Artificial intelligence has quickly become part of everyday life. It can summarize research papers, organize spreadsheets, analyze reports and answer questions in seconds. But in dairy nutrition, that raises a bigger question: Should AI be able to formulate rations for cows?
According to dairy nutritionist Steve Martin and data analyst Macy Brown of Dairy Nutrition and Management Consulting (DNMC) during a recent episode of the Roadside Ruminations podcast, the answer — at least for now — is no.
AI still has a place on the dairy farm. Both see plenty of opportunity for the technology to improve efficiency and help nutritionists and producers work smarter. But when it comes to balancing diets and making feeding decisions directly tied to cow health and farm profitability, they believe a human still needs to be involved.
AI is Already Part of the Workflow
For many dairy professionals, AI tools are quickly becoming as common as search engines or spreadsheets. For Martin, rather than replacing expertise, he views AI as a faster way to sort through information and surface useful insights. Brown notes that AI tools can help streamline tasks and improve efficiency, particularly when working with formulas and data sets.
“I have used it to help me try to create some better formulas, or at least double check my formulas,” she says. “It’s a great math help when we need it.”
This kind of use lines up with what a lot of nutritionists are already seeing. It can take some of the time out of sorting through data, help clean up reports and point out trends that might not jump out right away. But ration formulation is a different conversation entirely.
Ration Software is Not the Same Thing as ChatGPT
Today’s chatbot-style AI systems work very differently than the ration formulation software nutritionists already rely on. That said, Brown notes that ration balancing programs already resemble an early form of artificial intelligence.
“For all intents and purposes, the ration formulation models are a very early concept of AI,” she says. “You upload all your ingredient files, and you say, ‘I have these on hand, build a ration around it.’”
Martin agrees there is some overlap, but he draws a clear line between mechanistic models and large language models.
“The models that we’ve used are mechanistic models, they have equations and they’re dynamic. In dairy nutrition a lot of it has to do with rates, so rates and pools are just fundamental in how we formulate diets, and so there’s a ton of math there,” he explains.
Today’s modern ration software is built on biology, digestion dynamics, nutrient flow, passage rates and decades of peer-reviewed research. Systems like CNCPS and NRC/NASEM models are continually refined by researchers who evaluate data and decide how new findings fit within established science.
“The models we’ve used are mechanistic models,” he explains. “They’re built on equations and they’re dynamic. In dairy nutrition, a lot comes down to rates, so rates and pools are fundamental to how diets are formulated. There’s a lot of math involved.”
Formulating dairy rations involves far more than generating responses based on patterns or text. It requires biological modeling, math, and an understanding of how nutrients interact inside the cow. The duo note that a chatbot may sound confident in its answer, but that doesn’t necessarily mean the ration is accurate.
The Cost of Getting it Wrong
The biggest concern with letting AI formulate rations is the risk attached to mistakes. An inaccurate ration can impact milk production, feed efficiency, reproduction, animal health and profitability.
Martin says the current ration models already rely on extensive scientific oversight.
“Right now, there’s a handful of people at a handful of universities that are kind of the keepers of the models,” he says. “We put a lot of trust in those people to stay true to the science.”
Those researchers do more than collect studies. They evaluate research quality, account for confounding variables and determine whether findings truly belong in the model.
“I don’t think AI has the intuition to understand some of the things the keepers of the current models do,” Martin says.
He adds that dairy farms operate within highly variable biological systems.
“Maybe we only know about 20% of the influences,” Martin says. “The other 80% we don’t even know what they are.”
That uncertainty is one reason he believes dairy producers and nutritionists should be cautious about handing feeding decisions over to AI-generated recommendations.
“If we just walk away from the mechanistic models and let large language models build the diets, and we make a mistake, the dollars are just way too big,” he says. “The animal health risks are big. The human touch is required in that.”
Farm Data is Rarely Perfect
Even if AI systems improve mathematically, Brown says another major challenge remains: data quality.
Most dairy operations generate enormous amounts of information from feed software, milk systems, inventories, herd management programs and financial records. But raw farm data is rarely clean and consistent.
“Regardless of how detailed, specific, or meticulous an operation is, the data is never perfect,” Brown says. “We’re dealing with cows, and a lot of them, and we’re dealing with employees who do a really good job, but there are still so many opportunities for error.”
A few incorrect values or inconsistencies can quickly change results.
“It doesn’t take very much for one or two numbers to start skewing your results,” she says. “When you don’t clean it up, you lose confidence in it, because it’s like, well, this week was wrong — what other weeks were wrong?”
That is where her hesitation with AI grows stronger.
“I still don’t know that I trust any of the AI options out there to accurately be able to clean up data,” Brown says. “A lot of people say that they can. I’ve still yet to be convinced of that.”
The challenge is not just technical. Definitions and measurements often vary across a dairy operation. Something as simple as “cow count” or “shrink” may mean different things depending on who is looking at the numbers.
“There’s reasons for some of the gray area to exist,” she says. “Without a human touch in there at one or two points, it’s really hard to have the confidence level in it to know exactly what you’re looking at and why you’re looking at it.”
Where AI May Actually Fit in Dairy Nutrition
While both Martin and Brown are cautious about AI-generated diets, neither dismisses the technology altogether. Instead, they see opportunity in using AI as a support tool around the ration — not as the ration formulator itself.
Martin believes AI could be especially useful for evaluating how a ration is performing after it is fed.
In that scenario, a mechanistic model would still formulate the diet. Producers and nutritionists would then combine feed data, milk components, herd management information and other performance metrics. AI could help organize that information, identify trends and flag areas worth investigating.
“I think the path forward is to fully utilize the ability for AI tools to assess our current situation,” Martin says. “Let that system say, here’s what we think we’re feeding, here’s what we think it should produce, what’s it actually producing? And then I think there can be a meaningful conversation.”
But he still wants actual ration changes handled within traditional nutrition models.
“Where I want to draw the line is, I don’t want to get out of the mechanistic model and into the large language model to have that next ration be created,” he says. “I want it created in the mechanistic model.”
Brown shares a similar outlook.
“I’m pretty pro-technology,” she says. “On the surface level, the idea of using AI is really exciting to me, because I can see where it improves efficiency. I think we just have to think through how to use it and to make it a good tool for us and our clients.”
AI May be an Assistant — Not the Nutritionist
AI can help summarize reports, organize data, flag inconsistencies and identify patterns that deserve a closer look. Those tools may become increasingly valuable as farms continue generating larger and more complex data sets. But ration formulation remains rooted in biology, math, experience and interpretation.
At least for now, Martin and Brown believe the best approach is combining advanced technology with proven nutrition science rather than replacing one with the other.
