Accelerating Disaster Response with GiveDirectly's Nick Allardice - Ep. 287

[MUSIC] >> Welcome to the NVIDIA AI podcast. I'm Noah Kravitz. Our guest is Nick Aladice. Nick is president and CEO of Give Directly, a global platform that enables donors to send money directly to people who need it most. He's also the former CEO of Change.org, the online civic action platform used by hundreds of millions of people worldwide. Nick's here to talk about his work at the intersection of global poverty, technology, and philanthropy, and how Give Directly is using AI and disaster relief and other humanitarian efforts. Nick, welcome to the pod. Thanks so much for taking the time to join. >> Thanks for having me. >> So maybe we could start with a little bit of background for listeners who might not be familiar with Give Directly about the organization. I'd love to also ask you about your journey through working with technology, and platforms, and social movements, and all of the intersection points along that path. I'll leave it to you if you want to start with Give Directly, or if you want to start with your own background and work into it. >> Yeah. Happy to start with a little bit of my own background. Both my parents were social workers. I think that's a bit where I got my values from. I think I really had this sense of just extraordinary luck. Because of just the accident of where I was born, I had access to education, to health care, to security, to opportunity, and I just was always just incredibly grateful for that. So because of my parents, I think I had this sense of service that I wanted to give back, and I started volunteering at nonprofits, and I wanted to work on the biggest, most urgent problems that humanity faced. >> Sure. >> And I found a couple of things when I started doing that. I was definitely inspired by the passion, the care, the dedication that I found from people who were working in those spaces. But I also honestly was a little bit frustrated and disappointed. I found all these small piecemeal solutions that did good yes, but didn't have a meaningful pathway to scale. I found a lot of bureaucratic inertia that meant many weren't moving with the urgency that I certainly felt, that these problems deserved. And I found a lot of top down, I think sometimes pretty paternalistic solutions that were based on feelings and feeling good about what we were doing, rather than data about what was actually working. And so because of all of that, I found my way into tech, building software that empowered people to solve their own problems, and then scaling that to hundreds of millions of users. And I kind of became convinced that we could make extraordinary progress on the most important problems in the world, simply by equipping people with the resources and technology they needed to solve their own problems, rather than us solving it for them. And so along the way, definitely got obsessed with what it takes to build the organizations, the business models, the technology, et cetera, the enables like scaled change. But that's very much how I got to where I am today. And where did you grow up? I grew up in rural Victoria in Australia, a small town called Bendigo, a couple of hours north of Melbourne. OK, and so were you in Australia through this kind of formative time of starting to volunteer and experiencing everything you just talked about with ambition, meeting what happens when humans try to organize themselves? Initially in Australia, yes. But then as I started working in the space, I definitely was spending time in different parts of Asia, in particular, working across places like Indonesia and Thailand, and different supported some work in East Timor and places like that. Right, OK. And so how did this lead to change that, Org? Yeah, I mean, I had this experience when I started volunteering of working on a political campaign that was aiming to persuade the Australian government to increase the quality and quantity of its international development of its aid program. And it ended up being successful beyond our imaginations, Australian government committed next to $3 billion to Australia's aid program as a result. And I had this experience of being like, wow, policy and politics is this extraordinary scalable lever. And I really wondered, how do we take that even further? How do we enable anyone anywhere to kind of be part of and drive that type of change? And so I joined to change the world very, very early. You know, handful of employees. I was the first one of the first employees outside of the US. What year or roughly what year? This is around 2011. OK. And then over the course of the next decade, we went through all of the ups and downs of the kind of normal startup experiences. We kind of grew the team significantly, grew to hundreds of millions of users. But along the way, like the business model failed. And we had to kind of totally rebuild the technology. And the competitive landscape was changing all around us. And so I started more on the go-to-market side of things working on PR and campaigns and things like that, but then I ended up moving to technology and the product and engineering for several years and then I eventually ended up the CEO as well. Is your tech background your skill set to put it that way? Did you study technology, self-taught, kind of pick it up along the way as you were working? And, you know, I need to build a platform that can do this. I'm going to figure out how to do it. What's been your path to building technology skills? Yeah. I was very kind of self-taught in an entrepreneurial sense. A number of the things that I worked on early, I think I was kind of coming of age. I guess doing this type of work in the late 2000s. And so starting to see what was possible with technology and playing around with