Latest / Grounded Intelligence: Why Most AI Innovations Never Reach the Field / A Systems Approach to Agricultural AI: Stewart Collis on Data, Trust, and Farmer Access

Transcript

A Systems Approach to Agricultural AI: Stewart Collis on Data, Trust, and Farmer Access · Grounded Intelligence: Why Most AI Innovations Never Reach the Field

• 0:00
  Dr. David Bergvinson: Was the importance that data plays in developing these models and refining them to better serve farmers.
• 0:05
  Stewart Collis: Bringing together a community around how do we share that data, build a much bigger corpus representing smallholder farmers for pre-training and post-training was really important.
• 0:16
  Dr. David Bergvinson: This new AI era offers a similar promise and a similar opportunity to have an impact at scale.
• 0:22
  Dr. David Bergvinson: But it means that we're having to think through the responsible use of these technologies and to ensure that we don't betray the trust of farmers as we design, develop, and deliver these services.
• 0:33
  Stewart Collis: But, you know, interestingly, with these new conversational interfaces where you have two-way flow of information directly to a farmer, you can actually start to build up a profile about that farmer without having to do a big, you know, government-led data collection effort.
• 0:49
  Dr. David Bergvinson: One of the resounding messages was, yes, uh, we wanna take advantage and apply these new tools in the AI era, but we wanted to have some sovereignty over [00:01:00] the data, the models, the infrastructure to unlock that impact.
• 1:19
  Dr. David Bergvinson: Welcome to Grounded Intelligence. I'm your host, David Bergvinson, and for the last 30 years, I've had the privilege of working with farmers around the world in different production systems, hearing about their story, about the challenges they face, these aspirations that they have for their farm, their family, and their communities.
• 1:38
  Dr. David Bergvinson: But so often is the case they're held back by a wide range of constraints, whether that's rain-fed production systems impacted by climate change, whether it's access to markets that are fair and give them an, uh, an income that they can live off of, or just access to the right inputs so that they can grow the right variety, access the right fertilizer, [00:02:00] or get the right extension service for them to realize their full economic potential.
• 2:06
  Dr. David Bergvinson: I've been privileged to work with industry leaders around the world in crop improvement, and being a corn breeder or maize breeder myself for more than 20 years gave me a perspective on, uh, engaging with farmers to understand the requirements. What are the needs of that product? What does it need, uh, to, to contain in order, uh, for it to be useful?
• 2:26
  Dr. David Bergvinson: So we call that demand-driven innovation or human-centered design. I've also had the privilege of working with the Gates Foundation in setting up digital agriculture 15 years ago, and a lot of the lessons there, we had all this promise of this new digital technology, and yet it really didn't deliver at the scale that we had hoped.
• 2:46
  Dr. David Bergvinson: This new AI era offers a similar promise and a similar opportunity to, to have an impact at scale, but it means that we're having to think through the responsible use of these technologies and to ensure that we don't betray the trust [00:03:00] of farmers as we design, develop, and deliver these services. So I'm very excited about this topic.
• 3:07
  Dr. David Bergvinson: It's a topic which I've been passionate about for more than 15 years, and it's exciting to see it coming to fruition now with the advent of artificial intelligence. So join us on this journey. We look forward to your comments, your suggestions. Give us some hard topics to tackle on the podcast, 'cause we would love to make sure that we're capturing a wide range of perspectives and challenging topics to position this technology to deliver on its promise.
• 3:34
  Dr. David Bergvinson: I would just like to introduce you to a community that is really galvanizing around this important topic of the application of artificial intelligence to empower farmers, and that community is called AgX AI. It's a community that has built over a number of years on the use of information technology services, and this community is specifically focused on the responsible use [00:04:00] of artificial intelligence to make sure that we're not leaving farmers behind in emerging markets like Africa, Asia, or Latin America.
• 4:08
  Dr. David Bergvinson: And so you ask, "Well, what is this AgX AI community doing?" Well, it's doing many things. Uh, first, it's just finding each other, uh, you know, organizations and individuals with common cause to unlock the power of AI to empower farmers. We're also looking at a range of challenging topics that are important in creating the ecosystem or setting the table for artificial intelligence to be successful.
• 4:35
  Dr. David Bergvinson: So that includes, uh, data that is actually representative of the communities that, uh, we're, we're looking at and how we aggregate and deliver that data to a wide range of organizations. It's around model development that is tuned to the needs of communities that, uh, also represents those that often don't have a strong voice, such as, uh, minority groups or women or, [00:05:00] or youth.
• 5:01
  Dr. David Bergvinson: It's around benchmarking to compare models, uh, because so many of them are being developed now, but we don't really have a way of comparing them against each other, especially as it relates to agriculture and advisory services for farmers. It's around localization to understand the context of how these models are gonna operate and the value that they can deliver and making sure that that context is integrated into the design.
