Conversations around artificial intelligence (AI) can go in unpredictable directions; the technology is so unwieldy that getting one’s arms around it is difficult. Factor in AI’s potential, current and future applications, and possible impacts on every aspect of society, and things can get downright philosophical in no time. Dairy industry researchers are dancing as fast as they can to keep up with emerging and ever-improving precision technology tools, while simultaneously taking into account the industry’s evolution — which is these days to a large extent informed by analysis of data collected by those very tools. Penn State University’s Melissa Cantor is a case in point: as an assistant professor of precision dairy science, she is well aware of precision tech’s power — and how overwhelming the AI landscape can be for producers with limited time and budgets. During a recent Penn State webinar, Cantor narrowed the focus to speak about the confluence of precision tech, calf raising, and illness, asking if it’s possible for algorithms to beat humans when it comes to predicting and detecting calf disease.

It's an important question; more studies keep rolling in about the benefits of social housing, which Cantor said has been proven to produce more adaptive cows, beating out those raised individually in social ranking and competitive success. She stressed that 10 out of 14 recent studies found that individually raised calves had reduced growth and/or grain intake at higher feeding levels when compared with their pair-raised peers. If that’s not enough of a push to consider social housing, Cantor also pointed to consumer perception, citing some real-life examples and a survey indicating that the public is more amenable to the idea of social housing over individual.

But while there’s a case to be made for social housing, Cantor also acknowledged the conventional wisdom around raising calves in these programs: Assessing for disease can be a challenge in social housing scenarios. Typically, feeding time for individually housed calves is also assessment time, and when group housing is implemented, it’s often accompanied by automatic feeders, which knocks out that eyes-on-calf, boots-on-ground feeding observation that can pick up signs of illness.

So how can these precision tech tools contribute to the cause? Cantor laid out the basics of robotic milk feeders and their ability to collect data around metrics like milk intake, drinking speed, number of visits, and even teat-nudging activity in the case of some models. Radio frequency identification (RFID) technology tracks the individual calf’s feeding activity, which Cantor posited holds some potential beyond just showing past behavior, but by also predicting future illness. She created a promising algorithm that predicted diarrhea 24 hours ahead of its occurrence — but noted that because feeding protocols change calf behavior, the algorithm only worked when calves had nearly unrestricted access to milk. Activity trackers, which show potential for disease prediction in dairy cows, can also be used for calves; Cantor said that data shows the period up to 72 hours before clinical illness hits is often marked by behavioral changes such as decreased movement. Once algorithms are established beyond the proof-of-concept trial in Cantor’s research, they may be able to predict diarrhea up to 48 hours in advance. Specific to pneumonia, Cantor noted that there also is a correlation between a drop in milk and grain intakes and clinical onset of the disease.

With all this in mind, she pitted technology in the form of precision tools against humans to see which detects illness more accurately. The result surprised her, she admitted. “Precision technology is more accurate than humans,” she said, explaining that in this study, the “technology from the pedometers and the robots can classify a calf as clinically ill at 96% accuracy.” Meanwhile, humans topped out at 80% accuracy. The significance of that difference needs to be weighed alongside the economics of precision tech, Cantor said; farmers might not be able to invest in both automatic feeders and pedometers. A study on pneumonia diagnosis using various budget scenarios yielded the conclusion that just one well-chosen data collection system could contribute to better disease detection — but only after algorithms have been validated in peer-reviewed studies. “The idea is there,” Cantor said. “We are moving in that direction.”

To comment, email your remarks to intel@hoards.com.