a lot myself, building a bunch of software. And so that part of it was self-taught. At Tangerdorg, you know, I originally took over running our product and engineering teams at a moment of crisis in the organization when we didn't have anyone else to do it. And I was leading another team in the organization and I was lucky enough to have a bunch of mentors around me who had gone through the entrepreneurial experience and scaled technology several times and kind of learned a lot on the job and honestly just loved it. And so I ended up just getting obsessed with both the processes and the technology and really, yeah. I just love it. Yeah. Fantastic. When did you leave Tangerdorg? How did that come about? I left in 2023 and the organization had grown a lot and I think was definitely, you know, we'd achieved profitability as an organization, made a big transition as a platform that initially started out as a social benefit company, raised a bunch of money from investors, scaled a lot. And then we actually transitioned into being the largest nonprofit owned technology platform to support social change at scale. And soon after that, it was like the time was right. Right. And so give directly, let's talk about that. What in in that shell, what does give directly do, how is it started, and, you know, I'm very curious to ask kind of the, there's a bad pun about the million dollar question. But, you know, how you land it on the model of donors giving directly to the people. Yeah. And then directly sends cash digitally to people in poverty and crisis and then studies what happens. Right. And so you can think of give directly as a little bit like a cross between the FinTech company, humanitarian, nonprofit, and an economics research institute. Give directly came to be because about 15 years ago, two things happened. At first is you have, it's called the mobile money revolution, starting in Kenya with MPSR, but then increasingly in South Africa and then other parts of the world, a huge proportion of the world's population leapfrogged the banking system and start transacting purely with mobile money where SIM card is the same thing as a bank account and then you can kind of transact. And so you've got an extraordinary number of people who are becoming banked for the first time. And also as a result of that can be reached entirely digitally in a way that they couldn't be reached before. And so you've got this huge volume of people, many of whom are amongst some of the poorest in the world who suddenly can be reached digitally. So that's like the first big thing. The second thing that happens is it's like the RCT revolution, RCTs, randomized controlled trials. They like AB tests and smart economists in studying development economics started to figure out ways to rigorously measure the types of aid interventions that we've been doing for decades based entirely on when we thought it would work, but not actually being able to measure whether or not it worked. And as they started running these trials, we started learning a number of things. First, a lot of what we thought worked didn't, huge bummer. And well-intentioned programs made lots of sense, but actually the end effects aren't translating to changes in people's lives. The second thing we learned is when we give people the same amount of money as it would cost to implement those interventions, more often than not, a majority of the time, just giving them money achieves better outcomes by a significant margin. And so we have this RCT revolution where we're starting to do more and more of these AB tests and learning as a result that giving cash works surprisingly well. And as a result of the intersection of these two trends, we saw an opportunity for a new organisation to be born, give directly came to be about 15 years ago. Since then, about, give directly sent a pair of billion dollars to people in poverty and crisis, and there's been more than 25 independent studies of give directly's work. We work in Africa, in the US, and emergency responses around the world. And practically what that means is that if you're in an extremely poor community or you've just survived a devastating hurricane, you might hear from us via text message or by call centre, maybe someone visiting your community and then receive a one-time transfer that aims to cattle as you out of that situation, and it's entirely unconditional. You can do it with it what you want and our studies say people spend it really well. It's amazing, you know, we not to speak for all United States citizens, right? But amazing how I think a lot of us here and in, you know, sort of first world countries to put it that way. Think about the mobile phone revolution and think about, you know, well, I have a supercomputer in my pocket. I have a camera and it's connected to social and I can do my banking and I can do this and I can do that and all of it. But as you know, much better than I, the impact of mobile phones in these communities you're talking about and just they can bank for the first time to put it that way, right? And it's just, if listeners, if you're not familiar, urge you to go do a little bit of research, it's really eye-opening and makes you kind of think again about recent history from different perspectives. Oh, 100%. And just one thing I'll add there is like think about what it means for a community that might be hours away from a market where it might take hours to walk there and be able to check market prices with a phone call or a text message or something like that for the first time. Or get weather updates in a way that you previously were totally at the, you know, kind of a victim of just whatever was going on. So extraordinary access to information. Yeah. So you mentioned this and I wanted to ask what didn't went and ruptured