Harnessing AI To Feed The Future

How public research, tech companies and land-grant universities are pairing big data with biological science to deliver reliable decision support to farmers.

With artificial intelligence adoption rising on farms in the United States and beyond, the AgTech360 podcast is sharing insights from four leading AI-in-agriculture experts about what’s needed to ensure the technology can strengthen farmers’ bottom lines and their ability to feed a fast-growing world.

In a new four-part series, AgTech360 host and North Carolina Plant Sciences Initiative Executive Director Adrian Percy interviews leaders across government, academia and industry: Steven Mirsky of the U.S. Department of Agriculture’s Agricultural Research Service, Microsoft’s Ranveer Chandra, North Carolina State University faculty member Debjani Sihi and seasoned technology chief executive officer and board member Mohan Tavorath.

“AI is not a silver bullet that can replace human expertise, but … it can be a powerful decision tool.”

“Facing extreme weather, rising costs, and a shrinking farm workforce, growers increasingly rely on AI to plan and manage their operations,” Percy said. “AI is not a silver bullet that can replace human expertise, but when it’s farmer-centered and grounded in interdisciplinary science, collaboration and structured data, it can be a powerful decision tool that delivers practical value to growers and helps meet the growing global demand for food, fiber and fuel.”

Building Digital Infrastructure
"Cameras act as new eyes in the field. They allow continuous monitoring and turn observations into immediate decisions."

Mirsky, the USDA-ARS’ director of digital agriculture, is addressing a major barrier to agricultural AI: the lack of high quality, standardized training data and camera systems for deployment.

This is particularly hard in agriculture, he says, because farming involves constantly changing soils, microclimates and biological systems.

Through the Digital Agricultural Systems Hub (DASH), Mirsky’s team focuses on delivering what he calls “end-to-end solutions,” from discovery to deployment, of computer vision systems for agriculture. The project involves training machines to detect plants, know what kind of plants they are, recognize whether they’re under stress and, if so, where that stress is coming from – whether it’s a disease, insect, drought, heat or other problem.

The team’s efforts include deploying rugged, modular camera systems that can capture images of plants for a massive open national database of images, known as the AgIR, that is hosted on USDA supercomputers. The goal is to power more precise and automated agricultural production.

“Cameras act as new eyes in the field,” Mirsky explained. “They allow continuous monitoring and turn observations into immediate decisions.”

Automated camera systems like the ModCam system the DASH team uses allow a single worker to monitor vast fields, detecting pests, weeds and diseases in real time to enable quick action.

“DASH is here to enterprise computer vision and AI across American agriculture and deliver real-world tools to our farmers, breeders and researchers faster,” Mirsky said. “We have a lot of folks developing different computer vision solutions, and everybody’s collecting data in different formats. They’re using different camera systems, and so we’re trying to curate a way that we could create a lot of gravity around standards for a plant agricultural image repository and make that publicly available.”

Making Insights Accessible Anywhere
“Any farmer – anyone anywhere on the planet in the agriculture industry – should benefit from AI.”

While Mirsky discussed pioneering efforts to enhance computer vision for agriculture, Ranveer Chandra emphasized the importance of connecting separate software systems to make reliable predictions from data.

Generative AI, the Microsoft vice president said, is making data far easier to use and making predictions “just a chat away.”

“You just ask a question, and you could get an answer with all these GPUs, with cloud computing and everything else in the background delivering you that insight,” he explained.

However, applying AI to agriculture presents unique hurdles, particularly severe data sparsity in remote rural regions and the operational cost of cloud computing. To overcome connectivity and other constraints, research is shifting toward smaller, fine-tuned agricultural models operating on edge-computing hardware directly in the field.

Tailored systems can deliver rapid, low-cost reasoning by combining small language models built for speed, efficiency and targeted performance and training AI models by using feedback loops based on specialized, real-world agricultural knowledge rather than generic internet data, he added.

As Chandra observes, “Any farmer – anyone anywhere on the planet in the agriculture industry – should benefit from AI,” driving higher profitability, reduced chemical pesticides and synthetic fertilizers and greater environmental sustainability.

Bridging the Data Divide for Actionable Decision-Making
"AI models will only be trusted by the grower if they are correct."

