Ag Geek Speak
GK Technology Inc Team Members, Jodi Boe and Sarah Lovas talk about precision agriculture, agriculture mapping, agronomy and drainage.
Ag Geek Speak
9. Will AI Replace Farmers? The Role of Human Interaction in Artificial Intelligence Featuring Travis Yeik
We explore the future of human decision-making in precision agriculture and the role artificial intelligence might play in supporting – not replacing – farmers.
• Travis Yeik brings unique perspective from his background in farming, agronomy, and software development
• AI has existed since the 1950s but recent advances in computing power and algorithms have accelerated development
• Farmers must make complex decisions integrating agronomy, mechanics, business, finance and weather variables
• Current AI excels at specific tasks but cannot integrate all the complex knowledge domains needed for comprehensive farm management
• The true potential of AI is as a decision-support tool that helps farmers process overwhelming amounts of data
• Data availability remains a significant challenge for training AI models capable of nuanced agricultural recommendations
• Increasing farm sizes and tighter margins create a need for better decision-support tools, making AI potentially valuable
• Integration of AI should enhance rather than replace the human judgment and experience essential to farming
In part two of our conversation, we'll continue exploring how AI tools can help make farmers and agronomists more efficient without replacing their essential roles.
https://gktechinc.com/
Welcome back to Ag Geek Speak, and this week we have a very interesting topic and a great guest to help us talk about this topic. First of all, I'm going to introduce the topic and then we'll introduce the guest, because he's been here before. So the topic for this week is actually going to be about the human interactions with AI and precision agriculture, with AI and precision agriculture. So, in other words, what part of the decision-making process should humans be responsible for, as opposed to AI, helping us be more efficient with making these decisions, and what can AI do effectively? So, to help us with this conversation, of course we have our very own Jodi Boe, who is on every episode of A Geek Speak, and then, of course, we have our very own Jodi Boe, who is on every episode of Ag Geek Speak, and then, of course, our guest for this topic conversation is Travis Yeik. Travis, do you want to introduce yourself and make sure that we all know what you sort of do with AI here at GK Technology?
Travis Yeik:Yeah, thanks, sarah. I'm excited to be here and talk with you guys on the subject. I have a background.
Travis Yeik:I grew up on a farm ranch in southeastern Wyoming and we did a lot of irrigation and we raised beef cattle and as well as some dairy cattle for breeding. And from there I went to the University of Wyoming and I have a degree in geography or remote sensing and GIS and a minor in soil science, and then went on to the University of Nebraska-Lincoln for a graduate degree to get a better education or further education in remote sensing and specifically dealing with remote sensing in agriculture. And from there I worked for Valley Irrigation for a season or so and I was their agronomist for variable rate irrigation season or so and doing I was their agronomist for variable irrigation. And then I joined GK Technology back in 2014 and I've been with them for 11 years and I developed the ADMS software that is widely used by farmers and consultants, and so I guess one of my main goals then is to develop software for that's easy to use and intuitive and just helpful throughout the precision agriculture.
Sarah:Well, that's great. Thank you for that introduction. So I think you know one of the things that we talk about in precision agriculture a lot is that interaction or maybe the complicated relationship that exists between the practical agriculture people agronomists and farmers actually in the field and the computer programmers that are behind the software, trying to run the software to work with the practicality. But it's interesting that in your background you actually have that agricultural background and you've been out in a field so you understand some of the nuances and the unpredictability that can come with just common agriculture and farming practices every day.
Travis Yeik:Yeah for sure. Culture and farming practices every day yeah for sure. I think probably there's not a whole lot of people that go into the tech side or to the coding side, I guess, after doing some farming or having that background, to relate to people on that level. And I think it's super important though, rather than having somebody from another country developing software without being able to interact and provide that support, get that feedback directly from the people who use the software. Really.
