Abrindo a “caixa preta” da inteligência artificial para revelar as faces ocultas das decisões judiciais

Resumo

In this note, we present an innovative research methodology proposed by Ludwig and Mullainathan (2024), based on hybrid artificial intelligence, which combines machine learning algorithms with human evaluation to generate new scientific hypotheses. The approach is illustrated through the analysis of judicial decisions on pretrial detention, revealing how defendants’ facial features can influence these decisions, even after controlling for traditional variables. We also highlight the use of advanced techniques, such as convolutional neural networks and generative adversarial models, for the creation of manipulated images, enabling the identification of relevant visual patterns. Finally, we discuss recent advances in these technologies and the epistemological risks associated with the use of AI in scientific research.

Ficha do documento

Tipo

Estudo

Ano

2025

Instituição

Instituto de Pesquisa Econômica Aplicada (Ipea)

Fonte

Repositório do Ipea

Idioma

Português

Acesso

Acesso aberto

DOI

10.38116/ntdiest66-port

Identificador

oai:repositorio.ipea.gov.br:11058/17001

Licença

Licença Comum

Temas

Tecnologia

Palavras-chave

Inteligência ArtificialInteligência artificial híbridaAprendizado de máquinaDecisões judiciaisRiscos da inteligência artificialGeração de novas hipóteses científicas