• 5:26
  Dr. David Bergvinson: It's around delivering these, uh, outputs so that they're accessible and understandable to farmers so that they can act on the recommendations or they can access a market. Uh, it's around being inclusive in the access of these tools. So this involves not just the knowledge that's being delivered, but it's at a price point that farmers can afford.
• 5:46
  Dr. David Bergvinson: It's around policies that create the enabling environment, uh, for these AI tools to be used responsibly and not betray the trust of farmers, um, as we deploy them. So all of these and other issues are interconnected to unlock the [00:06:00] full value and to preserve the trust with farmers in this rapidly evolving AI era.
• 6:06
  Dr. David Bergvinson: So the AgX AI community is dealing with all of these through a wide range of what we're calling discussion papers. You're invited to contribute to these. We welcome your input. You can see them on the, on our AgX AI website, and the link to that is provided here. And we invite you to provide comments so that we can ensure that, uh, these papers are robust, that they can stand the test of time, and most importantly, stand the test of deployment and, uh, scale, uh, to deliver these services to farmers.
• 6:39
  Dr. David Bergvinson: Uh, the AgX AI community is also looking for opportunities to develop resources in a coordinated manner, uh, so that we work in concert towards unlocking the full value of artificial intelligence to empower farmers. And finally, it's just fun to get connected to a commity- community that is very progressive and helping stay abreast [00:07:00] of the rapid evolution of AI, not just in the ag, ag sector, but also in adjacent sectors that have relevance to the agriculture sector.
• 7:07
  Dr. David Bergvinson: So we're on a very exciting and fast-moving journey. We hope you enjoy our, uh, our Grounded Intelligence podcast, but also be active in the community and, and subscribe to, uh, the AgX AI website so that you stay informed about these developments and can contribute to the design of these AI-enabled services, uh, to empower farmers.
• 7:31
  Dr. David Bergvinson: Well, I'm extremely excited to get started with our first episode of Grounded Intelligence, and there's no better way to start that with Stewart Collis, who has a rich history of working in the digital agriculture space for more than 25 years. A pioneer in the delivery of weather advisory services by offering weather grids in the company he co-founded called Aware.
• 7:55
  Dr. David Bergvinson: And now leading the efforts around the responsible use of AI to support farmers at [00:08:00] the Gates Foundation. And so it's with great pleasure I introduce, uh, Stewart Collis to explore this fascinating topic and providing an overarching framework For Grounded Intelligence as we learn together around the responsible use of AI to empower farmers.
• 8:17
  Dr. David Bergvinson: Okay. Today, I'm very excited on Grounded Intelligence, uh, to talk with Stuart Collis, senior program officer with the Gates Foundation, who is leading their digital ag advisory services and many other domains related to the application of artificial intelligence. So this is a very timely and important conversation.
• 8:37
  Dr. David Bergvinson: Um, Stuart comes with a rich history within the IT space to serve, uh, smallholder farmers in particular in emerging markets. So his history includes working with Texas A&M around crop modeling and co-founded a company called AWARE, where they delivered, uh, weather advisories through, uh, [00:09:00] observed weather grids across the world, especially emerging markets.
• 9:03
  Dr. David Bergvinson: Uh, he also worked at ICRAF, a CG center focused on agroforestry, and, uh, is currently with the Gates Foundation, leading their strategy around digital agriculture, in particular the application of artificial intelligence, a very fast-moving and timely topic. So, uh, with that, um, Stuart, welcome to the podcast, and, uh, uh, just interested to, to know some of your top-of-mind, uh, thinking, uh, related to AI and agriculture before we dive in here.
• 9:36
  Dr. David Bergvinson: What's sort of top of mind for you today?
• 9:38
  Stewart Collis: Yeah. Thanks, David, and good to see you again. Um, yeah, I mean, top of mind just is that, you know, AI is obviously, um, having a lot of impact around the world and in many new and interesting ways. And I think, uh, you know, the catalyst was when ChatGPT came out, and, [00:10:00] uh, clearly our, uh, chair of the Gates Foundation is very interested in this topic, so it's a big focus for us is how do we ensure that this technology is equitably distributed around the world, including, you know, for smallholder farmers.
• 10:15
  Dr. David Bergvinson: Yeah. I, I recall, uh, the story of you first launching, uh, a, a contractor grant in this domain, and then literally the day after, you know, ChatGPT dropped, um, sort of- Yeah. ... rapidly by a change of mindset. Um, but I guess that's sort of been the status quo, uh, since that point onward. Uh, you know, things are just rapidly evolving, um, and just trying to stay ahead of it, uh, and making sure it's used responsibly.
