After seven years of heading up Google’s Mineral team focusing on developing technology for agriculture, Elliott Grant says this type of work is more like a marathon than a sprint.
And with Google dispersing Mineral — its patents, ongoing work and team — Grant says it’s like passing the baton.
“We chose to pass the baton at this time because it’s the best way to maximize our impact in agriculture globally,” he says.
As of this week, we know what entities are picking up the technology, patents, and ideas from Mineral:
- John Deere will acquire dozens of patents and a technology suite to support development of its See and Spray platform.
- Google Research Africa, The Alliance of Bioversity and CIAT will carry on the work setup by Mineral’s phenotyping methods.
- As previously announced in July, Driscoll’s will acquire and embed Mineral’s yield forecasting and quality inspection tools.
“This is the fastest way to transfer our technology to other players and for them to pick the baton and go and to take it to the next leg,” Grant says. “We needed to work with world leaders in agribusiness who we felt had to capacity to pick up this technology that we developed because it’s powerful technology.”
John Deere provided the following statement: “John Deere has acquired a technology suite from Mineral. This includes patents, pending patents, plant images, and machine learning tools developed by Mineral, a subsidiary of Alphabet. These assets will support the development efforts of John Deere See & Spray solutions to be delivered to more growers. See & Spray technology allows John Deere sprayers to see, target, and kill in-season weeds using advanced cameras and machine learning that distinguishes crops from weeds and selectively target sprays only the weeds. That means less herbicide, less costs, and less impact to crops and land.”
In Mineral’s work — which never had a business model to be sold directly to farmers but rather made accessible via partnerships — Grant says the team concluded partners with access to data, hardware, or end users will be ideally placed to serve farmers and solve the problems specific to agriculture.
“Our mission has always been to make a meaningful, positive difference to the global food system - which we knew was an audacious, high-risk undertaking–so we would regularly ask ourselves: “is this the best way to maximize impact?” and “are we reaching the diversity of farmers we want to, worldwide?” We’ve shared before that we’ve been looking for new and innovative partnerships that can transcend traditional approaches,” Grant says.
He goes on to share that over the past 18 months, the pace of change in foundational artificial intelligence improved even more rapidly than expected. This caused the team to rethink the value and differentiation it was bringing to farmers.
“Therefore we selected organizations to pass the baton to which had the talent, data, market position, and recognized the acceleration that Mineral technologies brings,” Grant says.
He adds about the assets John Deere acquired, “We had developed ag-specific ML Ops tools that were optimized for the unique challenges of training and tuning high performance plant perception models. The tools allowed a 75% reduction in training time per epoch (i.e. one complete pass through the entire training data set), and slashed training time to less than two days for 100,000 images. We knew that every farm and every crop and every season are slightly different — so the ability to retrain and fine-tune quickly is critical to scaling to more farmers and more geographies. These custom ML Ops tools also produced more robust AI models (that means models that work in diverse environments, in uncontrolled lighting conditions, and situations that differ from the training data), which is critical for farmers to capture the economic benefits that will drive adoption of new technologies like precision spraying.”
In addition to the work on the model training tools, Mineral also built and deployed robots, apps, sensors and ingestion pipelines which added up to dramatically improved data collection speed, fidelity and quality.
“While it can be tempting to focus on observable hardware alone, scaling to reach every farmer and achieving the promise of precision hardware requires a full stack of capabilities that are mostly hidden from view,” Grant says.
When Mineral started, in six months the team could collect data, train the model and then release it. This past year, that same work was able to be done in five days time.
“In many ways, that’s been the breakthrough — we’ve radically changed the velocity and the speed at which we could build something. Now it’s up to these new partners to make it available commercially,” Grant says.
Grant sees in the near future when we won’t talk about how AI is used because it will be fully part of products and tools.
“We’re at the beginning of a huge transition for the ag industry, which is the integration of AI into agriculture,” Grant says. “If AI continues to develop at the pace it’s going today, in ten years from now, it’ll be a thousand times more powerful.”
As for the team and talent Mineral employed, Grant says they too will be valuable across different businesses in the industry.
“I am proud those Mineral technologies will enable and accelerate some of the most forward-thinking and consequential organizations in global agriculture: John Deere, Driscoll’s, and CIAT. I’m also proud that the team we assembled are dispersing their expertise into the broader ecosystem,” Grant said.