you. But why does giving directly work so much better, at least in these cases and give directly his case, why does it work better than, you know, kind of more traditional means of donating and kind of things being organized and administrative by a group and then handed out, I think, use the word paternalistic before. Can you dig into some of the specifics of why it's so effective? Yeah. At the core of why it works is a few things. The first is who has the most information about what is going to be most impactful for you? And I think we've all experienced kind of faceless bureaucratic systems that you're like why is it structured this way and why is it making these choices for me? And the reality is that no matter how well intentioned, a group of people trying to understand what a community might need, they just don't have the same level of information, not just about community, but about you specifically. And so you specifically are going to be able to identify your needs far better. That's the first thing. The second thing is, it's far more efficient. We can kind of get those resources to you much, much, much more efficiently. There's not kind of overhead being eaten up by consultants and land rovers and like various things all night. And so you actually end up kind of receiving more than you otherwise would if it was kind of being done in kind goods. And then the final thing, and then I think this is extremely important. The level of like dignity and ownership in being like, oh, this is for me. And so I'm going to figure out how I can stretch this doll of the furthest. I'm going to do everything I can to make sure that this is utilized as effectively as possible. And people use it in incredibly diverse ways. I was reading this story from someone who received money from us just a few weeks ago. She's the 72 year old woman in rural Kenya. And it's this drought-stricken region that every day she's working to and from the kind of water source, which is more than five kilometers away. This is like, as a 72 year old, carrying Jerry cans and so on, like this is extraordinarily burdensome. So she received a one-time transfer from us, almost $1,000, which is fairly significant amount of money. We generally do a one-time transfer that aims to catalyze someone out of poverty rather than doing ongoing. Right. So we'll never come back. And she took $1,000 and she spent almost all of it on a giant 10,000-liter water tank. And now she pays for that water tank to be filled with clean water twice a month. And then she sells that water to her local community. She sells clean water to her local community and she earns a small profit. And so there's no organization in the world that is being like, you know what we need to do? We need to send 10,000-liter tanks to 72 year old women in rural Kenya. Right. And probably because they're not run by 72 year old women in rural Kenya, right? They don't know. Exactly. And but she can identify the unique opportunity that she gets in that moment. And so she can make that go far further by providing a service to her local community. So that's kind of in the kind of poverty context, but you know, in a crisis context, because we do a lot of crisis work as well, speed really matters. And kind of getting support to people really fast really is really important as well. So digital cash is kind of uniquely powerful. You don't have, you know, goods getting stuck at ports. And then having to figure out who needs tarpollons and who needs grain. And like there's a huge coordination problem there instead, you can go directly to people digitally fast. And then they can make the decisions about like, what is it that they need? And in the process, they can kick start a local economy. And then they can like support local businesses and things like that. Right. There's a situation like that in a crisis or disaster situation. Maybe you can kind of paint a picture for the audience and talk about, you know, you mentioned how critical speed is. What happens immediately after a disaster? What happens in those first, you know, 24 or 48 hours? And why is it so critical to get to the people to the families affected as quickly as you can? Yeah. But if you're an individual or a family that's experienced a disaster of some kind, it could be your community's been hit by hurricane or devastating floods or something like that, there's in the first 24, 48 hours a few things are happening. One, you're being like, okay, if you need medical attention, how do you get it? How do you get safe access to water, to food, like the basic necessities of life? Do we need to move from your home to an emergency shelter or a refugee camp? We even migrate further. Do you take kids out of school, et cetera? And so this is a moment, not just of acute vulnerability, where bad things happening is just like really bad, but it's a moment where desperation can cause people to make decisions that make it much harder to recover long term. Yeah. Often folks are forced to do things like sell livestock or property in order to meet their immediate basic needs, they sell it at a massive loss, and as a result they aren't in a position to recover long term. And so you want to try and reach those who need help the most fastest, because then they're empowered to make decisions that are actually going to help them recover long term, rather than be forced to make these awful trade-offs that ultimately leave them trapped and unable to springboard out of poverty. Right, no, it on a much, much less intense scale, I think we've all experienced that when in hindsight, like, oh, I only did that because only made that decision because I was under such stress and kind of didn't know what to do, but knew I needed to do something, and if only I had been clearer headed, so it makes a lot of