• 10:42
  Dr. David Bergvinson: Uh, you know, a lot of conversations around those responsible use of AI and, uh, that's what this AgX AI community is all about, is bringing together, um, like-minded people that want to unlock the power of AI but do it in a responsible manner. So, uh, thanks for your leadership in this, in this domain. I think, you know, [00:11:00] that leads, uh, you know, to one of our questions around, uh, the, the grant that you currently have, uh, with the AgX AI community and the topics that we're focused on.
• 11:10
  Dr. David Bergvinson: Can you give us a little bit more, uh, framing around what, uh, you're looking for out of this community, especially as it relates to the taxonomy and a, and a systems approach to responsible use of AI in ag?
• 11:25
  Stewart Collis: Yeah. I, I think, um, you know, when we first started looking at this space, as you said, it's fast-moving, it was new to everybody.
• 11:33
  Stewart Collis: You know, a lot of organizations came to the Gates Foundation, given our, uh, founder's, uh, background, you know, assuming that we would have some answers around how do we apply this technology in the development sector, and specifically, in my case, for agriculture. And I, I think early on, you know, we realized that everybody was learning, you know, learning quite quickly.
• 11:54
  Stewart Collis: Uh, as you said, we had actually deployed a grant right before ChatGPT came out, [00:12:00] and, you know, one of the issues that we had identified was, you know, when it comes to these advisory services, the information that farmers, you know, uh, uh, need to make better decisions to, to access services, um, to, to improve knowledge, um, was, was a challenge.
• 12:18
  Stewart Collis: And the area that we identified as a, as a real challenge is sort of like, you know, we could see that a lot of channels were available. So you had, you know, SMS and IVR services and apps, and there were many of those out there. Um, but it seemed like the research content, all of the new best practices were not flowing through to those channels.
• 12:40
  Stewart Collis: And so that was a problem that we had identified. And then ChatPT- Chat- ChatGPT came out and, and immediately it sort of showed us that there was this fantastic tool, this new tool that allowed us to synthesize knowledge and information in a very efficient way. Um, and so we were [00:13:00] super excited about that.
• 13:01
  Stewart Collis: But, you know, really at that time, nobody really knew how to approach it. And so that's why we thought it would be nice to have a community of practice around this. A lot of people learning, you know, quite quickly, a lot of new models constantly coming out. It's rapidly moving. You know, we ourselves, when we're making investments, our historical approach has been multi-year, uh, you know, long cycle investments.
• 13:24
  Stewart Collis: Um, but here we had to make a lot shorter investments because we didn't really know, you know, where things were gonna be in 12 months and even six months. It's, uh, hard to predict. So just adjusting our approach, uh, to that, and I think having a community of practice is super important for that. You know, cro- cross collaboration and learnings, uh, on one hand, but also I think the other aspect is, you know, AI requires a lot of data to really drive it, and we didn't have a lot of data representing smallholder farmers in these types of solutions.[00:14:00]
• 14:00
  Stewart Collis: And so, you know, bringing together a community around how do we share that data, build a much bigger corpus representing smallholder farmers for pre-training and post-training, uh, was really important. So yeah, I've been, been pleased with the, the progress made so far.
• 14:17
  Dr. David Bergvinson: Well, and, and you're, you're right on, on, you know, you've had to be very agile and responsive to this rapid evolution, so just how the foundation's operating in this new bold era of AI.
• 14:28
  Dr. David Bergvinson: Um, but also, yeah, there's a lot of constraints or gaps that, uh, need to be addressed so that we bring emerging markets along with us in this, uh, you know, rapidly evolving ecosystem. And, and so you've laid out really a systems approach, I think, for the community to not only address the issues of data and data infrastructure, but how models are developed to represent underrepresented communities, especially women and, and, and girls.
• 14:54
  Dr. David Bergvinson: Uh, benchmarking, so how do we compare these models? Uh, localization of not [00:15:00] just the data, but also how it delivers, uh, the insight to farmers in, in different contexts. Uh, delivery, which is very important, especially as we look to partnerships with private sector. Access to the, the policies that create the enabling environment.
• 15:14
  Dr. David Bergvinson: I mean, there's a lot of pieces to this complex system that, uh, that community is addressing. Are... You know, why is this framing so important to you, uh, as you look at a portfolio of investments?
• 15:27
  Stewart Collis: Yeah. I think it's, you know, it's probably coming from a place of, you know, lessons learned. You know, we would look back at the digital agriculture, you know, initiatives of 10 years ago, and you and I were both working in this space, uh, back then in different roles.
• 15:42
  Stewart Collis: Um, but, um, I think, you know, one of the, one of the key lessons was that, you know, look, we all thought digital agriculture was gonna solve all of our problems and, you know, all of these solutions were gonna reach farmers. And, uh, I think, um, you know, that's true. We did have some successes there with some, you know, [00:16:00] really great, um, you know, IVR solutions, SMS solutions, reaching small holder farmers.
