How John Deere Got Good at AI
Updated: Jul 16, 2020
Poised for prime time and now enabled by the Industrial IoT, AI and machine learning offer manufacturers and industrial equipment makers the potential to tap into a rich stream of information to do everything from improving the consistency and quality of visual inspections to gaining key insights into operations.
But for agriculture, it’s already here. Agricultural equipment of all types are leveraging the power of AI to help make farming more efficient, more sustainable, and more profitable.
Fierce Electronics interviewed Sona Raziabeegum, Strategy Lead, Digital Solutions at John Deere about the company’s transformation from a traditional equipment maker to one of the most technologically advanced in the world and what it takes to succeed at AI and machine learning.
FE: How did a manufacturer of farming and industrial equipment transform itself into an AI and data-driven business?
Raziabeegum: It’s important to know that innovation has always been a part of John Deere’s DNA, dating way back to 1837, when our founder created the world’s first self-polishing, cast-steel plow. Fast forward to 1919, when the company actually completely disrupted itself by deciding to get into the engine business. That was a pretty radical thing for a company that up until that point was dependent on animal horsepower.
Fast forward another 80 years to 1999, when John Deere entered the precision agriculture space by acquiring NavCom, which had the technology for correcting GPS signals down to an accuracy of a couple of centimeters. This was really the beginning of localizing machines to the field, which was another pivot in our journey. Then, a relatively short 18 years later, we entered into the realm of AI and machine learning with the acquisition of the AI startup Blue River, a developer of smart farm machines.
Throughout this journey of constant disruption, we obviously were very focused on innovation, but I think the company’s success is as much due to the fact that at our core we are customer focused. We put customers at the center of everything we do, always striving to develop technology to make their lives better, help make them more profitable, and be more sustainable in their operations.
In the end, being customer focused is essential for AI to succeed.
FE: Acquiring an AI startup seems like a great way to get a jumpstart on the technology, but any integration can be quite challenging. What did John Deere do to get this right?
Raziabeegum: Culture is so important and I think it’s the one thing you really need to make sure you get right with an acquisition. We have had experiences in the past where we’ve made an acquisition precisely for the innovation it could bring, but our own internal processes prevented us from fully exploiting the creative talent.
We’ve learned that in order to preserve the entrepreneurial spirit of a company that we acquire, we need to treat the company like an innovation hub and give the team the space to stay creative. We recognize that what we as a Fortune 500 company bring to the table is that we can identify real customer needs that the technology can solve and that we can effectively mainstream and scale the technology because of the size of our network. To do that, we have to be outcome driven and customer focused.
FE: What changes, if any, in the existing workforce did you need to succeed at AI and machine learning?
Raziabeegum: If you were to look back 10 or 15 years, our technical workforce consisted mainly of mechanical and electrical engineers. But we are slowly transitioning from being just a manufacturer of hard iron to embracing bits and pixels. It is a slow transition, and I want to emphasize that we still value the core engineering skills, because that is what built this company and made it successful.
As we aim to acquire new skills in AI, machine learning, and other areas, what’s exciting is that the new talent from Silicon Valley [from an acquisition] is heavily engaged in building and championing a curriculum on these topics. We offer the courses through John Deere University to anyone in the organization. Participating in this training isn’t mandatory, but if you are interested in developing your career and learning new skills, the company fully supports it.
FE: Can you talk about the inherent challenges of AI and Machine Learning in the ag space?
Raziabeegum: In ag, the problems are complicated and the applications take on a level of sophistication and complexity that most people probably don’t appreciate. For example, the job of an autonomous vehicle is to basically move from point A to point B. In contrast, a combine or sprayer is not just moving between two points, it has to do a job along the way, and it’s a job for which you have to have a high level of precision.
For example, our See and Spray technology targets and applies herbicide only to the weeds, reducing the amount of chemicals required by a substantial amount. But it has to distinguish between a weed and a plant on the fly in a matter of nanoseconds.
In ag, you’re dealing with millions of plants across thousands of acres, yet each plant has to be treated individually—not only is each plant unique, but you have to factor in things like their orientation and their maturity. The need of the hour is precision. You are not just needing to differentiate between a dog and a hydrant, the differences are much more nuanced when it comes to plants. Some weeds and plants are so similar it’s very hard for the human eye to distinguish between the two.
The other thing is that in ag, connectivity is not a given, especially in rural areas. And yet you may have to make a decision in nanoseconds. For that reason, most of our AI models have to reside at the edge, and these systems have to be custom built. So, from that perspective we really are pioneers in this space, because our AI architecture is so customized.
FE: What advice would you give to other companies looking to develop AI-enabled products?
Raziabeegum: The most important advice I could give is that it’s critical to not only test the technology in the real world, but also to develop a deep understanding of how the customer will adopt that technology in their world. I think too often companies are so focused on the technology—granted, that’s hard enough—that they discount the customer experience until too late in the process.
Trial and error is so important with AI and machine learning. We have the ability to create AI engines and machine learning simulations, and the way we do this is by using curated data to build the initial model. But no simulation is going to be perfect. Unless you take it out to the real world, you will not be able to solve for the actual problems.
A challenge is that real-world data is messy and there’s likely to be a lot of behaviors and practices that your customers follow that you may not have accounted for. You need to really understand how it is going to be used by the customer, and how it fits into their current workflow. You’ll also need to understand if the value of the product is aligning with the cost.
While it may be tempting, you cannot create a one-size-fits-all application that is so dedicated you can't modify it, because you’re likely to discover that some customers choose to use it in an entirely different way. And they’ll need your help in understanding how to use it.
In the end, success with AI is all going to come down to field trials. So, my final words of advice: Get out of the lab and get out into your customer’s world--now.