From do-it-yourself help (using ChatGPT) for Massican Wines’ winemaker to Atlas Vineyard Management harnessing AgCode data to giant Treasury Wine Estates (TWE) using predictive data software, artificial intelligence (AI) is making inroads in how growers are using AI to improve their bottom lines.
“Every year, farming costs go up and up,” said Kelli Cybulski, director of vineyard operations, Atlas Vineyard Management. “Labor costs rise. Everything is getting more expensive. So how can we use AI to lower the cost of farming?”
At a Napa Valley Grapegrowers event earlier this summer, growers and experts shared a range of options as well as their insights on what AI offers in the here and now and how to prepare for a future where its insights can lead to improved profitability.
Australian AI entrepreneur Ros Harvey headlined the Ahead of the Curve gathering, held in Napa. Founder of The Yield (acquired by initial investor Yamaha Agriculture in February), a sophisticated AI tool road tested with TWE in both Australia and Napa, she explained in common sense terms exactly how AI learns and what it offers.
The range of speakers demonstrated the scalability of solutions.
Kicking things off at the event, Massican Wines’ Dan Petroski shared his insights on how he uses ChatGPT (at a cost of $20 a month) to efficiently route his vineyard visits as he keeps tabs on veraison at the 17 vineyards he sources fruit from and looks at predicted harvest dates.
“My Sauvignon Blanc is in five locations that are as much as 200 miles apart,” he said.
Petroski has programmed the app to show him veraison and harvest dates as well as his best use of miles. “It has its limitations, but it’s very awesome,” he said.
You have to give AI feedback when it’s wrong, he said. “When it does give you the wrong information, I call it out, and it’s like, ‘You’re right. I didn’t look at it that way. I’m redoing it,’ and then it comes back with the right information. It gets lazy. It’s like humans.”
The process is iterative and dynamic, as Harvey explained in her keynote.
She illustrated that iterative nature with a graphic showing the different types of data that are inputs into the AI model. Data, data, data, these models need tons of it, she said.
AI uses data analytics to become intelligent. “It’s called inference,” Harvey said. “You’ve got inputs and you’ve got outputs to predict anything with AI. It’s putting it all into a computer, and it’s learning the patterns between it.”
“AI is half science, half art. What AI is doing is then finding the best model,” she said.
Modern AI thinkers featured at Google Talks are starting to use the word “co-intelligence.”
Harvey stressed that the models are dynamic, just like life. “We’re dealing with living plants outside in weather, and everything’s changing all the time. So it’s the perfect industry for AI. But it’s also hard because of that. You’ve got climate, you’ve got supply chain optimization all along the supply chain.”
Data is the “Fuel” of AI
“You’re not only using it on the farm, but you’re also using it in the winery, in sales, in marketing, in production, all of these different ways along the supply chain,” Harvey said. “Data is valuable up and down, and all this data, which is the fuel of AI, can be used to create models. So data is really important.
“There’s nothing new about this, in a way. If you think about all the industries that are using AI, we’ve already got autonomous vehicles. You can get in a Waymo in San Francisco, and off you go. It already exists.”
Interviewed via email, TWE reported that using the software did improve tasks they wanted it to address: “predicting yield quantity, forecasting harvest timing and tracking growth stages. Knowing why a prediction has shifted helps inform decisions by our viticulture and winemaking teams.”
The software accomplished some of these goals, TWE said. “We’ve seen increased productivity and operational efficiencies, including less time spent manually estimating yields, improved coordination across vineyard and winery teams, and improved intake planning.”
However, the co-intelligence model may more accurately represent its benefits.
“It supports decision-making across the season, though we still rely on local expertise to manage variability at the individual block level.
“It supports the planning process for our vineyard teams. It’s useful to have an additional data source that provides rationale for predictions in harvest volumes and timing across regions or major varieties, especially when paired with historical and in-season data.”
Harvey stressed that wineries need to get their data ready for AI. For many, their existing AgCode data is their first fuel.
Case Study: Atlas Vineyard Management
At Atlas Vineyard Management, Cybulski and her team took a hard run at using AgCode data and repurposing it for more viticultural feedback. The company’s focus on this project crystallized after a data-centric group of investors (who’d done well in pharmaceuticals due to data) became stakeholders.
To get their AgCode data AI-ready, the team created a data dictionary for scouting that standardized reporting. That enabled everyone to use a common set of definitions, something the industry currently lacks.
After trying to develop a custom solution with a data company, the team found the cost prohibitive. Then they found a better fit with another vendor, Orchard Robotics, which uses AI-powered cameras.
“They’re collecting three things in all our vineyards right now. They’re collecting virus status. They’re collecting production status, how much of the vineyard or blocks are producing? Are they fully developing? They’re also collecting crop estimates for us,” Cybulski said.
Harvey: “The Dream Is Very Real”
“The dream is very real,” Harvey said, “that we can get to a point where all this is integrated, where the tractor is going up and down. It’s reading what it’s seeing. It’s logging what it’s doing. It’s feeding it into algorithms. It’s predicting. And you’re actually creating this beautiful thing of a closed system.
“And you’ve got humans, incredibly talented humans, supervising it, all free from all that craziness of trying to wake up at 5 o’clock and work out what the weather is and can I spray, free from all that, and actually thinking about high-value problems of how to create a perfect, perfect vintage. That’s the dream, and all the technology exists there.
“The trick, the very real trick, is how do we get there [where actionable ag data is collected]?” Harvey continued. “How do we break it down in a way that we can get there? How do we reduce cost? How do we reduce risk? How do we increase value? Because if you can’t do those three things, you as the industry, as growers, are never going to pay for it.”
Nonetheless, she said, growers should get their data ready, because AI they will want to use is coming. The industry can shape AI, she said.
An afternoon conversation with Harvey and Kia Behnia, CEO of Napa-based precision vineyard management company Scout (www.agscout.ai), discussed future AI developments, including vineyard monitoring with automated photo collection.