Podcast Short: FieldLark AI Brings AEA’s Expertise to Every Grower

Hi friends, this is John. Welcome back to the Regenerative Agriculture Podcast. Today is a day that I've been dreaming about for the better part of a decade. Today we are announcing the launch of fieldlark, our AI agronomist chatbot that we are now releasing to the public. I've talked about our development work with Fieldlark on a couple of recent podcast episodes, so we've certainly been teasing it out there. There have been some social media announcements from Advancing Eco Agriculture, where we gave access to an early beta access group, and we've been continuing to develop it quite rapidly here over the course of the last several months and I'm quite excited about what we have to release. So you can find Field Arc on the Advancing Eco ag culture website advancingecoag.com and I think Fieldwork shows up in the nav bar. If I'm not mistaken. You can find it pretty easily or you can just do a search for it as well. Fieldlark is not yet available and I'm not sure that it will be anytime soon. It's not yet available as an app, but you can save it on your phone, at least on iPhone. I don't know how this functionality exists for an Android, but you can save it on your phone, just save it to your home screen. You can easily find instructions to do that on, how to do that. It's pretty easy and pretty straightforward, so it actually shows up as an app icon on your phone. And fieldlark, this functionality and this capability is something that I've been dreaming about for a long time because I realized, starting about there were several things that happened, several different pieces that came together that have had me thinking about this for the last decade. Plus one aspect was the realization that to do regenerative agronomy well, or to take this agronomic approach that is based on biology rather than based on chemistry, is very multifaceted. There is such an abundance of different information that becomes relevant that when you want to really understand how to manage the microbiome well, how to manage plant nutritional integrity very well, then all of a sudden you start paying attention to factors that you've never paid attention to before. Soil redox characteristics become important and the form of nitrogen that you apply and that the plants are absorbing becomes important all of a sudden. Ammonium versus nitrate versus urea. You start realizing that each of these different forms of nitrogen has very significant produces very significantly different outcomes on plants from a plant health perspective and from a microbiome perspective than the others do. They are not all created equal. And so there is just this abundance of different types of information that all of a sudden become very relevant, very meaningful and very impactful. You can make, you can have, produce significant yield responses and crop quality responses by managing all these various factors. And then there is also all of this adjacent discipline information where we use different language and different terminology to describe the same phenomena, but in one area it's understood much more deeply and perhaps much more thoroughly than in another. For example, I just mentioned reduction and redox environments. Understanding oxidation versus reduction. Well, when you go to human health, we don't really have conversations about reduce reduction versus oxidation in a human body, but instead we talk about oxidation and antioxidants. And it's the same thing. We're talking antioxidant is a reducing compound. We're using different language to describe the exact same phenomena. And so that's just one example. There is this abundance of, of discipline adjacent knowledge from other life sciences that is incredibly valuable and useful in the soil and plant health and animal health management and domain. That and the level of information, information and knowledge is growing so rapidly year by year that it becomes extremely difficult even for people who are deeply trained in a specific space to just stay abreast of and stay on top of all the new information that is emerging, that is coming forward. And so realizing that this different type of agronomic approach is much more knowledge intensive and in some cases, many cases, much more management intensive, I realized years ago that we need to find a way to democratize this information and to make it readily available to anyone. And so it was that. What is it now? I would guesstimate it's maybe half a dozen or more years ago. A friend of mine introduced me to a mathematician and engineer who developed the original concept of what today in the AI world are called large language models or LLMs. So I had the privilege of having a breakfast meeting in New York City with, with the individual who came up with the foundational math and the science behind what today has turned into chat, GPT and Grok and Gemini and Claude, etc. And when they described the capabilities, the theoretical capabilities of what this technology would be capable of, that it would be able to aggregate information and put together coherent explanations from across a broad array of different papers and research and domain knowledge, I became quite excited about the possibilities that were represented because all of a sudden you could now ask very specific detailed questions such as the effect of selenium on the fat content of forages, and you would be able to find answers in A matter of minutes that previously would have taken hours and hours of in depth research