Sarah:And I think, as we're prepping this conversation to that same point, you know okay. So Jodi and I we work on the sales side of GK Technology and Travis is working on the product development, the computer programming side of things. So Jodi and I are working with agronomists, we're writing maps, we're writing prescriptions, we are really on that practical side of making things work out in the meetings where we're trying to understand for sure how things are working on the farming side of things. How do you think about this and those sorts of things, to make sure that what you're programming is really, you know, practical, going into things. And I do think it's interesting, because there's been times where I feel like maybe I've asked some questions about how things work for you. I probably don't ask enough questions like that, though.
Travis Yeik:It is totally like a symbiotic relationship, right, you have to. I mean, we have to understand each other to be able to move toward that. Though. It is totally like a symbiotic relationship, right, you have to. I mean we have to understand each other to be able to move toward that goal. Yeah, yeah.
Sarah:I honestly think that's one of the greatest strengths of our company to be to be real about it. But you know we wanted to talk really about AI. You know artificial intelligence because you have done some programming with artificial intelligence for some products, have done some programming, um, with artificial intelligence for some products that hopefully we're going to be incorporating into our, into our software and our daily, daily, um, daily practice. So, travis, we hear AI all the time, um, and we're not in the cattle industry. So really what? And we're not in the cattle industry, so really what is AI when it comes to precision agriculture?
Travis Yeik:Yeah, it used to be artificial insemination right, that's how I knew it growing up.
Jodi:Yeah, I've talked to dad about it and and I wonder if he thinks which one is it? Yeah, travis, he's working for a software company, but he keeps talking about AI, like what what the heck is going on.
Travis Yeik:No, yeah. So artificial intelligence is, in its very simplest terms, is being able to mimic what humans can do to to achieve processes, and it's it was created back in, like the 1950s, which is really hard to believe that it was that long ago.
Sarah:I had no idea.
Travis Yeik:Yeah, yeah, and this term was coined in.
Travis Yeik:There was some models that were called Markovian decision process and bellman equations, and these were used back in the like 1960s and 70s.
Travis Yeik:Right, you watch those old-time videos I don't know if you've seen them, probably where, uh, they um have, oh, what is like war games that was, that was one that was popular. I think it came out in the 80s. But, um, yeah, where you have this ai that's playing games and it's and it's taking over the world and that's you know. So they had this stuff clear back when and it really isn't until recently 2014 ish or so, is when, uh, it has become popular and it's it's grown tremendously, partially by the equations themselves. Uh, there was a big influence or big kickoff with that, with Google developing some open source AI software and researchers being able to use this software in universities, and then the other part is now that we have computers that are able to handle the processing, such as the GPUs needed for processing all the data and the memory and storage that way, and so that's a big part of why it's grown here within the last what, even 10 years or so.
Jodi:So what I'm hearing is like this this concept of artificial intelligence. What it is is basically taking data, putting it into an equation and kind of predicting what would happen, or mimicking, like how humans would take an information and then make it quote, unquote, make a decision and like. Up until now, we've been limited in terms of like what those equations are themselves. So Google has helped to bring out a better one that can be worked on by researchers. Google has helped to bring out a better one that can be worked on by researchers, and we also have better processing, and these two things coming together have made it so that when we turn on the TV or look at the news, artificial intelligence is something that we see. Almost every time we look at something in the news, it's there. So that's huge and I feel like we'll probably probably this isn't going to go away.
Travis Yeik:Yeah, no, it's interesting. I was looking. There's a very popular equation and it's with reinforcement learning and it was developed by DeepMind, google back in 2013, 2014 or so, and it was one of the first ones that could truly play games such as Atari back in the 1980s, the Atari gaming system and this equation was able to. After it learned these games, it was able to play as well as a human can and so, and then you fast forward now, seven years later or so, and since then they've had other equations come out for reinforcement learning. Some of them are called the AlphaGo you might've heard that one that was popular here about five or six years ago, being able to play chess, or the game Go, which is a Chinese game, and beat the top Chinese player in Go, which is just huge, because this game has hundreds of different actions or possibilities that you can take within this game to complete your goal there. Take within this game to uh, yeah, to, to complete your goal there and uh since then.