• 16:06
  Stewart Collis: Um, and a lot of ag tech companies, I mean, it's been remarkable actually, and we've supported some of those at the Gates Foundation and, and you did here when you were as well, here as well, is, um, you know, a lot of these solutions in Sub-Saharan Africa, I think there's over 750, uh, ag tech companies.
• 16:25
  Stewart Collis: They've raised $1.5 billion over the last 10 years. Yet, when we look at the adoption of those services, been relatively low. You know, you, you have... Like a lot of farmers might register for these things, but are they actively using them? Are they actively engaging with them? And that's often less than 10% in many countries.
• 16:45
  Stewart Collis: And so I think- You know, a couple of the reasons why these solutions didn't scale. One, I think, is the, the digital rails weren't there. It, it's really hard for a small ag tech company that's gotten a little bit of [00:17:00] funding, or maybe they're funded by an NGO or a, or a philanthropy, um, to really scale their service because they don't have the underlying farmer registries, uh, they often don't have mobile internet in many areas.
• 17:14
  Stewart Collis: Um, they, they don't have soil maps, they don't have weather forecasts, and it's impossible for any one organization to build all of that. I think that was, you know, one key reason that these solutions didn't scale. The other, I think, is around the human-centered design problem. And that is, I think a lot of times we came in with a technology-first approach, uh, rather than understanding what the farmer's problem really was and whether this technology was gonna solve that problem and how to design it appropriately.
• 17:45
  Stewart Collis: So I think that's sort of the, some lessons learned and the rationale for taking more of a systems approach to this, like to really think about the end-to-end. Okay, you can deploy an AI chatbot, but, you know, is it gonna [00:18:00] be accessible in local language? Is it gonna understand local dialects? Is it gonna have, you know, local market prices?
• 18:09
  Stewart Collis: Is it even gonna have in the un- underlying model, you know, that you're using, whether that's Gemini or ChatGPT or Claude or one of the open source models from China, is it gonna have the data that represents those smallholder farmers in the pre-training? So I think we've really gotta think end-to-end.
• 18:28
  Stewart Collis: We need to think about, you know, the, uh, cost of these services. We need to think about the adoption and the active use. Uh, and we need to think about how to sustain them going forward because, you know, as we all know, AI is not a cheap thing to do. Uh, it's quite expensive. And, and how do we deliver at a per unit cost that's gonna be acceptable to a smallholder farmer or those that are funding and supporting, uh, these types of services for farmers?
• 18:58
  Stewart Collis: Ideally, it's free for, [00:19:00] for smallholder farmers.
• 19:02
  Dr. David Bergvinson: Yeah, you, you raise a lot of really important points here. Um, not the least of which is, you know, the business model for sustainability and scaling of these services, especially when you're dealing with, uh, you know, marginal farmers who, you know, don't wanna ha- don't have a lot of resources to invest in these advisory services, and yet they realize the value that they bring, especially because they're tailored to the requirements of their specific farm and driven by weather data, market data, and other considerations to position farmers for su- for success.
• 19:34
  Dr. David Bergvinson: So this is a real conundrum, I think, on the sustainability side. You also mentioned the digital public infrastructure. Uh, you know, Africa is not as advanced as countries like India, where we're drawing a lot of examples from. What's your sort of vision or path forward when engaging with countries that have, uh, relatively weak digital public infrastructure?
• 19:56
  Dr. David Bergvinson: Is there, you know, a, a strategy or plan in place to [00:20:00] help those countries think through optimal investments to realize the digital rails for, for these resources?
• 20:07
  Stewart Collis: Yeah. And f- yeah. Uh, I, I think fortunately there's been a lot of momentum behind the idea of digital public infrastructure being a, a requirement for, for countries to provide services, uh, to their populations.
• 20:22
  Stewart Collis: And India has some fantastic examples like the Aadhaar system and now, you know, the MOSIP na- uh, national ID platform, which is open source now in, I think, over 30 countries now. Uh, and so this-- there is momentum behind this, and I think a lot of, um, you know, governments are recognizing the need to invest and support these types of rails.
• 20:46
  Stewart Collis: Uh, because it just unlocks all sorts of, uh, you know, value-added services that you can build on top of, say, a national ID, uh, being able to then open a bank account and then have di-direct benefit transfers going to that account. But then when we think about [00:21:00] agriculture, how do we extend that to farmer profiles, to farmer registries is one thing.
• 21:06
  Stewart Collis: But I think it's a broader concept of, of an agri stack because you'd have lots of registries, you know, for a farmer to get a tailored piece of advice or to be able to get credit, uh, or, you know, to get an insurance product or, um, you know, be able to be connected to a market. You need a lot of additional data to support that.
