Audio description and artificial intelligence

for scientific accessibility

Abstract

Objective: To analyze the use of Artificial Intelligence, specifically generative language models, in the production of image audio descriptions for academic texts, considering the principles of information accessibility. Methodology: This is an descriptive study with a qualitative approach. Images commonly used in scholarly communication were selected, and the audio descriptions generated by Artificial Intelligence were evaluated according to the audio description criteria proposed by the National School of Public Administration (ENAP), considering aspects such as objectivity, informational relevance, semiotic organization, language adequacy, and communicative mediation. Results: The findings demonstrate that Artificial Intelligence has significant potential as a support tool for producing audio descriptions, particularly in identifying visual elements, spatial organization, and the description of colors and details. However, the comparative analysis revealed limitations related to excessive information, the presence of implicit judgments, and the absence of accessibility markers, highlighting weaknesses that may compromise the quality of information mediation. Conclusion: It is concluded that the ethical and critical use of Artificial Intelligence can contribute to strengthening inclusive practices in libraries, repositories, and academic platforms. However, its use should be guided by accessibility standards and supported by qualified human mediation to ensure more accurate, inclusive, and information-justice-oriented descriptions.