sense. And you talked about this a little bit when talking about, you don't need to worry about getting the resources into port, and then what do you do from port to distribute them. Are there other ways in which give directly his approach to disaster relief differs from traditional aid models? Yeah, I'd say the first thing I'll say is there's a lot of really important work that happens in traditional kind of aid response, but often, not always, but often. It is slow, it's rigid, and it's disconnected from what people actually need. So, as I said, shipments of tarps or grain get trapped in ports, sometimes you have this influx of free goods that flood the local markets, and unintentionally kill all small businesses, and so you have this situation, kind of what communities are like, not able to recover because they don't cut an income anymore. Many times the kind of traditional form of aid is like, either free food or vouchers. There meant to be kind of like cash, but it's cash that can only be used for food, and this is a very popular form of aid that is given in humanitarian response. Now, would you believe that refugees who are given cash vouchers that can be spent on food often sell them for half their value, because what they don't need is food, they might need shelter, or what they don't need is shelter as they need transport to get to a local hospital, and the type of needs that you have are so diverse, and the beauty of cash is incredibly flexible, it's like any problem that you have, and so that's kind of the traditional approach, and so I think our approach is we aim to be super fast. We aim to get in within days, which is extremely fast for the way that these things often work, and we just did a response in Jamaica to Hurricane Melissa a few months ago, and a month after that Hurricane hit, most humanitarian response actors were still waiting to deploy types of support, and we had been in and out within a few weeks. Right, amazing. So cash is more effective because it can be applied to more different problems, like who applied to whatever problem is the people that have. It's more efficient because you don't have to spend it all on like, you need to do the digital transports. It's very secure because we can kind of trace exactly who it's going through at an individual level. You don't have it sitting in ports for big periods of time or going through lots of middlemen, which sometimes unlock opportunity for people to just take small little bits along the way, and then the speed is just very, very compelling because it does mean that we can move in days rather than weeks or months. I'm speaking with Nick Allerdays. Nick is president and CEO of Give Directly, and he's the former CEO of Change.org, and we're talking about his life at the intersection of technology and philanthropic work, providing aid helping people have better lives, really, particularly in crisis situations and poverty. And Nick, we talked a little bit about it, but let's dig in a little more to the technology if we can. Can you talk about how Give Directly uses AI tools to identify people who need help the most? There's some other things we'll get into, but maybe we can start there. Yeah, so I think there's like three big buckets where we use technology and AI in particular. So the first is we have to decide whether to respond somewhere in the first place. And that involves an enormous set of inputs of imperfect data that is moving extremely quickly. And so actually our ability to synthesize, sort through, and then make high quality decisions at pace is significantly unlocked by generative AI, actually, as you can imagine. Is Give Directly a virtual organization? Do you have headquarters? I know you're in Brooklyn and New York. How's the organization set up? Yeah, so our global team is fairly remote. We have a few hubs. We have hubs in New York, in London, and Nairobi, the kind of three hubs, but we have a lot of remote stuff as well. So when you're talking about the first thing we need to do, it's, I'm imagining the global map with little lights all over the place, and that's your spread out, you're coordinating remotely and in person. Yeah, exactly. This is kind of like a virtual war room that's spun out. Understand all of the data that's coming in and sorting through that in making high quality decisions as a result of it is the first use case. The second, it's not very intuitive, but one of the hardest problems to solve when there has been a disaster is who do you support? How do you find them and how do you then target that group of people appropriately? Data quality in these situations is often incredibly poor. It's not like there's some government systems somewhere that has a perfect ranking of vulnerability and impact. Usually government systems have been hit hard by whatever disasters it is that has struck as well. And so we use a mixture of machine learning and AI to help us identify the most vulnerable communities that we want to go then support. And I can kind of dig into that in a little bit more detail, but that's using everything from telco data to satellite imagery and things like that. And then the final big picture way that we use AI is engaging recipients. People are actually receiving money from us who might receive money from us at scale. As you can imagine, we're trying to move really fast. We can't spend a week to train and scale up and recruit a giant call center that is capable of talking to thousands and thousands of people. And so our ability to kind of reach people in language that they can understand whether that's text or whether that's voice or things like that is also unlocked by a generative value. So I understand you're also using AI powered image technologies for things like assessing damage, forecasting, floods, and other