O impacto da inteligência artificial generativa na produção de notícias falsas com vieses políticos

Abstract

The application of natural language processing models based on deep neural networks with attention mechanisms has been an area of significant advancements in the last decade within the field of Artificial Intelligence. These models, such as LLaMa, have the ability to generate coherent and relevant text, making them useful for various tasks, from machine translation to content generation. Inspired by these advancements, this work proposes to demonstrate the potential for creation of politically biased fake news using a large-scale language model, LLaMa. The proposed model takes as input a news body and the desired political bias to generate a politically biased headline on the subject. The application of sentiment analysis methods to the generated headlines will be investigated in an attempt to relate biases toward certain political profiles with human sentiments. It is expected that this work will provoke reflections on the ethical implications and responsibility in content generation, as well as shed light on the possible ways in which various political actors globally conduct automated propaganda.