At the intersection of tradition and innovation lies the future of dairy farming, which is increasingly intertwined with artificial intelligence (AI) capabilities. AI has emerged as a transformative force as the industry responds to marketplace pressures and technological advancements. While integrating AI in dairy practices presents challenges, it also offers significant opportunities.
Recently, I sat down with two thought leaders in the space: Chad Jenkins, Ph.D., senior nutritionist at Standard Dairy; and John Goeser, Ph.D., director of nutritional research and innovation at Rock River Laboratory and adjunct assistant professor at the University of Wisconsin – Madison, to get an idea of the challenges, opportunities and future of AI in dairy.
How is AI currently being used in dairies?
JENKINS: Some companies in the industry are starting to aggregate on-farm data using data from behavioral monitors, such as specialized eartags or collars. Using that data, the software predicts when a cow has an upcoming health event and then informs the manager to check on the cow. When they return to the system, they are prompted with a question related to that predicted event. As that feedback occurs over time, the computer better determines when the incident will occur – ideally, improving the prediction of the problem before the cow shows clinical symptoms. You hear the terms “training the model” or “machine learning” – this is what that means. The more data fed into the systems, the better the predictions get over time.
Additionally, computer vision as a technology is growing. Using cameras installed on the dairy, software can monitor cow behavior or estimate body condition. These systems will be able to identify individual cows by their color patterns and track behavior and what that behavior might be signaling, whether it's a health incident or metrics like rumination time. I have seen this technology used in maternity areas. Based on behavior, the system notifies staff when a cow approaches calving, limiting the need for physical checks.
Another way AI is being integrated with dairies is through language learning models. These models go through herd management records and the notes associated with events. It collects all that information and interprets trends. It's a deeper look into the language applied to specific events.
What challenges does the dairy industry face regarding implementing AI technology?
GOESER: Picture the size of the Grand Canyon – that would be an anecdote to the gap between AI potential and our current data structure in the dairy industry. Jenkins described a couple of tools out there that are being commercialized. And yet, I contend that's just the tip of the iceberg of what we can do in agriculture with AI. Machine learning is the big piece of this puzzle. However, if we're going to look at any on-farm metrics and relate them back to any nutrition data, whether it's feed management software or the actual diets, we need standardized data. A notification system resulting from machine learning can be trained with informative data relative to an economically impactful output. To build these models, we need well-structured, robust data with inputs and outputs to have a fighter’s chance of successfully training models.
Dr. Robin Johnston has taught me a great deal here. For example, we have a lot of dairy farms out there that are using DairyComp, but unfortunately every dairy's data is just a little different. Dairy A might call something ketosis, and dairy B might call it BHBA (beta-hydroxybutyrate). If we don't have the same thing titled the same way, it introduces issues in terms of data cleanliness. That's one microcosm of the problem. There's great power and great potential out there. Yet, I must emphasize that our industry needs to work on aggregating and structuring data properly before we can unleash AI's mammoth potential.
JENKINS: Besides the inconsistencies in the data model, there is simply a vast amount of data being generated on dairies right now. That data is outpacing our ability to do anything meaningful with it. I wrote an article last year on the topic. I see AI as a potential answer.
How do we make the data usable for these opportunities?
GOESER: For a dairy farmer, find a good partner. Find a consultant who is knowledgeable in this space. Not all nutritionists or veterinarians will know how to support this in operation. There's very little that a dairy can do short of having a family member or an employee who is a software engineer or a data scientist who understands how to do this. Each dairy is unique, yet multisite dairy organizations will probably be able to implement a data management and utilization strategy a bit easier. First, given their larger businesses, they likely have more capital to work with. They will have similar data tables across their various dairies, making aggregating data into something usable for the models easier. Ideally, we want to pull in data from financial software, component and production software, feed management software and nutrition software to drive these models, which in turn aids decision-making. We've got gaps that we need to consider, and we need to find industry partners to help us bridge those gaps.
Also, start small. A few years ago, I tried to force through a project with one dairy. It was a manual effort with a few bright people coming together, and we failed because we tried to do too much. Start simple. Start with your herd management data and maybe one other data table, then add some different pieces and go from there. That is the prevailing theme.
How will AI impact the dairy nutrition field industry-wide? Will we see AI replace nutritionists?