disasters. Can you talk a little bit about how using AI and related technologies in these ways? Yeah. So one of the ways that we will identify the communities that need help is we use satellite imagery, pre and post disaster, combined with machine learning models that help identify damage assessments, looking at things like, you know, roofs and walls and infrastructure essentially. And then we'll overlay that with poverty data. So to identify the communities that are most vulnerable to then identify target geographies that we think contain the people who are most in need of support. So we've done this actually in Florida. We responded to Hurricane Ian and we partnered actually with Google.org Fellows to build a tool that combines satellite imagery, AI damage detection and poverty data to kind of identify the households and the communities that were most in need of support. To do work like that, do you need to broker agreements with governments, with private data collection organizations? You mentioned telcos. How does that work in just in terms of getting the data so you can get started? Yeah. Doing this work well requires pre-positioning, contractual partnerships, relationships, data pipelines. It's almost impossible to kind of spin these things up fast enough in response to an individual circumstance. And so, you know, as another example of one way that we do this in the Democratic Republic of Congo, in East Congo. This is a conflict-ridden region. This is one of the most unlucky places in the world to be born. There's been conflict there for the last 20 or 30 years with different rebel groups, conflict between countries, etc. And so people are regularly being displaced, having to flee their homes because of armed attacks from militias or various things. So what we've done is we've built a relationship with one of the telcos there that allows us to use a machine learning model that identifies when people have been displaced by violence and are on the run from their homes. And so we will get a trigger that says like, hey, there's been an attack in this region. We can then run the model on their data for how mobile phones are moving in that region. And we can see how there's like several thousand people who have been in this fixed location for the last two years and who show patterns of their phone usage that indicate high-level poverty and they're literally on the run and we can see it better. And we can reach out to them remotely via text message or call center and we can say, hey, we can see you've been affected by this attack. We can see that you're on the run. We're here to help. And just go through this small two, three minute enrollment process. We'll verify who you are. And then we'll get you some supportive cash that you can then use to make sure that you survive this situation. And that's something that having that system in place, it took us months maybe over a year to kind of build that level of relationship and to kind of give the data pipelines flowing and test and make debug and all the things that you can imagine. Once it's in place, we can reach people within 24 hours of an event. And that's almost infinitely scalable. And so our approach is like we're putting in place this in different parts of the world at different times to kind of make sure that we have global coverage to make sure that we can reach people, no matter the situation, no matter the circumstances, quickly, efficiently, and then our aspiration is not just to use it ourselves, but hopefully others will be able to use it as well. When you reach people, is the response generally accepting? I'm imagining, and again, thinking of my own perspective where things like information privacy, you know, it's in this context, I'm thinking what a privilege it is for me to be worried about, you know, my digital footprint, right? If someone in the Republic of Congo is on the run from an armed attack that just to place them from their home and they get a text message from give directly who I'm just going to assume they may never have heard of before, how do they respond? It's a great question and it is one of the big challenges that we need to navigate is like establishing and building trust such that people are able to engage with us. So what we find is the pattern that tends to follow is that a few people take the leap, a register, and then basically we send money to them quickly, and then they tell everyone that this is legit. It's legit, yep, yep. And a universal human way of doing things, yeah. Exactly. But, you know, this is something that we spend a lot of time iterating on, because you need to build trust with people quickly. And so that's everything from lots of tests on the language and on how we introduce ourselves and helping people understand what privacy mechanism we have in place and things like that, because it is actually something that people do care about kind of universally, although there are variations depending on the context. And we've learned a lot about what it takes to kind of build that trust. But then, you know, sometimes in the DRC, for example, we do know, we run radio ads to help people understand that this service exists. But typically there's nothing like the proving to people that exists, like sending them some money. And after that, we tend to overcome most hurdles. Yeah, that'll do it. Along those lines, are there other surprising challenges you've encountered, either on kind of the technical or operational end of things when it comes to, you know, using technology and deploying AI in particular in these real world settings? For anyone who's worked with machine learning AI, this won't be surprising, but data quality and quantity is a huge challenge. And we have enormous appetite for, you know, getting as much as we can to try and help our