| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 6.21 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A tecnologia de InteligĂȘncia Artificial Generativa (IAG), que inclui modelos de linguagem e modelos de geração de imagens, tem sido amplamente utilizada em vĂĄrios setores, desde a produção de conteĂșdo digital atĂ© o auxĂlio na tomada de decisĂ”es.
Contudo, com o avanço dessas tecnologias, surgem preocupaçÔes em relação Ă perpetuação deâbiasâque podem ter impactos negativos na sociedade ao fortalecer estereĂłtipos ou gerar desequilĂbrios.
Na presente dissertação, Ă© feita uma investigação sobre as origens e representaçÔes deâgender biasâem sistemas de IAG, examinando como esses modelos sĂŁo treinados com grandes conjuntos de dados e como a representação desproporcional dos gĂ©neros nos dados de treino afeta os resultados. O estudo investiga as repercussĂ”es concretas deâgender biasâem situaçÔes comuns, como ferramentas de design, e modelos de linguagem. AlĂ©m disso, sĂŁo abordadas tambĂ©m as implicaçÔes Ă©ticas e sociais deste tema.
A fim de reduzir esses biases, sĂŁo sugeridas soluçÔes tĂ©cnicas, como a seleção cuidadosa dos dados, a implementação de mĂ©todos justos de aprendizagem e a monitorização constante dos modelos. Para alĂ©m disso, a presente dissertação analisa polĂticas e normas que podem contribuir para o desenvolvimento de uma IAG mais igualitĂĄria e justa.
Generative Artificial Intelligence (GenAI) technology, including large language models and image generation models, has been widely used in various sectors, from digital content production to decision-making. However, with the advance of these technologies, concerns have arisen about the perpetuation of biases that can have negative impacts on society by strengthening stereotypes or generating imbalances. This thesis investigates the origins and representations of gender bias in GenAI systems, examining how these models are trained with large data sets and how the disproportionate representation of genders in the training data affects the results. The study investigates the concrete repercussions of gender bias in common situations, such as design tools, and language models. In addition, the ethical and social implications of this issue are also addressed. To reduce these biases, technical solutions are suggested, such as the careful selection of data, the implementation of fair learning methods and the constant monitoring of models. The present dissertation also analyses policies and norms that can contribute to the development of a more equal and fairer GenAI.
Generative Artificial Intelligence (GenAI) technology, including large language models and image generation models, has been widely used in various sectors, from digital content production to decision-making. However, with the advance of these technologies, concerns have arisen about the perpetuation of biases that can have negative impacts on society by strengthening stereotypes or generating imbalances. This thesis investigates the origins and representations of gender bias in GenAI systems, examining how these models are trained with large data sets and how the disproportionate representation of genders in the training data affects the results. The study investigates the concrete repercussions of gender bias in common situations, such as design tools, and language models. In addition, the ethical and social implications of this issue are also addressed. To reduce these biases, technical solutions are suggested, such as the careful selection of data, the implementation of fair learning methods and the constant monitoring of models. The present dissertation also analyses policies and norms that can contribute to the development of a more equal and fairer GenAI.
Descrição
Palavras-chave
Igualdade de GĂ©nero InteligĂȘncia Artificial InteligĂȘncia Artificial Generativa Direito da UniĂŁo Europeia Direito Internacional Direitos Fundamentais Carreira Profissional Educação
