AI at Scale

Alex Rykhva: Harvesting the benefits: AI and the future of agriculture.

In the fifth episode of the AI at Scale podcast Alex Rykhva, Chief Data Officer at Louis Dreyfus Company, shares insights on how new technology, especially AI, is changing agriculture. Alex explains the evolution of use of technology and data in agriculture, new methods of assessing and predicting yields of different products around the year, and some of the challenges and opportunities ahead.

Duration:: 22m
Broadcast on:: 22 Jul 2024
Audio Format:: mp3

The last time the agriculture has really changed was probably about 100 years ago when the agriculture moved from local and mostly manual process to much more automated process. We see that there is a similar shift happening right now where data is becoming a critical factor of growing food in a responsible manner. Welcome to the AI and scale podcast. This is a show that invites AI practitioners and AI experts to share their experiences, challenges and AI success stories. These conversations will provide answers to your questions. How do I implement AI successfully and sustainably? How do I make a real impact with AI? Our podcast features real AI solutions and innovations. All of them ready for you to harness and offer a sneak peek into the future. Hi, everyone. I'm Gosh Agorska and I'm the host of the Schneider Electric AI at scale podcast. I'm very pleased to introduce today Alex Ericfa, Chief Data Officer at Louis Dreyfus Company, a leading global merchant and processor of agriculture goods. There is a fun fact that 30% of what caught on is moving through Louis Dreyfus Company. This is really amazing. And Alex is an accomplished technology executive recognized for building, transforming and promoting digital business through data driven insights. He has an extensive experience across diverse industries, including manufacturing, logistics, retail, healthcare, insurance and professional services with Fortune 100 companies. Alex, I'm very glad to have you today with me. Thank you for the invite. You work in the area that all people on this planet care about, which is food and weather. What does LDC do and what kind of positive impact it brings with AI. Look, I think to your point, we do quite a bit in terms of moving the agricultural goods through the global supply chains, everything that people eat, but also what they close with, as you mentioned on cotton. The world is changing massively with the introduction of artificial intelligence. And not only recently, honestly, we've been using AI for a number of years now to both manage our operations, but also get a better view of the world. To give you the example, we see some positive impacts through enhancing some of the efficiencies, reducing the weight of the supply chain, we manage those supply chains scales. So AI really helps optimize the logistics, ensuring that the truly timely delivery of goods to where they are needed the most. We have ways and proven safety, food safety, but also the safety of the transportation. We do some automatic quality controls. We run traceability of our products with the use of technology. We support and sustainable practices in terms of precision and regenerative agriculture. We monitor the environment with the use of tech. And at the end of the day, we really want to make sure that we apply AI to drive better decision making, both within our organization, but also to provide better market intelligence. And I'll give you the example here, the intelligence of a, you know, given the, I'll combine the weather and supply chains together. This year we are seeing a switch from La Nina to El Nina effect, which does generate a significant shift of weather patterns around the world. So suddenly we see drier than normal Asia and then more, more in falls in Latin America. That has an impact on both how the crops grow, but also has an impact on how the crops get transported. So our forward looking prediction on not only whether they will be a switch, but also at which point of time and what's the probability of that particular switch between these effects is changing quite significantly how we organize our transportation, but also how the world operates a bit as well. So that's a meaningful example of combining food and weather. Sure, so we can say that some of the food industry is now running on data, right? How does the modern agriculture look like with all this technology? Yeah, I think, well, first of all, the agriculture, the last time the agriculture has really changed was probably about 100 years ago. When the agriculture moved from local and kind of mostly manual process to much more automated process with the use of machinery, the use of pesticides, the use of irrigation and so forth. We see that there is a similar shift happening right now where data has become in a critical factor of growing food in a responsible manner. So we see, again, I talked about precision agriculture, we see more and more utilization of data driven farming, where based on data collected, that's quite specific to individual plot of land where the crops grown. That data helps to make an informed decision on things like planting, on water, in an irrigation, on scheduling the harvest and so forth. So we see that it's been massively changing. We see the use of additional machinery, so quite some automation and robotics. So imagine in modern day, there are a number of tractors