
Host Scott Sorrell, Director of Global Marketing (Balchem), is joined by Dr. Miel Hostens of Cornell University, Dr. Mike Van Amburgh of Cornell University, Dr. Ryan Pralle, Steve Martin of DNMC, and Dr. Elias Uddin of the University of Connecticut.
The discussion opens by considering AI as a complement rather than an immediate replacement for nutritionists. After introductions, the panel explores using AI to understand what cows are actually being fed, compare performance with expectations, and suggest ration changes for evaluation through established nutrition models. (00:00 – 08:56)
The panel contrasts mechanistic nutrition models with large language models that produce probabilistic answers from available information. Dr. Hostens describes combining deterministic models with AI, while the speakers emphasize that predictions remain dependent on data quality. (08:57 – 17:59)
Dr. Van Amburgh highlights variables a formulated ration cannot capture, including feeder training, dry matters, load-cell calibration, overcrowding, days in milk, genetics, feed push-up, mixing, and sorting. AI could help organize these signals and train nutritionists to identify what to investigate next, while humans remain involved in complex decisions. (18:00 – 22:12)
The speakers ask whether large language models can reproduce the dynamic calculations of a mechanistic model. They favor pairing AI with deterministic models and sensor feedback rather than relying on AI alone, particularly because poor sensor data could drive incorrect decisions, reinforcing the need for a “human in the loop.” (22:13 – 27:38)
Dr. Van Amburgh illustrates changing cow requirements with a low-nitrogen study involving 144 cows. The starting diet was 14% true protein and 14.7% crude protein; cows increased from 4.6% to 5.3% fat, with some between 8 and 10.4% fat and producing 204 to 262 pounds of energy corrected milk. He says responses like these expose gaps where data and established knowledge remain limited. (27:39 – 30:46)
The panel then explores feedback loops using individual-cow information including milk weight and composition, body condition, body weight, weather, diet, and physiological measures. Dr. Van Amburgh describes using these signals to recognize changing energy and amino acid needs, while Dr. Uddin discusses the still-early possibility of robotics and partial mixed rations supplementing specific cows or groups. (30:47 – 40:00)
Turning to trust and outliers, Dr. Hostens compares AI learning with the scientific process of challenging common knowledge as more evidence emerges. The panel also discusses stochastic modeling to represent ranges rather than single values, potentially supporting decisions involving cost, nutrients, and environmental impact. (40:01 – 45:56)
Using corn silage changes as an example, the speakers consider whether AI could work backward from a desired outcome to recommend ration adjustments. Dr. Hostens sees particular value in tracing problems across herd, feeding, ration, and forage information, but the panel again favors human review before implementation. (45:57 – 50:52)
When discussing the “art” of feeding, the panel considers whether AI could learn from actual diets, feeding errors, forage analyses, cow responses, economics, ingredient markets, and milk solids markets. Nutritionists, however, currently integrate these factors when deciding whether a ration that differs from expectations actually needs to change. (50:53 – 55:08)
In their closing takeaways, the panel urges nutritionists to prepare for change, start conversations with clients, and identify practical short-term AI tools. Dr. Van Amburgh stresses governance and the risk of models being altered or misused, while Dr. Hostens envisions broader nutrition models that build on established knowledge while maintaining responsibility for avoiding mistakes. (55:09 – End)
If you enjoyed this episode, subscribe and share it with your industry colleagues so they can join us at the Real Science Exchange virtual pub table. Don't forget to register for our upcoming Real Science Lecture Series webinars.
Finally, if you want one of our Real Science Exchange t-shirts, screenshot your rating, review, or subscription. Then, email a picture to [email protected]. Include your size and mailing address, and we'll mail you a shirt.
More episodes from "Real Science Exchange-Dairy"



Don't miss an episode of “Real Science Exchange-Dairy” and subscribe to it in the GetPodcast app.








