THERE is probably no topic that is more front and center in our world today than artificial intelligence (AI). People are either excited, indifferent, or scared to death of this new technology. No matter if it is going to create the panacea that it promises, the end of our world as some fear, or something way less dramatic, AI is sure to shape our future.
Entering an AI era
Since it seems this train has already left the station, we might as well be thinking about how the nutrition sector of our industry can take advantage of these capabilities to hopefully feed cows better while remembering that these cows have complex, and often unpredictable, biology. Perhaps the sweet spot is to begin working in the realm of helping the AI machine better understand this unique creature we strive to feed. Wait a minute, though; Haven’t we already been doing something like this for at least 30 years? More on that in a moment.
I was thinking about what was the most commonly used AI tool that is well accepted by the average person. After a little web search, my first guess was correct. Using the map app on your phone, which is about as widely accepted as anything in today’s world, is full-on use of AI. It has been this way since Google Maps launched in 2005. So, we have all been using a straightforward example of AI for 20 years now.
Which route to choose?
Why the conversation about a map app on your phone in a dairy nutrition column? Well, the way we can use optimization models in ration software has many similarities to the way a map app gets you to the airport to return your rental car in a town unfamiliar to you. Even more, you probably use Google Maps to make your way to your hometown airport to park your own car in your favorite airport parking lot. This general approach for combining information you do know with things you don’t know, while adding additional information like driving preferences, is so very similar to using an optimization model that has been in almost every nutrition formulation software for at least three decades.
The surprising thing is that with the vast majority of dairy rations formulated every day, the process used is more like starting to go to the airport using your favorite way and not using the map app in your phone at all. We all know the reasons that it might make sense to go ahead and let the map on your phone help you get to the airport, even if you could drive it with your eyes closed. You can predict things like schools letting out or normal morning traffic, but not unexpected slowdowns, accidents, and road closures, to help you adjust accordingly. Even if you are in an unfamiliar city, you might know the general direction of the airport and the key roads to get you there. When choosing this route, you can look for airport direction road signs and “exit here” signs. No analogy is perfect, but this one is pretty good.
The routine reformulation of a diet that a nutritionist has closely managed for a long time is very much like driving to your local airport when you know all the roads. This drive and formulation can be fairly predictable and would seem to offer few surprises. Yes, you can pay attention to some unexpected things and make a few decisions along the way. In reality, if you used the maps app on your phone, you would get to the airport a few minutes earlier. Likewise, if routine ration reformulation utilizes the optimizer, you will feed the cows maybe a little better and most likely for a few cents cheaper.
Optimal optimization
Considering ration formulation optimizers, there are two main types. The first, linear optimization, has been used for building rations for livestock since the 1980s. I used an early ration optimizer in 1985 at Auburn University. These are mathematical manipulations that check and recheck many iterations of workable solutions in a short amount of time before choosing the best solution. These solutions are usually the lowest cost options, but other optimization goals can be selected. This linear approach assumes a constant relationship between the nutrients supplied by various ingredients to combine to meet the animal’s requirement.
The other type of ration optimization that is available is called nonlinear. The key difference between linear and nonlinear is the capability to account for the fact that the relationships between nutrient supply, animal requirements, and response is not a fixed ratio and is more complicated. In general, nutrients have reduced efficiency as supply and production go up. Additionally, a dynamic model that can include various interactions between ingredients in the same ration can better predict production when using a nonlinear optimization tool. In short, it is more complicated but offers a better solution with a better chance of supporting the desired milk production.
Unfortunately, in most if not all, nutrition models taking the step from linear to nonlinear optimization is a much more complicated and often frustrating process. It is a bigger jump to move from linear to nonlinear than to go from not using optimization at all and deciding to give the linear optimizer a try.
The goal here is to not give up. We can feed cows better at a lower cost when optimization is fully employed. No, the computer does not know cows as well as the experienced nutritionist, but it can do math better, multitask, and understand more completely how various feed ingredients can work together to create a better, lower cost ration. If the optimizer offers a bad and unworkable feasible solution, then ask it a better question and try again.
Tip of the iceberg
Driving to the airport and formulating a dairy ration aren’t the same thing, but the similarities are striking. As we are pressed with tighter margins, increasingly difficult nutrient management requirements, and numerous other goals, it may soon be old-fashioned to build rations for dairy cows without the use of optimization. Soon, the lines will blur between our current nonlinear optimizers and the same models with AI, machine learning, and real-time cost information — along with other influences like milk solids pricing.
The nutritionists of today have the opportunity to embrace these changes beyond levels of more than just penwalks. And no, we won’t hand it all over to the computer and lose our influence.