Travis Yeik:Now we have a recent one, such as uh equations. Again, they're called dreamer three or uh, or efficient zero or moonet, and these ones are now 500 times better than that equation that came out in 2014, right, and this and these, these ones came out in 2021. So that's seven years later and we're already 500 times better. So what's it going to be in another seven, 10 years? And we're going to be, you know that much more improved again.
Travis Yeik:So this thing I like to compare it maybe as like when we had computers, the internet, come back or come out in what 93, 94? I remember getting the first email address when I was back then and being able to surf the net, and since then, you know, computers are just in everybody's life, no matter what you do, like if you're a secretary or a contractor or a farmer, even right, everybody uses them now. And yeah, it could be that way with AI, where it's just going to be so necessary for our jobs to do, and yeah, so it's interesting how things are progressing and how we're at the beginning of this revolution. Probably is what it's going to be and where we're going to be here in the future we're going to be here in the future.
Jodi:So I've got a question of like clarification, so like when you mentioned equations, like you mentioned the AlphaGo equation, or would even like ChatGPT be considered like an equation.
Travis Yeik:Yeah, so they use several different ones. In its broadest term, chatgpt is called an LLM, which is a large language model, and but they also in the back end of those there is some reinforcement learning. So an LLM is it uses a variety of, I guess, equations. It's hard to explain.
Jodi:Probably don't put this on air, but I'm just thinking here out loud Because I guess, like the point I want to make here is like air, but I'm just thinking here out loud, because I guess, like, the point I want to make here is like it's more than just like a Y equals MX plus B, like a linear equation that we might have all heard or like think of something about. It's. It's a lot. It's like a big giant Excel, like if then statement like is that kind of how to think about it? Like, how, how do we think about an equation? Is it as simple as thinking about having an input and then that input goes through the model and we get out what happens next? Like, how do we think about these things compared to what we assume is an equation?
Travis Yeik:Yeah, so nowadays, when we think about the AI, we really are thinking about machine learning.
Travis Yeik:We think about the AI, we really are thinking about machine learning, and that breaks down into that deep learning or into reinforcement learning.
Travis Yeik:In deep learning there's training, whether it's supervised training or unsupervised training, and a lot of that is used for an image analysis or for these large language models that chat gptp uses, and it can then generate responses, right, or generate images in some sense, and so, as compared to then, reinforcement learning, which is more of your input and output, where you take, uh, we have a state or an observation that we're looking at and, based on this observation, we can have different actions, and it takes that, that observation and says, hey, based on this, let's do this action, and it is able to learn just similar to what, um, what, what babies or humans or dogs can do.
Travis Yeik:Right, if you want to tell a dog to sit, you train it and you give it a treat as a reward and after a while it says, oh, hey, if I come to you and I sit down, I'm gonna get a treat, and that's kind of what the reinforcement learning is, and so it's like chat gpp they use both of those in in some sense, um, but they're, yeah, so they're a large language model and in the background they kind of those in some sense. But they're, yeah, so they're a large language model and in the background they kind of use some reinforcement algorithms to help progress things and learn as your conversation goes on with it.
Sarah:So what kind of treats do computers eat?
Travis Yeik:That is a good question. That's a tough question.
Jodi:It's really just a negative one or one right. A negative one would be a negative. My computer like freezes now I'm just gonna scream that at negative one and hope that it does never do that again.
Travis Yeik:Yeah but I think about this like on the term. So I'm this is probably out of context and you can totally cut this out, but like, based on that, though, we keep hearing things of oh you know, ai is going to take over the world, it's going to take over farming, it's going to hurt humans in the future. Right, based on what it does. To me, it's all based on what we teach it. Um, if we have a say, we have an algorithm that says I want to make the environment safe, clean, the, the best thing possible. It's going to do that and it's going to get rewarded to do that. So all of a sudden, it says okay, well, humans are causing the environment to degrade or polluting it, or whatever it is.