to discover. So I saw these tools and this technology as a pathway to democratizing information. And I became quite excited about what that could potentially mean for the adoption and the thought development of a different approach to agriculture and to agronomy. So that was one aspect. A different aspect was when I when we first started using plant SAP analysis back in 2011, we did, we worked with SAP analysis, we started testing it in 2011. We used it through 2012, 13, 14. At the end of the 2014 growing season, I started realizing the implications of this technology that we could not just use this information to prevent diseases in insects with nutritional balance, but that we could also use it to predict disease or insect susceptibility into the future. And I realized that when we integrate this information from SAP analysis results with the various types of soil analysis results, Haney analysis was not really on the scene at that point. But even just looking at the total mineral assays that we were running at that point from the geochemical mining perspective and combining that with the more common Melek3ammonium acetate soil extraction, basically looking at the the intersection of soil analysis, different types of soil extractions with SAP analysis, I realized that this could provide a framework for completely reevaluating, not this is completely disregarding the biological component, just looking at the chemistry component alone, realizing that we could have a completely different framework for understanding soil chemistry interactions and how that corresponds to actual nutrient release and nutrient availability at different stages of plant development. And then when you integrate the modern technology that's just emerged in the last couple of years on microbiome analysis, where we now can measure the microbiome that is in a plant, around a plant root system or within the plant leaves or within the seed. We can essentially at this point we can measure the microbiome as I understand it, in any location in soils and on or within plants. We're not really there is we can extract DNA from all of these different locations. And now when you when you combine that microbiome profile information with the soil chemistry profile information and you do that and across a number of different locations, different interactions we've here on the podcast over the course of the last several years we've talked about the some of the foundational research work of Dr. Christine Jones. I've had conversations with Keith Burns at Green Cover Seed and the work that they are doing and specifically trying to as best they can trying to find cover crop seed that is grown in biologically active Soil and considering cover crop seed as a vector for microbes. All of a sudden what I expect to emerge is that when we combine this microbiome data with soil analysis data, with plant nutrient analysis data, with plant phytonutrient analysis, we start looking at the actual plant profiles of amino acids and carbohydrates and phytonutrients that, such as the carotenoids and so forth that plants are developing all of a sudden. I expect that with large data analysis we'll be able to see lots of interesting trends emerge. I expect that we will find plants in or crop and soil contexts where, oh, these plants, this particular crop are doing a particularly good job of absorbing phosphorus from the soil and that exceptional phosphorus absorption is associated with these particular microbiome characteristics. There is this group or this community of these 50 microbes that have synergistic community effect that are present in these 50 samples that aren't present in the other 500 samples that makes them outliers in phosphorus absorption. I'm just using this, this is completely hypothetical example of the things that I expect we'll be able to find once we start looking at this data in this way. And of course when, particularly when you start looking at microbiome data, you have this very large data set that is difficult to comprehend with, with a human brain just because there's such a sheer volume of data. And to make sense of this and to, to reveal the relationships that exists in the, in these, in these multiple dimensions. Actual plant measured nutrient absorption contrasted with soil geochemical analysis contrasted with soil agronomic analysis contrasted with the microbiome. When you start looking at all these different, different types of information, this is the logic, a logical place to use machine learning, to use deep learning and now also to use AI on top of that. And so this is, this is the, the next phase, the next iteration of Field Lark that we are working on. So Field Lark is now publicly available. Anyone can log in and can access it and use it. But what is present, what is publicly available at this moment is the, is essentially a glorified chatbot. Well, I shouldn't call it glorified. Yeah, it is a glorified chatbot. It is, it is equivalent to Gemini or Grok or Chat, GPT or Claude, but it is deeply trained specifically on agronomic relevant and adjacent information in addition to all of the information that we've developed internally here at AEA over the course of the last two decades. So, and we've, we've routinely tested it and I welcome all of you to test it side by side if you have familiarity and experience with these other tools as well. When it comes to, it's obviously a specialized and focused AI, and so it will, it, it's not designed to answer other more general questions, but when you focus specifically