Consider this situation: It’s 2005, and you’re about to road trip to visit family for the holidays. You carefully plan your route to the destination using a good old atlas, or if you’re really tech savvy, you may even print out MapQuest directions to follow. If you run into traffic, the best options include finding an alternative route using the map or waiting it out.
Now, present-day apps such as Google Maps or Apple Maps use artificial intelligence (AI) tools to help predict traffic to plan routes around rush hour, monitor increasing and decreasing traffic volume, and reroute us automatically. These AI-based tools can even help tell you when to leave to arrive at the airport on time.
Want to find a place to eat along the way? In 2005, road signs were your best navigational beacon. Now, we can plan hours ahead, mapping out potential stops and available options. We can even use language models within AI to search for such places without removing a finger from the wheel.
Most of us are using AI on a daily to weekly basis without even thinking about it. Although it has become commonplace in our standard daily routines, we don’t think about all the ways AI tools are and can be utilized on dairy operations. These powerful tools can not only simplify day-to-day tasks, but they can also help drive decision-making and profitability.
A quick search of “artificial intelligence” in the Journal of Dairy Science database yields just 18 appearances in keywords or article titles, 14 of which are in the past five years. Running a similar search for “machine learning” yields 114 results, with approximately 70% of those articles generated in the last five years. While there is still some hesitation about how and when to use these models, there is a hefty shift toward the exploration of maximizing their use.
At the root of AI and machine learning models lies large, robust datasets, which are pervasive in modern dairies. It seems that in every direction we look, there are more sensors, cameras and equipment capable of gathering and storing this data. However, imagine if all of the mapping apps on our phones were simply showing us the map. What if they simply captured the information but did no further processing? We gain little to no insight or efficiency from this data. When we tie data from these various sources together into an integrated dataset linked to productivity, health outcomes or profitability, we can use these tools to empower decision-making.
In quickly reviewing some of the aforementioned articles, recent research spans from using AI models to predict health outcomes based on data from things such as activity monitors, milk production and component data, rumination and cameras. It has assessed opportunities for genomic improvement using health outcome data paired with genomic data to identify single nucleotide polymorphisms (SNPs) that had not previously been evaluated.
Large language models (LLMs) have also been reviewed for their use in decision-making on dairy farms; however, until recently, few chatbots were specific enough to the dairy industry, which limited their applicability. ExtBot, developed by the Extension Foundation, which is powered by 112 land-grant universities, is an LLM tool designed specifically to assess peer-reviewed articles within the agriculture industry. Think Google AI but vetted for reputability and tailored to agriculture. This tool can make quick work of what would have been hours’ worth of research. For instance, if we ask ExtBot what to do when deoxynivalenol (DON) is detected in dairy feed, it offers several steps and solutions, including verifying sample accuracy and interpretation guidelines, and it even offers some potential actions, including segregation of the feed, dilution or blending of the feed. This tool can take a single point of data, which may not have much value, and pair it with existing resources to help make decisions and create action items to improve animal outcomes.
Although the future looks bright for these AI tools, there are still limitations to their use and uptake in the industry. One of the larger hurdles I have observed thus far in working with these datasets is the disjointed nature of our current data systems. Many datasets have shifted to cloud-based storage, which allows ease of access and easier transmission of data. Yet, bringing multiple datasets together from different formats or inconsistent unifying identifiers (to join the datasets) can also be challenging. Work from Dr. Grant at the William H. Miner Agricultural Research Institute found that rumination while cows were lying was positively correlated with increased butterfat production. When we tie behavioral data with rumination and milk production, we can then analyze which cows are more efficient at this behavior and potentially find factors that impact the likelihood of recumbent rumination. Whether it be stall design, location or general cow behavior patterns, being able to join these datasets provides additional actionable steps.
Some have concerns about overall data safety and security, as these have an impact on our overall food security system. These tools have created solutions to also decrease the labor needed on farms via robot post-dip machines, pre-dip machines, milking machines and even feed pushers. One downfall, however, is that when these machines run into mechanical issues, we must have labor and a plan in place to operate without these integral technologies.
It can be easy to feel overrun by the ever-growing amount of data we find ourselves surrounded by; however, if properly organized and aggregated, this data can provide powerful insights. Similar to genomic testing and evaluation, accelerating the year-over-year increases in milk production and energy-corrected milk (ECM) production, this insight can allow farms to tap into hidden efficiencies and increase profitability at a faster rate. With proper strategy and planning, putting data in the driver’s seat can take dairies to the next level.