• 21:25
  Stewart Collis: You sometimes need, you know, weather forecast, you need soil maps, you need, you know, transaction history about the farmer. How do you bring all of that information together for these use cases? And I think that's this concept of an agri stack, where you have, uh, a, a sort of, uh... You, you agree on these sort of protocols.
• 21:45
  Stewart Collis: Like there's a thing called Beckn protocol that underpins some of the, the, uh, DPIs, uh, coming out of India, uh, that can be leveraged in the African market. So we don't have to build it again from scratch. We can reuse a lot of these open source tools, [00:22:00] uh, that have already been deployed. Um, and I think we're working with, uh, World Bank and others, uh, on initiatives like the Digital Agriculture Roadmap to design a roadmap for how do you build these solutions in a country?
• 22:13
  Stewart Collis: How do you justify, uh, the, you know, the, uh, rationale for building those foundations through use cases? So what are the government's priorities on import substitution or food security or diversification into horticulture? And then, you know, building the digital rails you need to support those, uh, objectives, bringing in the best, you know, from other countries.
• 22:37
  Stewart Collis: Uh, and then, you know, trying to crowd in funding around the different pieces that are required, uh, because it can be, you know, quite expensive to build these things out. The, the other thing I think that's sort of interesting about the AI space right now is, you know, uh, I, I think it makes sense to build these, um, you know, agri stack deployments in country.
• 22:59
  Stewart Collis: Um- [00:23:00] But, you know, interestingly, with these new conversational interfaces where you have two-way flow of information directly to a farmer, you can actually start to build up a profile about that farmer without having to do a big, you know, government-led data collection effort. You can gradually build up a profile just like, you know, all of the, uh, Amazon or Netflix does with us.
• 23:23
  Stewart Collis: Um, but, uh, that's can be really, you know, an alternative way or in fact a complementary way to build some of that foundational information we need.
• 23:34
  Dr. David Bergvinson: Yeah, we're fortunate that India's made these investments, uh, more than 15 years ago, uh, that can serve sort of guideposts for other countries to adapt the, you know, that digital public infrastructure that is now well established and, um, apply it for their own context.
• 23:50
  Dr. David Bergvinson: Um, I remember a year ago in Kigali, uh, Rwanda, where the AI Summit for Africa was held, and one of [00:24:00] the resounding messages was, yes, uh, we wanna take advantage and apply these new tools in the AI era, but we wanted to, um, you know, have some sovereignty over the data, the models, the, the infrastructure to unlock that, um, that impact.
• 24:17
  Dr. David Bergvinson: And so I, I think that's, that's great that donors are working in concert in a coordinated manner to, uh, im- put in place that digital public infrastructure that's relevant for the African context. W- in, in doing so, what do you see as the biggest challenge in that coordination mechanism at a country level?
• 24:36
  Stewart Collis: Yeah, I mean, it's, it's... You, you can have a sort of control everything. I, I think a lot of people are really interested in investing in AI. Um, a- and I-- so I think it's, it's really about building consortiums at that national level. We're, we're sort of taking an approach of, you know, yes, people are gonna have their own reason for building out their own solution that, you know, [00:25:00] they maybe want farmers to interact with, and they may have a, a business model around that.
• 25:04
  Stewart Collis: They might be selling finance or insurance or market access Tools, input, input, self-tag, and so forth. But I think our, our hypothesis here is that, you know, it's very hard to go it alone, uh, especially given the data that's required to drive these models. And, you know, we don't believe that any one organization can really do that, um, that it will require a corpus of information that, uh, you know, you need language data, you need, um, you know, recordings of, uh, people speaking and transcripts of that.
• 25:40
  Stewart Collis: You need, you know, uh, market price information. You need, uh, all of the advisory content that is relevant to the crops and value chains of priority in that country and the agroecologies of that country. And I think it's just so much to pull together, it's ha-hard for anyone to do it alone. So that's why [00:26:00] we think, you know, sort of a consortium-led approach to this, maybe it's like a PPP type approach, where you could have a national, you know, AI-driven and advisory service that everyone has access to.
• 26:13
  Stewart Collis: And you could still, you know, provide your own services to farmers, but tap into this back end, uh, platform to, uh, to leverage all of that data collection and the fine-tuning that it requires to tailor AI models, uh, for smallholders. Uh, and I think that's another element is just the, you know, it's very hard to get the expertise to do this too.
• 26:37
  Stewart Collis: You know, to be, you know, if you're a small ag tech company, you could hire one software developer and one agronomist and come up with something that was sufficient. Uh, but when it comes to these AI solutions, it takes a lot more than that. And, you know, you can't, at least at this point, exactly use the off, uh, off-the-shelf, you know, out-of-the-box AI solutions for [00:27:00] smallholders right now because they don't have the, the, the localization and the context.