on agriculture or agronomy related questions, it exceeds the output of these other AIs by a very significant margin, particularly when you start asking it about how to solve specific problems, how to solve disease or insect problems, and how to, how to solve ecological interactions. The depth and the quality of the information that emerges from fieldlark is very different from what we've been able to find anywhere else. So that is what exists right now. But the, one of the strengths and also one of the possible downsides of these large language models is that when you ask that there is this inherent variability. If you ask exactly the same question multiple times, you're going to get slightly different responses every time. And particularly if you ask the question slightly differently, you're going to get variable responses. And those variable responses can be valuable for revealing edge cases, edge information. And you can also customize with the prompting. You can ask it for a detailed explanation or a thorough and comprehensive explanation and you will get very exhaustive information back when you, when you specifically ask it for that, or you can just simply ask it for a summary and it'll give you a summary of the information. But there is, there is this kind of inherent variability in how a chatbot functions and there is a, there's a time and a place for that. And there is also a need for precision. So the next stage of Field Lark that we are actively working on at this point have been for some time is an algorithmic engine that is not variable. Where you provide it with, you provide it with the context. This is the historical context, this is the historical fertilizer applications, pesticide applications, soil amendment applications, these are the historical yields, Here are the historical soil analysis. And you can, you can train it, you can build in all of this information, train it on all the contexts, it'll look at location, it'll take historical meteorological data into consideration. So there's, there's a lot of pieces that are some that have been built, others that still remain to be built. And the output from the algorithmic engine is not variable. So it will make very precise and repeatable recommendations to say that, well given, given your context and all these factors, here is a set of recommendations for precisely which cover crops are the best fit for your context. And to solve your agronomic problems. Here are Here are specific soil amendments. Here are the foliar applications that you should consider. Here are the specific microbes that are the best fit for as a seed inoculant to shift your soil microbiome. So it will make very precise recommendations that are consistent and repeatable based on the information that you provided. And the vision is that we develop these two components to communicate with each other that you have the algorithmic engine and the chatbot, and as you interact with it and you work with it to ask it questions about your operation, it will and you provide the algorithmic engine with all of this contextual information. Those two will integrate and you will get ever better and better more accurate recommendations and be able to ask questions to understand those recommendations. Ask well, why exactly did you recommend this microbial inoculant? Why exactly is this cover crop a good fit? And be able to get explanations to understand why that is being recommended. So there's There have been some interesting conversations. If you've listened to the podcast episode that we released last week, I was interviewed by some friends of mine for another podcast and we talked a bit about Fieldlark and the the intention, the thought process that has been behind it. I started down this pathway. Again, uncertain of timing here, but it's been a decade plus. I first read Ray Kurzweil's work on the approaching singularity and it's given me a lot to think about how we develop technology. And I wrote an article last fall, it was published in Ag Funder News and I think perhaps also in Acres. I'm uncertain right now, but yeah, I think it was also published in Acres and I originally titled it Designing Technology for Regeneration. And so I've thought deeply about how do we develop technology, how do we design technology so that it facilitates regeneration and not further destruction? And one of the foundational aspects is so much of technology removes people from their relationship with the landscape. And so I think it's important for us to develop technology that has the opposite effect, that increases people's connection to the landscape. So one of the tools that we haven't yet released publicly, but in conjunction with the algorithmic engine, we also have developed a crop scouting app. And these tools that we are developing and Fieldlark already facilitates this to some degree. I believe that from a foundational perspective, technology should not be designed to replace people. Instead, it should be designed to help make people better. Because there are things that technology can do better than people, such as just sheer computing power and analyzing large data sets. But there are Other things that human intelligence far exceeds machine intelligence. And Gary Kasparov actually wrote an interesting book. I was having a conversation this morning with the same friend who introduced me to, to the original developer of large language model technology. And he was describing how when Garry Kasparov and this famous incident where he was the first time a computer beat