target in me better and to serve people more effectively. And we're operating in incredibly data poor environments often. And so we have to spend a lot of time cleaning. We have to spend a lot of time kind of integrating that. And that I would say is probably the number one challenge is getting the data sources that allow us to do this effectively and kind of make the right decisions. Transparency and trust are big ongoing issues in any use of machine learning. And now what we call AI, when it comes to making predictions and acting on those decisions and, you know, things like financial decisions leap to the head of my mind, right? I apply for a loan and is an algorithm deciding, do I get it or not, and why did it make those decisions? How does the idea of transparency and trust come into play when it comes to the predictions that give directly is acting on? Yeah, it's a great question. There's something we've sent a lot of time kind of thinking about and talking to our the communities that we work with about in terms of just make sure we understand how they think about it as well. So I'll answer in two parts. The first is when do we confidently deploy a new approach to identifying or selecting communities as an example that is a result of data? And so we will run tests to make sure that we understand how the more automated machine learning driven method performs against the kind of more traditional methods. So to give one example, we have done this in a few countries now, Togo, Malawi and Bangladesh, where we have tested anonymized phone usage patterns as a predictor of poverty levels. And so you can imagine that if you are someone who doesn't have much access to resources, you use your phone differently than if you have lots of resources, you call at different times of day, you may be accept calls, you don't make them, etc. So we get the anonymized access to that data and then we run experiments on it to understand how well can we predict whether or not these are actually people who are in need of support as a result of this data and compare it to the normal method that other organizations deploy. And the normal method is to send people in person to go household by household and then ask a series of questions like do you have a teen roof, how much salt do you have in your house, etc, etc. And those types of in person surveys are known as the gold standard for essentially identifying vulnerability. As you can imagine, these are very slow and these are very expensive. And so what we do is we kind of run these trials against each other and understand what level of predictive ability does this mobile phone data have versus their decreases in cost and the increases in speed. And how do we feel about that trade off? And only when we're feeling like it's comparable and that it can form as good as would be that would be when we would actually deploy it in a real world context. The other thing that we've done, we do focus groups and we talk to a bunch of the people who receive our money to understand just like what's important to them about this question of technology and privacy and so on and so forth. And when we talk to the communities that we work with, they will say if it helps you reach us faster, we are really in favour. We really value speed. But we want to be really confident that about privacy as it relates to our neighbours. We think we can trust you and we want to understand what the privacy protections are that you have in place at an institutional level for using this data. But what we care about most is fairness within the community with in which I kind of operate and privacy of my own data amongst kind of neighbours. And so when we do these focus groups, when we have these conversations, we're then able to take principles out of that and then apply that to how we develop our own privacy policies and the kind of protections that we put in place. You've spoken to a couple of example, several examples of crises and just empowers conditions that give directly is engaged with, can you walk us through maybe another example where AI really changed what was possible for give directly to do? One of the most exciting, I think, frontiers of this work is predicting in advance of a disaster happening that a disaster is about to happen and getting support to people before the disaster hits. Right. Okay. And so this is something that we've done now in Nigeria, in Bangladesh and in Mozambique, where we've used flood forecasting models to identify vulnerable communities that are at risk of flooding and identify triggers that indicate that there are likely to be floods in that region within the coming days and then actually got support to them days before the floods here. And so if you think of an ounce of prevention, it's better than a pound of cure. Right. Think about what it might mean for you. If you knew it was coming and you had extra resources to prepare for it, like how much further is the money going to stretch? If you're able to move your assets, if you're able to actually get to higher ground and so on and so forth. And so that's like one very tangible way and I think this idea of predictive support, we call it anticipatory action is I think one of the new frontiers that is being unlocked by you know new generations of weather modeling and disaster forecasting that is still to be clear imperfect. We have had at least one occasion where we predicted that communities were about to be flooded. We sent them money and then it wasn't them that was hit. It was like some other communities nearby. Right. And so we're still ironing out the the kinks in the model, but yeah, I think that's a really exciting frontier kind of along those lines. Then what other ways are there for tech companies and researchers to get involved and engage with humanitarian