Information retrieval on the Web with Bard:

an experiment with Google's generative artificial intelligence

Abstract

Bard is Google's new Artificial Intelligence (AI) service, launched in Brazil in July 2023. Like other AI models, the proposal resembles a chatbot that interacts with users in natural language to retrieve information on the web and in private and dedicated databases. This experience report is part of a larger project that aims to analyze information retrieval on the Web with AI tools. To this end, three criteria were initially created for selecting a domain: (1) the authors' prior knowledge of the subject; (2) a global context; and (3) results that could return textual information and other digital objects. After defining the domain, which selected the open data theme, ten search expressions were created, in the form of questions, associated with the typology and need for information. Four classifications were created to analyze the answers: totally pertinent, pertinent, partially pertinent, and incorrect. The questions were run on Bard in July 2023 and the three drafts offered by the tool were analyzed. Most of the answers obtained were partially pertinent (40%) or pertinent (33.3%), two answers were incorrect (6.6%) and six answers (20%) were totally pertinent, as they met the need for information and had the expected typology. Bard proved to be an interesting tool for information retrieval, but it still needs improvement, especially about the source, which is not informed, and the accuracy of the results, which could be optimized.