Send message to Authors

Paper short abstract:

Farm data offers real-time insights into poultry health and welfare, but improvements rely on actions by farmers and farm staff. Using established tools and processes linking farm data to an AI engine, we explore the potential for language models to help farmers and workers act effectively on data.

Paper long abstract:

Farm data offers real-time insights into poultry health and welfare, but improvements rely on actions by farmers and farm staff. Poor accessibility of data has been noted as a barrier to action. To overcome this data-action gap and effectively manage socio-cyber-physical systems, various costly and time-consuming interventions have been proposed, e.g. improving the advisory system and opportunities to upskill the workforce. A potentially more efficient solution could involve language models, or natural language programming. This would use generative AI trained on farm data, ultimately ending with a visual, text, or audio management instruction to the farmer/worker alerting them to an issue or suggesting corrective action.  

This pilot research, working with a company that has established tools and processes for sending data it has captured to an AI language model, explores preferences for how data is presented. This work also explores the potential of language models to help farmers/workers act effectively on data – an area which has so far received little to no empirical on-farm evaluation. We also critically explore the social and ethical implications of using AI language models on farms. 

Interviews with poultry farm staff explore how they make decisions from data and technology and the challenges they face, if any; and their preferences and views about the potential for language models to overcome the data-action gap, as well as how they might affect the nature of work and human-animal relationships. These insights will improve understanding of the presentation of data insights in ways aligned to individual preferences.