← This Week in Global Development
This Week in Global Development
Global Progress in the AI Era: How GiveDirectly is using AI to deliver cash faster
00:42:57
In this episode of Global Progress in the AI Era, Nick Allardice explains how GiveDirectly is using AI to deliver faster, more direct aid — from anticipatory cash transfers in Bangladesh to near real-time responses in crisis zones. The conversation...
Loading summary
Transcript37 lines
- [00:00]
A
Foreign. Defining forces of our time. It's reshaping how money flows, how services reach people, and who holds power. I'm Katherine Chaney, senior editor for special coverage at devex, and this is Global Progress in the AI Era, a special edition podcast series exploring what happens when AI collides with the world's biggest challenges. In each episode, we speak to leaders from philanthropy, government, civil society, and the private sector who are responding to this technological turning point. They'll break down how AI could unlock breakthroughs in health, agriculture, and education, and what it will take to avoid the risks of deepening inequality. We'll tackle difficult questions and spotlight promising ideas that will determine whether AI accelerates global progress or leaves more people behind. Artificial intelligence is already reshaping how global development works. And that means that some of the most important questions facing leaders today are no longer theoretical. They're operational. What does it actually look like to use AI as part of humanitarian response? How do you move faster without sacrificing trust? And what happens when technology starts to reshape not just how aid is delivered, but who gets it and when? Today, we're joined by Nick Allardyce, who is leading one of the most ambitious efforts to answer those questions in practice. As CEO of GiveDirectly, he's helping build a system designed to deliver emergency cash to people anywhere in the world within days of a crisis and increasingly, even before disaster strikes. From flood forecasting in Bangladesh to rapid response systems in conflict zones, this conversation explores how AI is changing the space, speed, the scale and the philosophy of aid and what it means for the future of humanitarian response. So, Nick, welcome to the podcast.
- [01:54]
B
Thank you for having me, Katherine.
- [01:55]
A
Thank you. So I want to start very simply. I've spoken with some of your colleagues about this, so I have a sense, and it's part of why I was eager to bring you on. But I want to hear from you. When you say GiveDirectly is using AI, what does that actually mean in practice?
- [02:11]
B
And we are using AI across almost everything that we do. But there's a few core use cases where I would say it's particularly important. And the first is in our humanitarian work and as context. GiveDirectly does work in development contexts, and we also do crisis response, humanitarian work, and increasingly so, I think increasingly so. It's a huge area of focus for us, and we've delivered, maybe we've delivered about a billion dollars in unconditional cash transfers over the course of the 15 years that GiveDirectly has been around and the whole way through, we've been very, I think technology enabled. The whole reason that givedirectly exists is the advent of mobile money and the ability for us to extremely efficiently reach people in poverty without needing to have an extraordinary number of kind of intermediaries or people on the ground or things like that. And so that's always been core to the model. But as we have lent more and more into crisis response, we are using AI at every step of how we respond. That means, and maybe to step back. Our goal is to get emergency cash within five days of any crisis to survivors of crisis anywhere on Earth.
- [03:37]
A
I know you describe that as sort of a moonshot goal for the organization.
- [03:40]
B
I think that's right. You know, I think there's a lot of been a lot of talk about, like, what a humanitarian reset looks like, and like, how is it that we make humanitarian aid work in a more efficient, accountable, dignified way for beneficiaries. And for us, we see a huge opportunity for cash to be at the core of that because it is so efficient, it's so dignified. Like, the evidence is that it works, but there's a lot of technical problems that need to be solved in order to make that possible. Like, what does it actually take to get digital cash into the hands of people anywhere on Earth within five days? And the answer is that actually AI is a big enabler of that. And so we think about it at a few layers. First, we think about it when it comes to how do we identify vulnerable communities that are most in need, and how do we do that quickly? That's the first problem. Second problem is, like, once we identify them, how do we communicate with people in ways that they can understand, engage with, trust our relationship with in a way that allows us to then enroll, verify, and, you know, make the decisions necessary? And then how do we actually, like, get the cash into their hands and track and make sure that it's happening in the way that it does. And so we use AI across each of those different steps. I'm happy to kind of dig into each of them, but at a high level, that's everything from flood forecasting models that allow us to identify kind of communities before they're hit by disasters, to kind of recipient communications that is using AI to essentially aggregate incoming information from people on the ground and make sure that we can process that faster and then communicate with them in ways that they can understand and things like that.
- [05:23]
A
Yeah, I would love to ground this in a specific example, just to help people wrap their heads around it. And we actually had a reporter recently travel to Bangladesh and see some of this work up close. I think that's a really interesting example where some of this is happening. So can you talk us through kind of in those three categories you mentioned? What does that look like right now in Bangladesh?
