The research evaluates a multistage approach to AI deployment, with each stage involving a more accurate method of inference. The method aims to maintain user trust by quantifying model uncertainty and making confident deferral decisions. The AI is currently being deployed to thousands of cotton farmers across India to aid in pest management decisions. The AI models identify and count bollworms from photos taken by farmers, and based on these counts, provide pesticide spraying advice from reputed agricultural bodies.

 

Publication date: 2 Feb 2024
Project Page: agri-ai@wadhwaniai.org
Paper: https://arxiv.org/pdf/2402.00015