• 27:07
  Dr. David Bergvinson: I mean, the sharing of data, uh, uh, you know, we sometimes refer to as a federated data, uh, system where different organizations are contributing different pieces to that corpus of data. Um, it-- One of the challenges, it's, it's been a historical challenge for research organizations in particular, but, you know, even companies and, and governments, uh, as well, has been the incentives for sharing that data.
• 27:34
  Dr. David Bergvinson: Um, some thoughts on how we can incentivize data so that, for example, a company could share their data, but not all the proprietary data, but some aggregate of it that then delivers value to a broader, um, system?
• 27:49
  Stewart Collis: Yeah. Uh, uh, it's definitely a challenge and, and, uh, I think in the agriculture space, it's been a tendency to not share data when you...
• 27:57
  Stewart Collis: You know, I was recently at a round [00:28:00] table with, uh, some of the big ag, uh, companies, uh, of the world, um, with their chief information officers and chief technology officers. And, you know, the, the historical approach has been to, you know, build out, uh, new IP, protect that IP, and then, you know, sell into the market and try to dominate that market.
• 28:20
  Stewart Collis: Um, which makes sense. You know, that's, that's sort of been the landscape for many years and... But very little sort of sharing of knowledge between those organizations. But I think right now even they are recognizing that there are some challenges that they individually can't solve for, and that, you know, one of the other organizations might have the data they need to solve for a certain problem.
• 28:42
  Stewart Collis: So let's say a mechanization, uh, company has a lot of images of, uh, some disease or a certain crop or value chain, say your blueberries or something like that. Another company, you know, selling inputs or treatments for those problems might require that data to build a model that [00:29:00] correctly identifies that disease through, you know, through computer vision.
• 29:03
  Stewart Collis: That's a place where you could have some collaborations around sharing data and then building models. I think the other aspect is how do you, uh, again, sort of access these AI expertise? So one of the investments we have is with the Institute for Agriculture and AI in, in the UAE at the Mohammed bin Zayed University for Artificial Intelligence.
• 29:25
  Stewart Collis: And really that's about matching these AI researchers with some of these agricultural problems, um, and trying to bring the best, you know, AI researchers in the world to some of the best agronomists and agricultural researchers, uh, and, and try to resolve some of these things. On the, on the data sharing, so I think if you can sort of set up a framework like that where there's...
• 29:47
  Stewart Collis: people are getting value out of it, there's a reason to share that data. But even beyond that, I'd say, you know, we, we do need new thinking around some of the licensing and terms under which you share data. I've seen some recent [00:30:00] licenses where, you know, you can sort of clarify whether my data can be used for training a model or only for post-training, or yes, it's fine to use it for training, but you can't replicate my publications, you know, through the AI model itself.
• 30:15
  Stewart Collis: So I think getting some clarity around some of those terms will really help unlock, uh, some of this data. And I think also some of the new New technologies that maybe they're not so new, but they've been around for a while, but maybe not deployed sufficiently as a federated system. So you can still host your data, you know, within your own organization.
• 30:35
  Stewart Collis: You don't have to provide it all over to other org-- You can sort of selectively decide what you're gonna share with different organizations very easily. Uh, so I think there's some, some exciting things happening in that space. We have some investments with the, uh, CGIAR on this, with a project called Fairgrounds.
• 30:53
  Stewart Collis: Uh, and I think, you know, we're looking at some, some new approaches to building that corpus [00:31:00] and, um, you know, building some of these vector databases so that it's much easier to incorporate into fine-tuning of AI models.
• 31:09
  Dr. David Bergvinson: Yeah, no, the... I mean, as you pointed out at the, the top of the interview is the role that data plays in, you know, the, the old slogan, garbage in, garbage out.
• 31:17
  Dr. David Bergvinson: So, you know, making sure we have quality data to do the training, uh, make the refinements to these models so that they deliver value, uh, to patients generating the data, but most importantly, in generating value for farmers.
• 31:31
  Stewart Collis: Yeah, most, uh, it's interesting, like most, you know, many governments actually that we speak to about AI, you know, we start out the conversation about AI advisory, they wanna have a chatbot for their farmers or something like that.
• 31:43
  Stewart Collis: But very quickly, it gets it down into data, like that's often where the conversation ends up. And, and within governments themselves being able to coordinate, you know, they have many departments. You've got a livestock department over here, a horticulture over here, agriculture here, and they [00:32:00] have siloed data sets.
• 32:01
  Stewart Collis: And so how do you integrate all these many, uh, data sets together, even just within the government, not even talking about going out to private sector? Um, and that's where another investment we have called, uh, Open AgriNet is an attempt to do that. It's getting some traction in India. Hopefully, that's something we can reuse in the African market as well.