Travis Yeik:Well, it's going to learn that solution to fix that right, but it also has to be trained then, like, how does it get rid of humans? Does it get trained to kill them? Does it also has to be trained then, like, how does it get rid of humans? Does it? Does it get trained to kill them? Does it get trained to poison them? And so then it has to learn that as well. Which are, you know? I don't know. So, to me, I don't see ai as as that level where it's learning like set several different tasks that are completely unrelated. Uh, really, we're training on one single task, and to be able to to kill humans is just to me it's like intuition, which it doesn't have. It's just training based on on what we want to teach it.
Sarah:This is a really great conversation right here. This is really interesting and very pertinent to agriculture right now, because one of the things that I am hearing in the countryside is fear over not being able to drive a tractor anymore. My ability to go out to the field and spend time in the field, which is the part of farming that I love, is going to be taken away from me because of AI, and so, really, what you're saying, travis and correct me if I'm wrong is that unless we train it to do that, it's not going to be that way.
Travis Yeik:Yeah, totally.
Jodi:And adding on that too, like from what I think I just heard from you, Travis, is that these models need some sort of definition of what winning is and what their goal is. They need a human you know this human interaction part of AI. They need to know what they're moving towards right Like with Atari, it's really easy to see like or a video game, you have an end goal most of the time, unless it's like animal crossing or some open world, but they're you're still rewarded in that game setting of like what to do next and how to move and how to operate yeah, so it's like this tiny little model they've learned.
Travis Yeik:They've learned to just play that one game right now, you can't just turn around and have it uh, intuitively know how to tie a shoe right. They're completely unrelated things and so they would have to be taught on this task and that task to be able to, and then to put those tasks together, like whether they're related or not, like that's. That's like truly human interaction, like you know what, what we do in our brains and and deductive reasoning, and to do that with an ai boy we're I.
Travis Yeik:I feel like we're generations away from that still that's a really good point yeah, this is a great conversation yeah, you were talking, though, on like yeah, and I think when you first sent out the invitation for me to talk you, I think you sent it out. As is AI going to take over agriculture?
Jodi:With that guy with the beard?
Travis Yeik:Yeah, yeah. So what's your thoughts? Is it going to?
Sarah:What do we mean by take over agriculture? I mean that's a great, like what you know. Do we mean that humans will literally not be farming anymore? Because, essentially, if AI were to theoretically down the line, take over agriculture and we don't need farmers for the entire decision making process anymore, why do we need farmers?
Jodi:That's a good question and I think it comes back to what we just talked about right, like right now, it seems like AI models can be trained to do one thing, but in agriculture there's so many things going on all the time, right, that's why we talk about farmers as wearing all these hats because they're mechanics, they're expert tractor drivers, they're decision makers, they're agronomists, they're financiers, they're CEOs there, they're agronomists, they're financiers, they're CEOs, like there are all these different things that don't really connect or like they don't really fit into one area. So I feel like, in order to really create a model to replace a farmer, I mean, you're asking a computer genius to build that to make it, I think, successful enough to rival a farmer.
Travis Yeik:Yeah, Maybe it's like a better thing to say is like what can it do now? And you know if we're still in the beginning of the AI in that revolution, and what can it do now and what is the future of it going to be, and what are we really looking for? Are we looking for it to take over, you know, all human interaction in farming? I would highly doubt it. I don't know if that's the end goal or not. It would seem to me that there's always going to be that farmer that owns the land. Right, we're not going to have robots, AI, owning land, and so it's going to be a farmer making those important decisions.
Sarah:Okay. So I think it's interesting because, really, the three of us come at this whole angle, from trying to use technology and precision agriculture to make good agronomic decisions, right To try to be the most efficient that we possibly can be with our agronomics, whether that's drainage, whether that's fertilizer, seed, chemical. How are we going to manage that? But farmers, you know, they are making economic decisions. When do I sell the grain? How do I sell the grain? When do I decide to buy that next piece of land? When do I decide to buy that next piece of equipment? Oh, the equipment broke. How do I fix that equipment? What is the right?