Mohan Tavorath, who has led and overseen technology ventures and data analytics platforms in agriculture, financial services and other industries, points out that agricultural data sets are highly fragmented across various stakeholder groups. Translating raw agricultural data into meaningful operational decisions therefore remains one of the greatest challenges in digital agriculture.

Unlike centralized industrial supply chains, agricultural data originates across millions of independent farms and is scattered across many categories that include soil health, genetics, farm machinery, irrigation, disease and pest management, crop protection inputs, harvest estimation, yield, climate, storage and transportation logistics.

“While AI is helping connect the dots between all these different data sets, one of the biggest challenges is getting those data sets structured, cleansed and consistent so that you can analyze them well and provide insights back to the grower,” Tavorath explained.

The goal, he noted, is to accelerate growers’ understanding of field conditions so they do not have to act as their own systems integrators.

Tavorath stresses that for AI tools to earn more widespread grower adoption, data platforms must initially deliver narrower solutions that are highly accurate and consistent.

“AI models will only be trusted by the grower if they are correct,” Tavorath observed. “It cannot be a beta product that’s being tested on the grower. The grower is so focused on their production operation that they don’t have the time or patience for data companies that oversell and underdeliver.”

Tavorath added that adoption also accelerates when technology fits seamlessly into existing farm workflows without requiring growers to alter their routines. He thinks multi-modal input will become increasingly important, with growers supplying data by voice and video as they work in the field. As AI models grow more sophisticated, they will better translate that input into semantic structures that map to agronomic, operational and crop models, returning actionable insights to the grower.

Looking ahead, Tavorath sees the future of digital farming evolving to support “co-intelligence,” a seamless partnership between human operators with deep domain expertise and autonomous AI systems working together to optimize regenerative agriculture and sustainably feed the world.

Unpacking the “Black Box” with Disciplinary Science
“We need disciplinary scientists who … have the knowledge of the system, … software scientists who can give it life and … practitioners … . And we need to work together.”

Even as AI platforms become easier to use, Debjani Sihi says agricultural ones must remain rigorously anchored in physical and biological realities. This requires grounding in subject-matter expertise reflecting how agricultural systems actually function on the ground, across space and time.

Debjani Sihi is a biogeochemist and an NC State assistant professor working on applied AI in plant and microbial biology and crop and soil sciences. She stresses that machine learning algorithms must not operate as unexplainable “black boxes” detached from biological facts.

Her research integrates advanced machine learning with classical process-based mechanistic models to better understand and predict how agricultural systems respond to management practices and changing environmental conditions.

While these models use mathematical equations to simulate physical and biological laws, pairing them with machine learning allows scientists to generate rapid predictions without sacrificing accuracy.

Sihi underscores that unlocking true predictive capability requires tight interdisciplinary collaboration: domain scientists who understand ecosystem mechanics, software engineers who build digital infrastructure that won’t break when deployed outside the lab and extension agents and practitioners who test tools in real-world farm conditions.

“We need disciplinary scientists who actually have the knowledge of the system, we need the software scientists who can give it life, and we need the practitioners, … and we need to work together,” Sihi concludes.

The Interdisciplinary Imperative
"By grounding artificial intelligence in high-quality data and making the technology easy to use, affordable, accessible and responsive to farmers’ most urgent needs, we can equip growers with decision-support tools that can help ensure the kind of … agricultural sector we need now and in the future."

The overarching takeaway from these four pioneering experts is clear: Technology alone cannot solve the grand challenges of food production, natural resource conservation and supply chain resilience.

“Success requires bridging federal research initiatives like USDA-ARS, global technology leaders like Microsoft, land-grant university institutions like North Carolina State University and the N.C. Plant Sciences Initiative and others,” noted host Adrian Percy.

“By grounding artificial intelligence in high-quality data and making the technology easy to use, affordable, accessible and responsive to farmers’ most urgent needs, we can equip growers with decision-support tools that can help ensure the kind of strong and secure agricultural sector we need now and in the future.”

About Adrian Percy and the AgTech360 Podcast
N.C. Plant Sciences Initiative Executive Director Adrian Percy is an advocate of the need for and benefits of modern agriculture. As host of AgTech360, Percy explores emerging agricultural technologies with a 360-degree view, interviewing academic researchers, industry experts, farmers, Extension specialists and others. Look for AgTech360 on Spotify, Amazon Music, Apple Podcasts or your favorite podcast app.