challenges that disaster prediction example is you know, it's it's a great example. I mean, it's a great thing, but it's a great example because it's so clear to kind of understand, you know, it's not just warning get out, but here are the resources so you can move your livestock, you can you know, protect your your other assets. What are some of the other ways that tech and doing good can kind of come together along these lines? Anything that is focused on accessibility and when I say accessibility, I mean in particular language and connectivity is incredibly powerful. You've got to remember that some most of the communities that we're working in low connectivity environments, maybe you get some 2G or 3G every, you know, half the day, when the wind is blowing in the right direction. So adapting the kind of technological tools that we're building for low resource environments is a huge enabler because it unlocks not just kind of access on a day-to-day basis, but it means that in contexts where power lines are knocked down or where connectivity is suddenly hurt, like that goes a long way. Language is another big one. There's a lot of people working on this. I think there could be more working on it. Next gen AI models at this point are like really highly performant in high resource languages and something that we're really thinking of a lot about is how do we get the training data and how do we mobilize the resources necessary to make sure that as AI models becoming more and more powerful, that's something that, you know, those with the least aren't left behind and that they can access those models. I also think the thing that I mentioned around just like trusted scale, understanding and working towards technology that is understandable and accessible in art from a digital literacy perspective is also I think incredibly powerful and important. You mentioned the under-resourced languages being included, maybe I don't think is a word that does enough here, but included in the next gen frontier models. Are there other gaps that you see where AI innovation, you know, not only can really help, but is really urgently needed? Yeah. I mean, the truth is we have the most training data for high resource context problems. Yeah. And so if you're a marketer in a SaaS company, you've got all the data. There's so many things to draw on. And yet, if you're in rural Malawi and getting access to an LLM for the first time, its ability to support you in things like medical decisions could be the difference between zero medical support and any medical support. For that to be unlocked, for that to actually be something that has functional use, we need to be generating the data and ensuring that the kind of fun tuning happens to make sure that these models perform in these lower resource contexts. We cannot have chat GPT medical, which I think just got launched in the last several hours. You know, recommending that people go to a hospital when there are no hospitals, there's no hospitals. We've got to have it just as performant, just as confident and accurate in supporting tropical disease, diagnosis in low resource contexts as it is in, you know, what men should do about going bald. And so I think that that set of problems, I think, is incredibly high impact. And I would say that that's true across medical issues, it's true across education. This really interesting data about how I was helping students in low-income countries potentially leapfrog what was otherwise possible because of the potential of personalized tutors, but all of this only as possible if we have the right context and kind of training data. And it is actually designed to work in those contexts. And is that the kind of undertaking that give directly other humanitarian or aid organizations can or need to start taking on themselves? Is it a matter of brokering relationships with the right tech companies to use your example, you know, calling Sam up and saying, look, GPT medical is great, but we need it in Malawi. Like, how does that, you know, how does that happen? Yeah, there's a lot of work going into that question, but I would say it is a minor fraction of the amount of resources that is going into, you know, how do we make SaaS marketers marginally more productive? And so we think about this for give directly in a few ways. We're working on the neglected language accessibility issue because we are engaging with people at scale in a large number of low resource environments. We are asking questions like, what type of data can we be capturing that actually allows us to make sure that the next generation of LLMs are performing in those neglected languages. And so we kind of see a role for us to play, that's something we're super excited to work on. It's like helpful to have call centers that are literally operating in 60 or a hundred kind of low resource languages, which we do. And so that's like one way we're working on it. Other problems that I think need to be solved in this space. I think we're in desperate need of benchmarks. Throw a cat and you'll hit 20 benchmarks. When it comes to how LLMs are performing on everything from math to speed intelligence exactly. But I think that benchmarks are a huge unlock for its ability to do everything from provide the right medical support in neglected resource environments to how it's working in those languages, to educational outcomes, and so on and so forth. And so I actually think there's a huge opportunity for, I don't know if it's the next ten leaderboards on hugging face to start benchmarking the types of use cases that can radically catapult some of those in the world with the least up to a standard of living that I think we can all be proud of. I don't know when this article was published. I read it yesterday as we're recording this early