre o paradigma da racionalidade e inferência e o paradigma da imaginação

Abstract

The thesis proposes the hypothesis of an inseparable connection between legal interpretative reasoning and language and, based on this hypothesis, examines the possibility of artificial systems performing legal interpretation. These ideas are investigated through qualitative exploratory research, conducted by integrating philosophy of language, natural language processing, and theories of legal interpretation, organized into two opposing theoretical models: the Paradigm of Rationality and Inference and the Paradigm of Imagination. This theoretical foundation demonstrates that, from their conception, artificial systems must follow a specific theoretical perspective to process language and formulate heuristic rules to artificially replicate reasoning. Furthermore, it highlights that this perspective will have direct consequences on the results generated by these systems and on the concept of language comprehension. From this perspective, legal artificial intelligence systems and their developments in analytical, deductive, or analogical reasoning models are analyzed, identifying vulnerabilities related to how artificial intelligence processes language. These vulnerabilities result in different forms of opacity, biases, projection issues, the need for linking explicit and implicit knowledge, flaws in relevance selection, difficulties in performing deductive, inductive, and abductive inferences, challenges in handling common-sense knowledge, problems of overfitting and underfitting, and complexities in safeguarding different spheres of the principle of transparency. These challenges can affect procedural constitutional guarantees, such as due process, adversarial proceedings, judicial impartiality, and the reasoning behind judicial decisions.