that don't need to have any manual labor on it as well, so they can run at the best schedule without really being dependent on all the human being running it. We see drones being used quite a bit on doing some aerial surveying, monitoring the crops, seeing one is the good time to plant the seeds and so forth. There's a similar technology when it comes into utilization of biotechnology, so modification of crops to better fit the environment where they're grown. But most interesting stuff is obviously happening in the world of AI and machine learning, where we see quite a bit of predictive analytics, so making sure that we're able to predict both the weather conditions, but also the yields of different crops who are predicting multi-market trends relevant to what's important to the farmer and so forth. So certainly quite an exciting time to be in agriculture right now. Yes, so how do you use satellite data to predict the yields of different products around the world? So satellites, look, how do we do that? We collect quite a bit of data from both publicly available satellites, so the projects, the likes of Sentinel and Landsat. We combine that together with more of a private sources or more of a high-resolution sources, then we apply analysis on, again, on both public weather data. Everything is available through the public flow providers like ECMWF or JAXA on realized weather. We look at something called NDVI, which is a vegetation index, and ultimately we're trying to assess three things. We're trying to assess the overall planting, so acreage of different agricultural commodities, how many beans are growing in Brazil, how much sugar is being planted in Thailand, was the plantation of grains in Ukraine, and so forth. That we can then, we're trying to identify the total farmland, we're seeing which crop is being put on that land, and then the real holy grail is understanding the yield. So understanding how much of the real product and quality will be produced in that environment. There is quite a bit of sophistication when it comes into both taking these datasets from satellites, but also using what we call ground-through data, so understanding of the actual plots. And then we use both public algorithms, the likes of segment anything model from a Facebook, which is exactly the same model they use to clarify whether it's a cat or a human being on the picture. So we use the same on applying classification of fields on those images. And then we have some number of custom-built models that are proprietary to LDC, that are done with our mighty data science team here in Toronto. Yeah, I'm very glad that you brought this example about the cats, because I think people are a bit tired of these kind of examples that, you know, this newest technology is actually capable of recognizing cats and dogs. While actually it can do much more, as you mentioned, you use basically image recognition to assess what is growing where, and then based on this you are making business decisions, right? Well, correct, it's a business decision. It's also, it helps us to better understand the markets. It helps us also defining a bit of the longer-term trend on climate change. In the same topic, we take the internationally accepted consensus data on climate, and then we are able to apply that to a specific crop delineations that we do through the use of AI. And then that gives us a bit of the better view of the longer-term shifts of climate around the world as well. It's very useful for things like understanding the best farming practices, irrigation practices. It gives us the ability to create joint programs as the farmer to ultimately take care of their land better versus what it was done in the past. Yeah, that's the topic that is very interesting for me. How do you work with farmers and how do you maybe share some of this information back with them so they can actually do better in terms of yield, in terms of taking care of the crops, for example? Look, we operate a very large supply chain system, which includes over a million farmers around the world. We work with them, of course, through the commercial deals. So we discuss commercially what makes us to do what not. And then every farmer for us is a partner. So we were trying to create a program that fits best both the objectives of LDC, but also the objective of the farmer. There are a number of examples. Our regenerative agriculture program is a joint program that allows us to put certain investments into new farming methods. Another example would be the work that we do through something called Louis Threyfus Foundation, which is a non-for-profit foundation that's focused on improving the life route of the farmer, especially more of a difficult places around the world. Some of the places are in Africa, some of the farmers that have to deal with very poor living conditions, educational conditions, and so forth. So there are a number of methods by which we engage the farmer. And similarly, I think we are trying to make sure that we introduce our customer base to the better farming conditions as well, so that there is a similar communication on the consumer side. And how easy or difficult it is to basically capture the relation between the agriculture and the climate change, because I heard that you also have some initiatives related to exactly the long-term weather forecasting, extreme weather conditions. And this is impacting really the