Experts Tell House Subcommittee AI Is Accelerating Agrochemical Discovery, But Data, Funding And Regulatory Bottlenecks Remain

At a hearing of the House Science, Space, and Technology Subcommittee on the Environment titled "Innovations in Agrochemicals, AI's Hidden Formula for Driving Efficiency," witnesses from industry and land‑grant universities described concrete ways artificial intelligence is changing discovery, manufacturing and field application of pesticides and fungicides.

The panel heard that AI can compress discovery timelines, help manufacturers optimize production and generate site‑specific treatment maps that could sharply reduce chemical use. Committee members and witnesses urged continued federal support for public datasets and for research partnerships so AI outputs have reliable inputs and can be validated before field deployment.

Brian Lutz, vice president of agricultural solutions at Corteva Agriscience, told the subcommittee that AI is changing discovery by allowing researchers to move from randomness to prediction and design. "AI has revolutionized discovery by allowing us to trade randomness and chance for prediction, specificity, and design," Lutz said, describing work in which models analyzed roughly 10,000 molecules in weeks and flagged dozens of candidates his team would not have found otherwise. He described three distinct AI roles: discovery, development and manufacturing, and field guidance. Lutz said Corteva invests "nearly $4,000,000 every day" in research and development and cited a manufacturer site in Midland, Michigan, as an example of using AI to optimize fermentation and reaction conditions.

Daniel Swale, associate professor at the University of Florida, testified that AI is opening access to natural‑product chemistry and can help universities re‑enter early‑stage discovery that historically has been dominated by private companies. "The outputs of AI driven discovery are only as good as the inputs you provide the system," Swale said, urging public investment to create and share foundational datasets that currently are often locked inside private firms. He also described efforts to find alternatives for pests such as the Asian citrus psyllid and noted that synthetic compounds his lab has identified face a prolonged and costly regulatory pipeline.

Boris Cameletti, assistant professor at the University of Illinois Urbana‑Champaign, described using satellite and spectral remote sensing with machine learning to detect Red Crown Rot in soybeans, a soilborne disease that can cause "yield losses of up to 50%." His team trains models with visible to near‑infrared spectral bands and ground truth observations to generate prescription maps that target only affected patches. "If only 25% of a field is deceased, our approach could reduce fungicide yield by up to 75%," Cameletti said, adding that on‑farm trials are planned to compare traditional uniform applications with AI‑guided site‑specific treatments.

Members pressed witnesses on safety, validation and the pace of regulatory review. Several members reiterated a recurring theme in the hearing: AI models require robust public data. Chairman Franklin told the panel he had secured $4,500,000 in federal funding for an AI center at the University of Florida and noted EPA's reported backlog of "over 504 new chemical reviews and 12,000 pesticide reviews" under the Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA) and review timelines under the Toxic Substances Control Act (TSCA). Members from both parties asked how to speed needed products to growers without sacrificing scientific review.

Witnesses described concrete benefits and limits. Lutz said a fungicide‑timing model combining field‑specific information with internal data has helped some growers increase yields by "4 to 10 bushels per acre" in pilot work. Swale argued that natural‑product pathways can shorten registration timelines for some compounds compared with synthetics and highlighted a Boulder, Colorado, company (Invata) that his team is working with to identify structures in complex natural mixtures. Cameletti emphasized that land‑grant research, checkoff funding and public computing infrastructure enabled the work at Illinois, and urged continued public support so tools serve the public good.

All three witnesses acknowledged technical and governance risks. Lutz and others urged model validation, human‑in‑the‑loop testing and regulatory clarity to prevent misuse or overreliance on unvalidated outputs. On risks and regulation, Lutz warned against a patchwork of state rules and called for federal clarity on oversight and testing frameworks.

The hearing record will remain open for written questions and comments; members may use those submissions to press agencies on data access, funding priorities and options to accelerate registration while preserving safety reviews.