January, but about a new tool and evaluation tool that instead of measuring AI models against academic tasks, you know, GPT-5 can pass the bar, that kind of thing. They're measuring them on real world professional tasks. And so, you know, skewing towards digital tasks for many reasons I'm sure, but that's an obvious parallel to the kind of thing you're describing. Yeah, exactly right. I mean, I think a lot of people at the moment are grappling with, yeah, when do we start to see capability translate into task change and autonomous performance and things like that? And, you know, one of those frontiers is it's one thing for model to be giving the right advice. It's another thing for that to be able to be a performant in these, these low-resil environments. Absolutely. So looking ahead a little bit, and as I was preparing to ask this question, I realized that five years from now is 2031, which, so let's just call it 2030 because that's sort of mind-blowing enough to say out loud. How do you envision the future of AI and humanitarian action evolving over the next, you know, three, four or five years? I do think that this frontier around anticipatory action is a huge one. Yeah. The ability to reach people faster and faster and faster may sometimes mean before. Not to get to my narrowly report on everyone, but I do think that when it comes to it. I think in this case it's okay. Exactly. But I do think that that's like a huge frontier that I think is very exciting. And within that, getting more and more granular and precise in reaching the people who absolutely need it, because when you don't have data and when you're working in physical goods, sometimes the money doesn't go as far because you kind of have to saturate, you kind of have to reach everyone, but if you can get really precise about like who are the most vulnerable people in a community, you can actually stretch them a dollar a lot further. So I think that's one thing. The anticipatory action, the hyper precise targeting to reach the most vulnerable, and then the speed that is unlocked by this like increasing digital revolution. At the moment, that 70% of the world has access to a mobile phone, and that number is growing over and over time. And that means that it's increasingly possible to reach people in some of those hardest to reach places. And so I think that's like a huge part of the future as well. And for AI researchers, tech leaders who, you know, we have a few in the audience, I think we're listening and thinking, you know, I want to do something good with the skills I have. What do you say to them? Quit your tech job and come join, give directly, stay where you are, but work from the inside to make sure that less resource languages are included in the next frontier models. What do you say to those folks? Definitely. We're hiring all the time. So I'm not going to say no to that. And I do think directly is somewhat unique in our ability to create an environment in which people with deep technical depth can apply their skills in incredibly important globally scalable ways. And so, you know, we have a highly technical team and kind of working on this all the time and definitely welcome people to join and reach out to learn more about that. I do think that within whether it's labs or research organizations or tech companies in general, I think one language accessibility is a sure bet, a enormous impact across a huge number of people that I think is just really underappreciated by a huge number. And so I think that's a big thing. Accessibility in particular, making sure that things are performing in lower connectivity contexts. Correct. It's hugely leveraged impact right there. This is kind of a sort of a little bit of I don't naive or sort of tangent question, but I have a background in covering the mobile phone industry is some particularly interested. When you're talking about some of these areas, you know, you mentioned like you might be on 2G, 3G for an hour if the wind blows the right ways. What kinds of devices are do people have? Are they, you know, what we think of as future phones? Are they smart phones? Is it SMS? What are they working with? Yeah, it's predominantly, you know, the updated cheaper versions of the Nokia brick phones that many of us were using 20 years ago. So they're home phones. But recent generations of cheap smartphones are getting increasingly powerful. And so in a number of countries, you're starting to get quite high adoption of smartphones in a way that is giving access to the internet and modern apps for the first time ever. And doing that for $15 or $20 for the kind of total cost of a smartphone. That's amazing. Yeah. Nick, there's so much to say. But I think it boils down to, you know, thank you and keep going with the amazing work. And along those lines, for listeners who want to find out more, maybe engage directly, maybe give the website givedirectly.org. Are there other places, social media? Where can they go to find out more, follow you, engage with the organization? Yeah, definitely givedirectly.org and definitely will, on the shame at least, say that if folks feel so inclined, dominating goes an extraordinarily long way. It's really incredible how much difference $500 makes for a single person and cash transfers is one of the most scalable and effective interventions out there. Connect with us for me on LinkedIn. We're kind of regularly, you know, we kind of take the philosophy of like building public. And so we are often talking about the hard trade-offs that we're navigating and the kind of ups and downs along the way. And really encourage folks to kind of follow along. Great. Nick, Aloe days, thank you again for taking the time to come onto the show and all the best to you and your team. Thanks so much for having me know. [Music]