agriculture globally, right? Well, yeah, look to put a couple of numbers. Again, don't quote me on this. I give them more as the broader sense, but agriculture is what is about 10% of the world's greenhouse gas emissions. The ultimate problem of agriculture is there is too much carbon in the air, and there is too much too little carbon in the soil. So more or less, the solution is you take the carbon from the air, you put it into soil, and some of the best carbon capture technology is called forests or trees. So it just happens to be that the world has the solution for this as well. Now, how difficult it is to drive? I think the world is kind of aligned to scientific numbers related to the warming, again, mostly driven by the carbon. So reducing the carbon and the atmosphere is the goal that has been named and called out and agreed upon in multiple different cross-national conferences, like COP and so forth. So I think right now there is no real debate whether it makes sense to fight the carbon emissions. I think the question is how do you do it? And that's where we come in. We were looking at the ways of reducing deforestation. We have a commitment on deforestation-free supply chains, which is quite a big difference. The difficulty here is from one side is measurement, so being able to capture the right data points frequently. For example, there is no definition of forest in the world, so we basically say, let's make sure that we fight deforestation, but what is the forest? It's not an agreed-to term yet. We hope to fight that. And then from there, turning that into commercial practices. So ultimately, I think that the intent is to make it a win-win. We really want to make sure that from one side we reduce the carbon emissions. And then from the other side, we want to make sure that the industry grows in a sustainable manner, which means that the farmer has to have the bread at the end of the day. They also need to make money so that they can invest into those better practices. And I think the combination of two should help quite a bit. So I'm quite positive and optimistic about the future of sustainable agriculture. Indeed, it's hard to discuss about responsible agriculture without mentioning deforestation. And I guess also in this area, you also use satellite data to capture where the forest actually grows, whatever the definition it is, as you mentioned, which is actually adding to the complexity of the topic. But you do use the image recognition to capture where we have a forest and where the local farmer is contributing actually and keeping the forest rather than rearranging it into the field, right? Yes, well, at the minimum, what we call our first line of defense monitoring is done based on a globally scaled product. So you have to use that type of imagery. That's exactly what you do. If there is no deforestation line too, and I have a number of algorithms that will help me with this definition, then it's fine, then business as usual. If for some reason it's been flagged as a deforestation, we would typically see whether there's a false positive first, and there are a number of false positives. If there is no false positive, we will typically deploy commercial team to go and discuss that problem with the farmer directly, and most importantly, to act on it. Either through reducing our commercial activity altogether, or through building a partnership that helps us to rebuild the landmass, the tea cover mass, so that there is at least the same or more forest on the land, as we had before the cut-off dates, which are typically at around 2021. The way that you describe it, it actually makes me think about something like a digital twin of the whole world, where you can see actually where different crops are growing, where the cotton is being produced, and then you see how this material is actually flowing from one space of the world to another. How do you manage the data quality of this scale of the model? No, look, we certainly partner in terms of the actual infrastructure and technology. We partner with the large hyperscalers. It's the same as the Microsofts of the world, and that helps. So we certainly rely on those partnerships. We also made a number of good, important decisions to utilize public data, because it's also the one that has the most transparency. So at any point in time, anybody in the world can look at exactly the same data as I'm looking at and come up with their own conclusion. I just happened to do it at a large scale in an automated way, but everybody could go and look at the image and then see what the conclusions were making are consistent with the individual conclusion. And then in terms of the actual ecosystem, we managed different products. So by design, we made a choice to architecture to build all the different components of our crop monitoring system as the individual products that can talk to each other. And why that's important, because from one side, we have a very good product that just does the field delineation, and then that same product I can apply to other use cases as well. So I can see whether we can, you know, on field delineation, I see what are not the fields and say maybe what are the roads and then see whether we can optimize the road traffic at the same time. Then I have the optimizers that I plug together on top of this particular data set, and I'm able to get a certain product. So my key point here is architecture instead