• 32:21
  Dr. David Bergvinson: Yeah, I think a lot of the lessons from those, those grants and adapting them to other countries is gonna be critically important to accelerate the whole process. So it's great that you've got an ecosystem of investments to help harvest the knowledge and practical application of AI for ag. Uh, you know, one of the things that I think is a real challenge too, and we touched on it briefly, was the business model for this.
• 32:44
  Dr. David Bergvinson: Any thoughts on, you know, how do we sustain these efforts after an investment is made by the foundation or the World Bank? Uh, you know, how... What, what's sort of evolving, uh, from your observation around business models that, uh, can sustain these efforts?
• 32:58
  Stewart Collis: Yeah, it's, it's [00:33:00] definitely top of mind for us because we're closing our doors in 19 years, which sounds like a long time, but, you know, we've gotta sort of be wrapping our grants up in about, uh, you know, 10, 10 to 12 years.
• 33:12
  Stewart Collis: Uh, and so that's something top of mind for us is... And I think we're sort of experimenting right now. We're, we're trying, trying a number of things. A lot of it depends on the, the specific country and, you know, what sort of budgets the government might have. So in a place like India, where the government tends to have, you know, sufficient budgets to cover these types of, you know, services to reach millions of farmers, you know, part of the question there is can we get the cost down low enough, uh, for a government to be able to budget, uh, for covering the entire cost of inference and, you know, managing the service and, and all of that.
• 33:52
  Stewart Collis: So, you know, with- and the benchmarks we have are these more traditional ICT services like SMS and IVR. In Odisha, [00:34:00] for example, we've been able to get the cost for those, you know, I would call them generation one digital services, down to about 18 cents per farmer per year. Um, now I, I think the AI solutions are gonna be more expensive than that, but that gives us a benchmark to say that the government's willing to cover that cost for up to seven million farmers.
• 34:20
  Stewart Collis: Um, one of the challenges we have with AI is that, you know, traditional digital solutions, you know, have a, um, economies of scale, so the cost comes down as you scale. It's sort of inherent in, in the design of those solutions. But we do have a challenge with inference, is that not nec- doesn't necessarily come, you know, the per unit cost per call to these AI solutions, uh, doesn't reduce with scale.
• 34:46
  Stewart Collis: It's sort of stays consistent. And so I think there's, you know, the companies themselves that are offering these AI solutions, uh, uh, uh, are maybe, you know, trying to figure out the pricing models that work, uh, at massive [00:35:00] scale. Um, there's a lot of interest in open source solutions for that reason because you can host it on your own servers and you can...
• 35:06
  Stewart Collis: But a lot of countries don't have that capacity right now. They don't have the GPUs to run these things. So, you know, in the African market, we're tending to look more at PPP models. So can we have some combination of government and private sector? Sometimes that would be, you know, sort of a co-designed, co-developed, you know- Do, do you think
• 35:26
  Dr. David Bergvinson: the telcos are gonna play a large role in the case of Africa?
• 35:29
  Stewart Collis: Yeah, I mean, you know, we've talked to them for a long time and, and still discussing with them as to, you know, is there value in them being able to provide an AI advisory service. I, I think some of them have tried to do this on their own, and they've found it too expensive for even them to do. And can we-
• 35:47
  Dr. David Bergvinson: Processors or aggregators, do you think there's an opportunity for them to offer that as a service?
• 35:51
  Dr. David Bergvinson: They pay for it but- In return, they get a more reliable, higher quality supply of raw material?
• 35:57
  Stewart Collis: Yeah. Well, that's, that's where I was gonna go next was [00:36:00] exactly that is in a place like Nigeria, we have some pretty significant input suppliers and off-takers. So, you know, your Indoramas or OCPs, Flour Mills of Nigeria.
• 36:10
  Stewart Collis: You know, these types of companies, they, you know, would like to see better quality output from farmers. They'd like to see farmers increase income and buy more of their products. So there is incentive there for them to invest in something like this. And I think, again, it's hard for them to do it on their own.
• 36:29
  Stewart Collis: You know, could you offer a national platform that they contribute something to, uh, with the, you know, idea that this is free to everybody, you know, free to all farmers, uh, but there's benefits to those input suppliers and off-takers. So we're quite excited about some of the work in Nigeria for that reason.
• 36:46
  Dr. David Bergvinson: Yeah. No, a lot of exciting times ahead in learning. Uh, you know, just to close off, uh, Stuart, again, thanks for your time here. Um, your, your vision of success five years from now. You know, you're looking at, uh, you know, a smallholder [00:37:00] farmer, a woman sorghum farmer in, in Nigeria, let's stick with. You know, what does her life look like, uh, once, uh, services are rolled out?
