Evaluating the capabilities of artificial intelligence for interpretation and retrieval of research data stored in a repository

Abstract The current format of data repositories like Plant Disease Management Reports (PDMRs) makes recovery of critical information such as treatment efficacy and pathogen control difficult. Artificial intelligence was employed to retrieve and evaluate data from the subset of reports related to turfgrass disease management. Artificial intelligence has the potential for application across a broad spectrum of report types and could enhance the efficiency of translating research data into management recommendations. A customized large language model with access to 1074 PDF files containing turfgrass PDMRs was asked a series of queries. The answers were recorded, and the accuracy of the given responses was verified. From these responses, we can conclude that while the queries were often answered in a thorough manner as specified in the chatbot's configuration, the accuracy and consistency of responses given was highly variable and dependent on which PDMRs the chatbot deemed relevant to the query. Further modifications and validation of chatbots would be required before effective utilization as an extension and research tool.