The use of ChatGPT in indexing Access to Information requests on Fala.BR

This study aimed to analyze the indexing of requests for access to information from Federal Universities in Brazil with the aim of proposing improvements, through ChatGPT, on Fala.BR. The research was structured based on the concepts of access to information, classification and artificial intelligence for indexing Subjects, Sub-subjects and Tags in Fala.BR. This is an exploratory and explanatory research, of an applied nature, whose approach is qualitative-quantitative and the technical procedures used were bibliographic and experimental, using Fala.BR and Arquivistas users as a sample. Research instruments such as the controlled e-government vocabulary and the classification code for archival documents of the core activities of Federal Higher Education Institutions were used to generate prompts in ChatGPT. Then, the results were compared with the texts of archivists who used only the classification code and, even so, reached surprising numbers in the Subsubjects and Tags fields. To the point of inverting the values of these fields, which were negative, due to the absence of completion in the original information access requests, they became positive due to the degree of similarity of substantial agreement between man and machine.