• 37:09
  Stewart Collis: Yeah, I mean, hopefully better. Uh, and you know, I think if, if you think about the potential here, you know, if you could have a, a true- Sort of knowledge thought partner on your farming and, and, and, you know, what you're doing. It's like you have it in your pocket, you can ask it a question anytime you want.
• 37:32
  Stewart Collis: Um, it understands you, your local language. I mean, the first time I've seen, you know, uh, women farmers engage with some of these solutions have been the first time they've been able to speak to a device in their local language. It's just remarkable. You, you just get immediate engagement. And so I think if they can trust it, if, if, you know, they're continually getting trustworthy information and they act on it and it works, then you, then you build confidence that, you [00:38:00] know, they can take actions based on this information.
• 38:03
  Stewart Collis: It's not just knowledge though, it's also access to services that they can then get access to finance that they need so... Or they can, you know, buy a product and get recommended the right product. They can order it, that, you know, they can know that it's available locally. So all of these things adding up to, you know, then be able to get more out of their farming, um, you know, uh, operation, that they're increasing their income.
• 38:29
  Stewart Collis: There's economic gain there, uh, whether that's increased productivity or reduced costs. Um, I think that's, you know, when I look back to when I first started this journey, you know, about six years ago when I first joined the foundation, I was speaking to one of my colleagues on the DPI team, and we were just asking a question like, "Why isn't there an Alexa for farmers?"
• 38:49
  Stewart Collis: Like, you know? And what we were imagining was, you know, an Alexa Dot and a group of, you know, a women's group sitting around asking it questions. Like it's, uh, sort of like [00:39:00] some organizations now have AI as part of their board. You know, it's just an advisor you can, you can sort of tap into that massive, uh, sort of wealth of knowledge accessible through that very simple interface.
• 39:12
  Stewart Collis: And I think this is gonna be the interesting journey here, is like how do we integrate these types of solutions into very natural, easy-to-use, um, workflows, uh, that probably currently exist but can be augmented with AI. Um, and I think I'd like to see that just be a, a very natural thing that people automatically go to and see a lot of benefit from.
• 39:37
  Stewart Collis: So, um, super excited about that future.
• 39:40
  Dr. David Bergvinson: Oh, very exciting times. And, uh, and, and thanks for this opportunity, Stuart, to reflect on some of the challenges, but also the, the vision and promise that, uh, this technology can deliver if used responsibly. And I'm glad you put out the word trust there, because for agriculture, that's a really important component of [00:40:00] technology adoption, is farmers' trust in a technology.
• 40:02
  Dr. David Bergvinson: So, um, again, thank you for, uh, sharing your insights with us. And, uh, yeah, let's make it real. And, uh, that's why Grounded Intelligence is a discussion around how do we build partnerships, uh, to make this bold vision a reality. in the shortest possible time. So again, thank you. Thanks,
• 40:19
  Stewart Collis: David.
• 40:20
  Dr. David Bergvinson: What is it we're talking about here?
• 40:22
  Dr. David Bergvinson: We're talking about the responsible use of AI to empower farmers to produce more, to make farming profitable, and also to preserve our natural resource base. Now, farming is an industry based on trust, and so we need to make sure that artificial intelligence is used in a responsible and trustworthy manner.
• 40:41
  Dr. David Bergvinson: And so that's what Grounded Intelligence is all about, talking to leaders in this field to come together to draw on best practices to empower farmers to realize their full economic potential. If that sounds like a topic that's of interest to you, then please hit subscribe and join us on this very [00:41:00] exciting journey as we look at the deployment of artificial intelligence to empower farmers around the world to produce nutritious food in a sustainable manner.
• 41:10
  Dr. David Bergvinson: Today, we're gonna talk about oranges and actually how they relate to models. So when you pick an orange and it's way too early, it tastes very, very bitter. And on the contrary, if you pick an orange when it's ripe and leave it on the windowsill for a month, it also doesn't taste right as, as well. And so what we're gonna, uh, talk about with models is something similar as far as they offer us the guidance of knowing when is the optimal time to eat that orange, but depending on the context, uh, that could shift either way, uh, depending on the farm operations.
• 41:50
  Dr. David Bergvinson: Well, I hope you enjoyed episode one where we explored, uh, the Gates Foundation investments in this exciting area of artificial intelligence to empower farmers. One of [00:42:00] the core themes that came out of this conversation was the importance that data plays in developing these models and refining them to better serve farmers.
• 42:08
  Dr. David Bergvinson: And so our next exciting topic on Grounded Intelligence is just gonna be on that, data corpus, these data systems, how they're used, how they're curated in order to deliver the impact that we're aspiring to. So join us on the next session of Grounded Intelligence to learn more about the world of data.