Recently, the ag tech company Mineral graduated from “X, the moonshot factory,” to become an independent business within Alphabet Inc., Google’s parent company. The name Mineral is meant to convey how essential, yet invisible, the company hopes to be in the ag value chain—as ever-present yet out of sight as minerals that are a key part of soil nutrients, according to CEO Elliott Grant.
Mineral developed a crop data collection module, a robot rover that trundles over row crops but can navigate all types of fields. The rovers, which continuously gather and transmit field data, are crop-agnostic and have already operated on a wide variety of crop types, from row crops to vegetables and fruits.
The data is only the raw material. Grant told High Plains Journal the most common complaint he hears from farmers, is, “I have all this data, but no insight.” Conversely, some farmers complain they don’t have enough data. “The amount of data needed to create robust models is vast,” said Grant. “We have lots right now, but we always need more.”
What farmers want from tech companies is not merely more data, said Grant, but a “consigliere,” or trusted adviser. The data itself is not the point; it’s the actionable insight into their fields that results from it. Farmers already have relationships with trusted ag retail suppliers, and that’s one reason why Mineral will be working with and through them, signing partnership agreements with them rather than dealing with farmers directly.
“We want to empower your decisions. Farmers already have strong trust relationships with their providers; we don’t want to disrupt that. We will help make the companies you already deal with better and smarter,” said Grant. Syngenta and Driscoll’s, the California-based grower of strawberries, are two early customers.
Machine learning
What makes the difference, said Grant, is machine learning, or artificial intelligence. “How do we bring the power of AI to farmers?” he asks. Machine learning means not only that the algorithms will be continuously learning and self-improving, but also that they will, in turn, be teaching and improving human experts.
As an example, he cites Mineral’s work with strawberry growers. It’s very difficult for human experts to predict strawberry yields, he said. Minerals’ algorithms not only outperformed the human yield forecasters but taught them new things. “It’s a back-andforth learning relationship,” said Grant. “Our process doesn’t replace the experts; it helps make the experts better.”
It’s scale that leverages the power of AI; the more data algorithms have to work on, at greater scale, the better and more precise their calculations will be. That’s why Mineral is making genetic seed data publicly available to the Crop Trust, including from many seeds that are not commercially grown today, but which include traits that might be valuable for crops of the future.
While Mineral recruits plenty of students in computer science, machine learning and other tech fields, Grant insists that all employees spend at least some time in the field with farmers. He himself grew up in East London and worked in aerospace before spending 17 years in the ag industry. When it comes to agriculture, he uses a word not often encountered in the tech industry: humility. He speaks of the data captured by Mineral’s system as part of the age-old crop wisdom used by growers for thousands of years, but that was not always articulated or recorded.
Edge computing
One of the features of Mineral’s in-field equipment is that it uses today’s more-powerful chips to ensure that it can continue to operate and calculate even in the absence of a continuous broadband signal. “We are not betting that rural broadband will be available everywhere tomorrow,” Grant said.
The concept is called edge computing, and it’s a recent development. The way it works is that a device with the right chips is constantly monitoring signal quality, and connecting or disconnecting as needed, invisibly and in the background without the operator having to do anything.
"The advanced chips with low power requirements (developed by Google) ensure that enough computing power is available, even disconnected from the cloud, that operations are not disturbed. “What we do would not have been possible before edge computing,” said Grant. “It means we can run machine learning at very low power.”
David Murray can be reached at [email protected].
