Speaker
Description
Farmers’ plant disease management (PDM) decisions are prone to various types of (perceived) risks concerning the likelihood and impact of the disease and resulting yield loss. Additionally, the time lag between decision and outcome impedes a direct evaluation of the chosen management (1–3). Decision support tools (DST) help farmers make the optimal decision in uncertain and complex PDM situations, but adoption remains low due to farmers' risk perceptions (4,5). Green insurance options aim to mitigate this risk by compensating farmers for yield loss even if the recommendations were followed (2). Further, novel, artificial intelligence (AI)-based DST promise to provide improved recommendations that increase PDM efficiency. By analyzing complex data and using reinforcement learning, AI-DST has enhanced prediction abilities and can provide various information on the resulting recommendation (6).
However, it is unclear how novel AI-DST for PDM might influence farmers’ risk perceptions. Therefore, we aim to investigate how far the provision of information by AI-DST recommendations might help mitigate farmers’ risk perceptions and thereby foster their follow-through on these recommendations. Specifically, we aim to employ an adaptive experimental design with dynamic treatment allocation 7 using probabilistic programming (8). This means that in each round of the experiment, treatments are assigned based on the posterior probability that each treatment is optimal with the aim to identify the best among a set of possible treatments at the end of the experiment.
This approach allows combining different treatments concerning 1) the provision of recommendation information (e.g. uncertainty, error-cost ratio, complexity, loss insurance, various objectives, learning) and 2) the time aspect (i.e. outcome/ payment in short-, middle or long-term) and to detect the combination that maximizes farmers’ willingness to follow the recommendation. Concretely, we will be able to identify for which type of decision (short-, middle- or long-term) what kind of information provision maximizes the farmers’ willingness to follow the recommendation.
With our study, we aim to examine how the abilities of novel AI-DST can help mitigate farmers’ risk perceptions on AI-DST use in PDM and thereby foster farmers’ willingness to follow recommendations. Based on our findings, tech developers can improve tech design on which information is communicated to the farmer. Further, the results can inform the design of policies and insurance schemes. Methodologically, we demonstrate how probabilistic programming can complement experimental design, and we are, to our knowledge, among the first to employ adaptive treatment assignment in the agricultural decision-making context.
| Keywords | Risk, Decision Support, Artifical Intelligence, Crop Management, Adaptive Treatment Assignment |
|---|---|
| Status of your work | Experimental Design |
| Early Career Researcher Award | No, the paper is not eligible |
