HERMISTON — As the use of artificial intelligence becomes more widespread and accessible, people are beginning to grapple with its benefits and challenges, but U.S. Department of Agriculture researchers believe they’ve trained AI to assist potato breeders and farmers.

During the 51st Hermiston Farm Fair on Wednesday, Dec. 4, at the Eastern Oregon Trade and Event Center in Hermiston, Max Feldman with the USDA Agricultural Research Service presented his team’s approach to making potato breeding more efficient by evaluating the spuds with a computer program.

The two-day trade show and seminars brought in about 150 people looking to earn continuing education credits on its first day, with many tuning into Feldman’s talk on creating artificial intelligence for potatoes.

The system should help select for preferred phenotypes — physical or visible traits, like weight or skin color — to improve the offspring of genetic crosses, getting more potatoes with good traits by industry standards and fewer with bad ones.

“If we can measure everything about their offspring, we’ll know which parents are better for the traits that we care about,” Feldman said after the presentation. “AI is what enables us to capture phenotypic measurements at scale.”

Humans can do this work, but on a large scale it is labor- and time-intensive, which is expensive for producers. Feldman said his team wanted to know if artificial intelligence could determine how many potatoes in a sample have a given characteristic that would be useful for breeding.

“We can get more measurements for less labor, I think, using some of these approaches,” Feldman said. “What that translates to is being able to make better choices, either through directing our crosses or through selecting offspring for the next round of crosses.”

Using a camera and conveyor belt, Collins Wakholi with the USDA trained artificial intelligence to analyze photos and detect various characteristics, such as selecting potatoes without growths, cracks or sprouts, or identifying them by size or color.

Looking at the results will help scientists determine how likely a given characteristic is in a parent population. For example, if one type of potato tends to have more defects, its offspring are likely to have more defects, which ends up costing producers.

“It tells us, you know, from our 1,200 samples we ran last year, which ones tend to have more sprouts or more growth cracks than the others,” Feldman said during his seminar.

Feldman said he expects to analyze 75,000 more potatoes this winter, which should help strengthen the model’s accuracy — which showed mixed results in terms of true positives and false negatives for identifying various defects.

He said he’d encourage potato producers to evaluate whether having more data and knowledge like this could be useful for their operations, as he expects similar programs to be accessible in many areas.

Feldman said he doesn’t expect this form of artificial intelligence to replace humans, though.

“Now, we actually get to talk and think a little bit more about what’s coming across the line,” he said after the session. “So I think that it enables us to do more with what we have and be more precise.”

But crop consultant Kevin Baker, an attendee of the presentation, isn’t so sure. He said artificial intelligence is “definitely going to replace people,” and noted recent changes in the dairy industry because of modern technology.

Despite that, Baker didn’t say he believed artificial intelligence is bad for farming. In fact, he said it will be a “very useful tool when it’s figured out.”

“It’s going to cut down on time and labor, and right now growers can’t afford any more expenses,” he said. “The more we can save them — or the more they can save — the better off and the longer they’re going to be in business.”