Bridging the Data-Action Gap: Potential Opportunities and Risks for Generative AI Language Models in Communicating Data Insights to Poultry Farms

Precision livestock technologies enable continuous animal monitoring or earlier health and welfare interventions. However, if data cannot be easily interpreted and acted upon by humans (e.g., farmers), benefits may not occur in practice. This case study explores the potential for Large Language Models (LLMs) to overcome the data-action gap by analyzing big data and communicating clear extension instructions. We worked with a company which has developed an AI model that predicts water intake deviations on poultry farms and identifies the possible cause, using an LLM to communicate farm data insights. Through qualitative interviews with developers and poultry farmers, we critically evaluate the potential role for LLMs in agricultural extension, exploring user preferences, design possibilities, and social/ethical opportunities and risks. Interviews uncovered an interesting extension journey, from initial manual data entry and reporting, towards the development of an LLM to automate the process. LLM insights informed work prioritization, but application varied across the chain of command in the poultry farming business as senior executives oversee multiple farm sites, each with a farm manager, some with a farm worker. Risks of using LLMs in extension were generally downplayed. The potential for LLMs to be used in ways that perpetuate inequality and tradeoffs, such as human surveillance for transparent animal welfare, is explored. Our work makes a theoretical contribution to emergent social science literature that has started to explore the application of LLMs in on-farm agricultural extension (e.g., Tzachor et al., 2023; de Clercq et al., 2024; Marinoudi et al., 2024), critically evaluating the potential role of LLMs in agricultural extension. We will shed light on the opportunities and risks of LLMs, as well as explore the potential role for user-centred design. Practically, we identify tangible factors that can optimize the use of LLMs for on-farm agricultural extension.