Sarah:So there's all of these other decisions that also go into being a farmer, and I think that's something that's really important for the industry to remember.
Sarah:For the industry to remember, you know, because in our little agronomy world which, when you think about what a farmer does every day to Jodi's point that she brought up earlier a farmer has to wear all these hats, right.
Sarah:But in our little agronomy world that's just such one small piece of the puzzle and even within that it can be so hard to have a Technology make good decisions, because we are dealing with life science for our decisions. You know there is it's not just an equation out there that makes plants grow. We have biochemistry in that plant. We've got enzymes in that plant that can make chemistry equations balanced that would never be able to balance outside of that system. We've got slightly different shades of green across different plants that you could never describe, except that they're just naturally a little bit of a different green. So there's all of these things that are occurring. Oh and, by the way, did I mention the weather that always tends to throw in a monkey wrench into a lot of things that we're doing. So these challenges from a computer programming standpoint have just got to be complicated.
Jodi:My thought in this, too, is like I just don't think there's enough data out there to train a model to do these things. Like I think about you know what as a farmer, like what I would want AI to help me with on the farm. Like one thing I would love to help get help with is like, hey, do I decide to replant or not? But guess what the AI model doesn't have? It doesn't have what my soil temperatures have been. I don't record that I should, but I don't have that data. I don't have a yield monitor on my combine.
Jodi:So like it's got no yield data about, like my specific area, but even, and even outside of that too, though like it needs data about what canola yields, like when it's planted on a specific date, that specific hybrid, and like we have you know research that's done by universities, but like that's a low amount of data and that would also need to be inside of a model. Like there needs to be good data, I think, to teach these models. And like right now we collect some data in agriculture and that has been, I think, a critique of precision ag these last 10 years is that there's so much data and not you know enough things being decided on that data, because, I mean, who has the time to do that? Maybe that's where AI helps us, but like there's not even enough, like data, I don't think that we could build a model to help us make some of these hard decisions that, as farmers, we have to make.
Sarah:And Jodi, to that very point. When you think about replanting canola, that is one small decision, one small agronomic decision for an entire farming year, and think about how much data is required for just that one small decision.
Travis Yeik:Yes, so you guys both bring up great points. So yeah, like the weather AI is now, it's actually doing pretty decent with weather. It used to here five years ago it could predict out maybe six hours, right, and now it can actually predict out maybe a day, two days, and in the next 10 years it might do better. But as you say that's one system, say that's one system, and so again, uh, kind of my my now or my what I said earlier, how saying, hey, this can play atari and now can tie your shoe, well, it's the same thing. Well, it can predict weather really well. Well now, how does that relate to, uh, the soil nutrients and how does that relate to disease?
Travis Yeik:and and the all these systems come together and they kind of mesh into one big net and to be able to have an ai to say, okay, these are related in some tiny way, and to be able to model that and now change it just a little bit and put put us in north dakota and now put us in california or wherever else, right, and how the entire system just changes completely again based on all these different factors. That goes right into Jody's points. How much data is needed to model that? That would take an enormous amount of resources and time to be able to gather all that data. To say that it could yeah, I might be able to to understand that in maybe not with our algorithms now, but 10, 20 years, sure, why not Based on you know how we're progressing, but you have to get that data to do it. Wow, I that that's quite the feat to come to.
Sarah:So let me ask you this, to this, to this, to this whole point of this conversation. So when you think about, like weather data you know we just talked about, you know, taking that decision model from North Dakota over to California just based in the weather data alone, would we have? Do you think, based on your, your knowledge, your work in AI, that we would have to have two separate AI models for those environments to make that, to make good decisions work? And I know that's a loaded question.
Travis Yeik:That is a loaded question.
