Herbicide reductions of about 50% are already common where AI-guided spotting systems target weeds plant by plant, according to research discussed at the Ames, Iowa-based Council of Agricultural Science and Technology's webinar.

Alex Thomasson, professor and director of the Agricultural Autonomy Institute at Mississippi State University, cited John Deere's See and Spray system as proof that artificial intelligence has moved from concept to field. See and Spray uses AI to apply herbicide only where the machine detects weeds, rather than across the whole pass.

Trials at the University of Arkansas on more weed-specific herbicide use in soybeans back the number up. "Researchers' studies suggest that it's fairly common for a herbicide reduction of about 50%, so value is being created in that context," Thomasson said.

He and Andres Ferreyra, a data asset manager with Syngenta, co-authored an assessment of AI in agriculture roughly 18 months ago.

What does AI actually mean on farm?

"When I use the term AI, I'm talking about the field of computing and focused on simulating human intelligence, and in that, I'm including perception, learning, problem solving, and decision making," Thomasson said.

That definition matters for farmers deciding where the technology pays. Historically, digital agriculture collected, stored, displayed and communicated data. AI goes further: it interprets it.

"It will identify patterns in the data, it will predict outcomes and it will even recommend actions," Thomasson said. That does not require robots. An irrigation system that switches itself on based on AI predictions counts too.

Why is agriculture a hard environment for AI?

The variable, outdoor, biological nature of farming makes it demanding territory for algorithms.

"Agriculture is a biological system, and we know that biological systems are inherently variable," Thomasson said. Weather changes continuously, and pests, diseases, weeds and insects are unpredictable and vary from one field to another.

Soil conditions vary too. Fertilizer decisions shift field by field, which is why the underlying data has to be highly accurate. Systems must also be safe, reliable and economically practical — the standard tests any grower applies to a new machine.

The payoff is getting closer to plant level. "AI is taking us to the point now where we can have plant level detection and AI can identify individual plants at a particular location very quickly," Thomasson said. "If we are able to do that then we can have targeted inputs."

That capability is the bridge to autonomous field operations.

In livestock, AI enables continuous monitoring humans cannot match. "We can do things that humans have never really been able to do, and that is continuously, 24 hours a day, monitoring health and behavior of animals," Thomasson said — catching disease early, detecting stress or spotting reproduction events.

What does reliable agricultural AI depend on?

Thomasson listed the prerequisites:

• Scientifically credible, high-quality data
• Broadband connectivity to move large data volumes
• GPS and global navigation satellite systems
• Large-scale computing capacity
• Interoperability between equipment and software
• Validation and trust built through rigorous field testing

Cybersecurity and human oversight matter as much as the hardware. "This is a new technology that we're dealing with, and the agricultural workforce in general does not have the capability to work with it well," Thomasson said. "Workforce development is going to be essential."

Fragmented data remains a practical barrier. A farmer planting with one machine and harvesting with another may find the two data streams do not integrate easily. Data also varies by crop and region, which is why Thomasson called for benchmark datasets and common evaluation methods to reduce duplication and strengthen scientific credibility.

"Independent field validation is important as AI takes on greater decision-making responsibility," he said. "If we start using autonomous machines in agriculture it's very important that these machines are doing what we expect them to do."

Where is AI already working in agribusiness?

Ferreyra noted AI adoption has run ahead of production agriculture in other parts of agribusiness. In animal processing, AI tools flag to managers that something has gone wrong, or predict it could without corrective action.

In poultry processing, where supervisors watch for misshapen wings or bruising, cameras and image-processing models linked to AI can improve efficiency, he said.

Both authors flagged the tension between AI's infrastructure demands — data centers, utility investment, permitting and water-use decisions — and agricultural and community needs. Thomasson argued the technology should expand grower capability without reducing choice, with land-grant universities, Extension services, community colleges and industry all playing a part.

"Policy will influence infrastructure standards, competition, workforce preparation, and trust," Thomasson said. "We want to capture AI's benefits while recognizing and managing its problems."

Watch for independent field-validation programs and benchmark datasets as the next benchmarks of trust — and for plant-level detection to move from trial plots toward targeted-input autonomy on commercial farms.