of building one monolithic solution. Who ended up building with a smaller, a bit more complex ecosystem, about five, six products, but each one of them is able to provide extreme value individually, but also altogether. And coming back to your point around digital twin, I think the world is certainly moving in the direction of building a digital twin. There are a number of very interesting initiatives, again, the one from Microsoft, which I think they call it the planetary computer. I've seen a fantastic work done by NASA and NVIDIA and the Lockheed Martin and billion Earth observation digital twin. I see a number of other companies doing that. It's a very complex task. So I certainly were not yet at the point where we can completely model the earth with all of the intricacies of the earth, but we're certainly moving in that direction. So we see more and more data, more and more precise data, more and more understanding of different signals and how they correlate with each other between weather, the atmospheric patterns, the agriculture growth patterns, the consumption patterns, many other patterns. We see that at a certain point in the future, probably sooner than later, we will get a much better view of the earth. And that would help everybody. I think that's our path towards a much more database evaluation and an action on the world. So again, I'm quite positive that technology is a huge help in defining a better sustainable future. Thanks Alex. Could you give us a more specific example of a project where the data, AI technology was actually helping local farmers doing their job better? Sure. In fact, let me give you a bit of a detailed example. We have about 30% of the farm in land that we manage for our orange juice belt in Brazil. It's about 10 million trees under management, about 30 farms. And one of the big questions there is how come some of the trees are yielding in a certain result, say a good baseline, whereas the other trees are only producing half of the oranges of the trees. And typical example would be, hey, maybe we should irrigate more, maybe we should put more fertilizers, maybe we should put more and more and more. So the typical answer would be what else can be done with those trees with the use of data. It's quite interesting. What we've collected is we will learn that our farms are further south from the belt, which makes them actually more prone to a more rainy conditions, which also makes them more prone to a certain diseases and bacteria. So interesting enough by using of this data, looking at the location, looking at the precipitation levels, looking at the growth of the crops, all the way from flowering to the final collection of orange. What's behind the scenes is collecting lots of data sets into a common kind of platform, doing multiple different analysis of looking at yields, you're looking at the predictions, looking at what changes versus the baseline, and then adjust our farm and practices to those best analysis that we were able to pick up. Okay, I really like this example. So you have good data, then you act upon it, and then you have more juice. That makes sense. Which is also good for you, by the way. It's good for humanity as well, but yes. So actually we are coming to the end of our conversation, and my last question that I wanted to ask is, how do you see the role of AI in the change that you just described? So how important it is, actually, how much this is helping into optimizing all of the flow of the goods of the food around the world, so that we are actually meeting the customer demand, as you mentioned, but also we can actually help solve some of the issues related to climate change and extreme weather conditions. Look, for me, it is the solution. The problem is that the world is so complicated and so complex that none of the human beings is able to properly act on it. We also find that most of the action has to be done, while it has to be globally coordinated, a lot of action has to be done locally, again, so it has to be done at a massive scale. So for me, our only hope is to get AI to help on this journey. We also, I must admit, what we've seen is the move or the transition to a sustainable agriculture has been taken a bit longer than expected. And for me, that's the other role of AI to accelerate that transition, either through better automation or better evaluation of where the opportunities are, or better prediction of where we'll have a specific negative events. And then I also hope that AI will not be as an energy inefficient in the future, so we see that the transition to AI should be supported by the energy and computerization as well. So for me, that's the solution to the problem that we have. Yeah, I think that's a good vision. Thank you so much, Alex, for your time. I really appreciated our conversation. No problem at all. Happy to be here. And look, I really like what you're doing. I think it's a fantastic podcast. And I hope to see more of an interesting speakers on it. I certainly subscribe to it myself. Thank you so much. Thanks for joining us today on AI at scale podcast. Be sure to visit our sc.com/ai website and learn more about our AI at scale solutions. Head over to our Schneider blog platform to read more. Don't forget to subscribe to the show on your preferred platform and share it with your network. Thank you for listening and stay tuned for the next episode. (upbeat music) (gentle music)