Closed-loop artificial intelligence agents for animal epidemic prediction and decision support in livestock farming: a review

Abstract

Animal diseases continuously threaten livestock production and public health, yet current surveillance approaches remain largely passive and fragmented. Conventional artificial intelligence models typically function as open-loop predictors, delivering static outputs that fail to accommodate the dynamic operational needs of veterinarians in real-world farm settings. This review explores an emerging paradigm of AI agents for animal disease monitoring, emphasizing how such agents can transition from passive observation to active intervention through a closed-loop perception–decision–feedback architecture. We first classify multi-scale sensing technologies: at the micro-scale, automated molecular diagnostics and biosensors; at the macro-scale, computer-vision-based phenotyping. These technologies collectively provide the data streams required for continuous surveillance. The review then discusses how multi-modal inputs, time-series forecasting, resource-aware decision-making, and reinforcement learning can be integrated into an agent-based workflow. Importantly, most current veterinary research evidence supports only the individual components of this framework, rather than the deployment of fully autonomous, iteratively evolving closed-loop agents in actual field environments. Translating closed-loop AI from a conceptual framework into a reliable veterinary decision-support tool will therefore require further advances in validation, standardization, governance, and human oversight.