JENKINS: My immediate response to the first question is that we will see performance improve, and the herd will become more efficient and consistent. Diets and nutrition will get more dialed in. However, all the same dairy management challenges will still exist. This brings up the question of whether AI replaces the nutritionist. It may change our focus on where we spend our time – more on the implementation side of nutrition and less on the ration balancing because that could be partially automated.
However, many activities, like physical data collection on the farm, will still be done by the nutritionists – walk-throughs, shaker boxes, total mixed ration (TMR) audits and manure analysis, all of which can be the source of more data fed back to the model. There is a skill set that is unique to a human nutritionist. And then, as Goeser was saying, getting a data aggregating system like this in place and managing it could fall to the nutritionist. We may spend more time on data science as a whole and less time specifically on ration balancing.
GOESER: Consulting needs to evolve, and it will evolve. In 10 years, we'll have a fraction of the dairies we have today. The consultants and nutritionists who will be successful in the future will recognize these opportunities and embrace them. As Jenkins said, there's no reason a human needs to sit behind a computer crunching numbers. Recently, I analyzed a subset of 20 dairies shipping around 7 pounds of components with 100 pounds of energy-corrected milk. When I peeled back the onion layers, I asked, “What's the feed conversion efficiency and the financial performance of those high-output dairies?” This data shifted the paradigm that if we're shipping 100 pounds of energy-corrected milk, we will make money hand over fist. There was a difference of around 10 pounds in dry matter intake among those 20 herds, at least $3 to $4 income over feed cost difference. My question was, “What are the more profitable dairies doing differently, shipping the same amount of milk?”
AI will help get to that answer by transitioning from strictly human intelligence and experience to AI insight. Is it the time of day of feeding? Is it temperature? Those are the types of insights we can derive from machine learning. This is again some coaching from Johnston with Orugen. He taught me that there are things that we would never look for, just in behavior or timing. It might be thought of as the “art of feeding cows,” but we can tie it together in terms of science by using data science and AI modeling to understand the factors that put cows in a position to be more efficient and profitable.
JENKINS: To add to that, being profitable by putting the best cost ration in front of cows happens with standard linear programming. However, machine learning and AI will inform those constraints that go into the linear model to deliver a more refined and accurate best-cost diet. From the nutrition side, my vision is to start from the base nutritional requirements of dairy cows, as with the Cornell Net Carbohydrate and Protein System (CNCPS) or the National Academies of Sciences, Engineering and Medicine (NASEM) model. We then formulate a diet and feed it to the cows. Then, data from the herd management software, information from the milking parlor and the nutrients in the feed will be used as a training platform for machine learning. It can then say, “OK, how is the diet actually performing relative to nutrients delivered?” It can help us determine where changes should be made to maximize income over feed cost. Nutritional requirements for a dairy could be based on its unique circumstances and performance outcomes, and you may start to depart from the basal requirements with machine learning, helping guide the nutritionist's decision-making process to be dairy-specific – fine-tuning in an automated way.
Can you predict what you might see in the future with AI and dairy nutrition?
JENKINS: To start, AI in the dairy space is inevitable. How fast we get there is up in the air. But what you are seeing in other industries today is crazy, and we are just talking about the last handful of years. Ag may be slightly slower to adapt, but it is undoubtedly coming. I think it will look like some of the things we already talked about – automation, instantaneous feedback from performance and behavior, and all those things we can measure to help inform nutritional decisions or changes on the front end.
GOESER: All of the methods we’ve discussed today have the horsepower to get there. Though I'm going to come back to that data piece again. We need the information to train these models to turn loose our skilled people and modeling. And then, if we make headway on the data and get the information fed to these models, we can proceed rapidly. I wouldn't be surprised if, a year from now, we have some of the tools that Jenkins and I are speculating might exist on the horizon.
Where are we going to be in 15 years? I have no idea – absolutely none. I know we will have a fraction of the dairies. They're going to be larger, and they're going to be multisite or further diversified organizations. And there will be a fraction of the operators out there. They will adapt these technologies, and the world will continue to look at us as leaders in this space. I can say that wholeheartedly. We will be working and supporting producers, nutritionists and consultants throughout the world.
Erica Louder contributed this article on behalf of Standard Dairy Consultants.