Travis Yeik:I'm good at those.
Travis Yeik:I personally don't deal a lot with weather, but from what I know is that AI can take in just tons and tons of data and filter through all that and say, hey, this tiny little bit of information here is important, and so when it goes through in this AI, they go through these neural networks is what they call them right, and so it's kind of like a branches on a tree these neural networks is what they call them right, and so it's kind of like a branches on a tree.
Travis Yeik:We're at the bottom, we have that trunk, and then we have a little bit of data that separates out to the different trunks, and then it separates out again and again and finally we've got a million different nodes that reach out to all the different branches. And so it's kind of like our weather system, right, Like we have some things the trunk of the tree which might be universal to all the weather or whatever, such as rotation of the earth or however. But then as we keep going into little branches, each one of these observations changes and changes and changes, until we get different actions with each different observation.
Travis Yeik:So, yeah, ai can take in and process that all weather data into one single algorithm and that's a lot of data that it has to process. But now to take all that weather data and also take all of another system's data, such as disease or nutrients, and add that and then add another one and another one, and add that and then add another one and another one, that's where I don't think ai has that, um that ability to do stuff like that yet, yet being the keyword.
Travis Yeik:I'm not very good at predicting the future, um, but that's fair.
Sarah:Yeah, and especially for this conversation, because I think there's people out there that want ai to predict the future for us.
Travis Yeik:Yeah, and that's important too. It's like, right, we, we don't want to replace farmers. Even even me as a coder, like I don't want to replace farmers, like they're super, super important and I don't want to replace jobs, but it's all. It's a tool, right, that you use. And that's really what it is is to help help make decisions Right. And if you can have I don't know let's say, 50 different models, one that predicts each and every little thing, you as a human can put those together and say, ok, I see, based on this information, given that this might be my best decision for this small, tiny little aspect. And now you can relate that to say, ok, well, based on this, I need to make this financial decision or this small, tiny little aspect. And now you can relate that to say, okay, well, based on this, I need to make this financial decision, or this time to plant or this amount of nutrients to put in, or whatever it is. And I think putting that all together is a tool to help us make better decisions.
Sarah:And let's be real about this. Farmers have to make big decisions every day, and the market is demanding that. The margins that farmers have to work with are so tight that the market is demanding, you know, nutritious food that's affordable. And the farmers have tight margins on the backside and so the demands of them to go through and be as efficient as they possibly can, that's there. That's why we see increasing farm sizes, economies of scale. So, you know, farmers are getting over larger acreages at a time, but we're able to variable rates so we can address those, those, those nuances within the fields to make sure that we're the most efficient right. That's, that is what the market is demanding. And at the same time, if they're, if those farmers are larger, farmers are dealing with more decisions as well, and more detailed decisions, than they ever have been in the past.
Travis Yeik:Yeah, we are inundated with the amount of data coming in from all these monitoring systems and to be able to make knowledgeable decisions or process all that data coming in, you've got to take time in your day to do that. All that data coming in, you know, you got to take time out of your day to do that. And how important whether that is, you know financially, supporting financially on whether that data is or if we need. Yeah, where am I getting at here? Sorry?
Jodi:I think this is really good fodder for a second part of the episode, because what we're talking about is that there is more and more incentive for farmers to be more efficient. With their time, there's a lot of data that could potentially help farmers be more efficient with it. The missing piece is, you know, could AI be that tool that condenses this data down and helps farmers become more efficient? And I think we could have a really good conversation about, like, what are some ways that would work? Where are some ways that humans still need to be a part of that? Because I think there still would need to be.
Jodi:But that's kind of where I think this is all leading to is how does AI help us be more efficient not replace us as agronomists or farmers, but how does it make our jobs better or our lives better? So thank you so much for joining us on this first part of our conversation with Travis. This has been fantastic thus far, so please stick around. We'll continue this fantastic conversation in part two, and with that, with GK Technology, we have a map and an app for that.