a human in playing chess. And Garry Kasparov was the global chess champion at the time, and he wrote a fascinating book called deepthink. If I recall correctly describing the differences between machine intelligence, or what today we would call artificial intelligence and human intelligence, they both have inherent strengths and they both have inherent weaknesses. And it so happens that the inherent strengths that humans have the strengths of intuition. And there's one that I'm forgetting, but there's intuition and there's long term strategy, long term strategic thinking where humans historically are much better than what at the time would have been machine learning or deep learning. And I, and as I understand it, even AI as it's developing with the large language models, it just so happens that many of those skills and strengths that humans are better at than AI are exactly the skills that are needed to be the best growers, the best farmers, the best agronomists. And so the way that we should be thinking about developing technology is not to replace people, but to make people better, to get us to be better observers, to ask better questions, to be better informed. Those are the things, those are the pathways in which AI can help us. We should be thinking about these large language models as an extraordinary librarian that we can ask questions and it can, it can pull all the books in the library and all the data and all of the knowledge that it has access to and derive the answer from all the information that it has. But simply having access to a good librarian doesn't make us a better farmer, better farm manager. We still have to be able to be out in the field and to make those good, good observations. So it's, it's, for me, it's quite an exciting day there. There's a lot of momentum with Fieldlark. I encourage all of you to please try it and let us know what your, what your experiences are, how well it's working for you, what you're discovering. I mean, we know that AI is still imperfect. There are things that the largest models like Gemini and Grok and ChatGPT still are still make mistakes on. We should expect, we should not expect these tools to be perfect. We should not expect them to not have errors. We need to constantly guard against hallucinations and there is still absolutely the need for human review. So these technologies, the chatbot technologies, are still very imperfect, but they are able to organize and pull information together in a truly remarkable way that far exceeds what we are able to do unless we, well, even I think at this point, AI has exceeded what even the most knowledgeable and advanced humans can do from a memory and a recall perspective. But that, of course, is very different from creativity. What else is important that I've missed thinking about? One other important aspect is the vision for the future as we continue to develop these technologies here at Advancing Eco Agriculture. Our vision, building on what Garry Kasparov wrote in deepthink, is the ultimate, the ultimate vision, the ultimate goal, the highest horsepower, if you will, the most effective, if you will, is not artificial intelligence alone, it's not human intelligence alone, but it's actually when those two work in concert in conjunction with each other, when you have the combination of human intelligence and artificial intelligence. And so the. I expect we need to continue to see how this evolves in our work at aea. But I fully expect that there, there is, will need. There is going to be a rapid and growing need for farmers and agronomists to work closely with these tools. The most effective, the most effective agronomists in the future, and I think the most effective growers in the future are going to be those that combine their skill sets and their strengths with those of an incredible librarian, an incredible tool that can pull a lot of information together and help them to make strategic decisions and to be more creative, which is something that those tools cannot do. They cannot be creative. But humans can bring creativity. So please let us know and also please let your friends know. This is. We're not investing a great marketing budget in launching Fieldlark. We're a very small company with limited resources. And I would ask you if you find Fieldlark to be useful and if you find that it brings you delight, as I have been delighted by the answers and the responses that it has returned, please share it with your friends. Please post about it on social media. Please tag me, tag aea, and let us know what your responses are. I'd love to hear from you. Thank you all for what you bring to the world and may we create a better future that we can all be delighted by. Thank you all.

The team at AEA and I are dedicated to bringing this show to you because we believe that knowledge and information is the foundation of successful regenerative systems. At aea, we believe that growing better quality food and making more money from your crops is possible, and since 2006, we've worked with leading professional growers to help them do just that. At aea, we don't guess. We test, we analyze, and we provide recommendations based on scientific data, knowledge, and experience. We've developed products that are uniquely positioned to help growers make more money with regenerative agriculture. If you are a professional grower who believes in testing instead of guessing, someone who believes in a better, more regenerative way to grow, visit advancingecoag.com and contact us to see if AEA is right for you.