Os grandes modelos de linguagem e os vieses de raciocínio lógico

Abstract

The definitions of Artificial Intelligence (AI) include characterizing algorithms as those that: think like humans, think rationally, act like humans, and act rationally. Logic, as a formal framework, enables the creation of algorithms capable of thinking rationally by expressing realworld situations in a language that allows for valid and rigorous reasoning. On the other hand, Large Language Models, such as ChatGPT, Gemini, Copilot, and the latest model, DeepSeek, represent algorithms that act like humans, especially in tasks involving natural language understanding and generation. However, these models can exhibit logical reasoning biases, which are tendencies that impair the ability to reason logically. This research project proposes to identify and analyze the logical reasoning biases committed by ChatGPT, Gemini, Copilot, and DeepSeek, in comparison with those made by undergraduate students in Information Technology-related courses who are beginners in Logic studies.