- [05:45]
B
Yeah. So in Bangladesh we're doing, I think, a really exciting kind of pilot to where measure the value of speed is at the core of what we're trying to do there. And I think it makes intuitive sense that if you're experiencing crisis, the faster you get support, the better. But how much should we value it? What's the value that we should place on the ability to reach people within 24 hours of crisis versus 7 days of crisis within 2 months of crisis? How much better, better kind of outcomes do you get for people when you're able to move at that speed? Or even how much better outcomes do you get if you're able to reach them before the crisis itself? And so in Bangladesh, unfortunately, it is just entirely predictable that there's a huge number of communities that are affected by pretty devastating floods every year. Now, which communities in which ways is uncertain. But that there will be floods in most years that affect an extraordinary number of extremely poor people is something that is unfortunately very predictable. And so we have pre positioned a set of infrastructure that allows us to predict in advance of floods happening that certain communities are going to be affected by those floods. And we partner with google.org and a bunch of kind of machine learning models to help us identify this. And we have pre enrolled huge numbers of people across these communities so that if a flood forecasting trigger goes, we can actually get cash to those communities in advance and say make your homes more resilient. Now is the time to move your livestock. Now is the time to kind of take precautionary measures. And a whole, whole theory here is that an ounce of prevention is worth a pound of cure. But we are going to be running some trials on this to actually prove that that's true. And so we have a few different arms of an rct, a randomized control trial that is going to. Different communities are going to be getting cash essentially at different speeds. Some pre crisis, some immediately post crisis, and some like normally on the normal timeline that you would expect. And we're going to be able to kind of see exactly how much value can we get by kind of pushing that speed barrier.
- [08:07]
A
One question I have for you in terms of challenges that come up is when I think about communities like those in Bangladesh who Could most benefit from this kind of anticipatory cash. There are basic, probably technological infrastructure challenges, things like access to smartphones. In fact, the reporter that we had in Bangladesh said in the community she visited, about half of the people she met had smartphones. And so can you just kind of talk us through how do you overcome that as what I would assume is a big barrier?
- [08:38]
B
Yeah, the about 70% of the world now has mobile phones of some kind. And that number is just growing every year. It is not yet like we cannot yet reach every vulnerable person through a mobile phone of some kind. But that world is not so far off in the future. And so the approach that we are taking is like we want to skate to where the puck is going to and say as more and more of the world is connected, we want to have a solution that is extraordinarily fast and efficient and effective. That does mean we need to pay extra attention now to who might get left behind by that approach. Now, some of the ways that we're thinking about that, first of all, in most of the communities that we're doing this work, smartphone usage is even lower than what you're describing. In Bangladesh, 50% penetration of smartphones is actually pretty decent. And, and so we primarily optimizing for feature phones and giving people opportunities to enroll and verifying who they are based on feature phones rather access rather than smartphone. Now smartphone makes it easier actually because we can, they can take photos of IDs, we can do like all sorts of other bits and pieces that otherwise wouldn't be possible. But the first answer is build solutions that work for feature phones, not for smartphones, and build solutions that rely, that don't rely on high connectivity or anything like that. The second answer is spillovers. So we have done a lot of research on our programs in a development context that establish pretty strongly the positive spillovers of cash transfers, not just to recipients. To give an example, there was a. We did a five, seven year long study on a series of cash transfers that we did in the late 2010s and what we were able to establish through some pretty granular economic measuring across multiple communities is that people who lived within half an hour of direct beneficiaries benefited all almost as much as those who got the direct cash from a consumption perspective, from an outcomes perspective. And that's kind of a crazy, crazy thing, right? The idea that someone who lives within half an hour can benefit just as much as someone who like receives cash directly. And that's because one of the unique things about cash is that it recirculates within the economy, like people spend it locally, it kickstarts those economies, there's more money, there's more support. And so where one of our big areas of research is like, how do we design our cash transfer programs and the amounts that we're sending in order to maximize those spills, spillovers. Because then that allows us to reach people who may not have the technological access to be able to kind of do things in the way that you're describing.
- [11:33]
A
That's a great point. I actually want to. When you mention positive spillovers, it makes me think of a question I wanted to ask you based on an interview I had with one of your colleagues. So I was interviewing your colleague who leads this emergency cash work and is kind of working toward that moonshot we talked about earlier. Um, and one of the things we talked about is how the humanitarian community tends to be more false negative prone and givedirectly is more false positive prone, meaning you'd rather include some people who might not need help, as much help as others, rather than miss people who do need help, that is attention that maybe givedirectly would be more comfortable with. And a donor, say, that is using taxpayer dollars might not be able to do as easily. Can you kind of talk us through that tension and how, how is AI kind of bringing this tension to the fore?
- [12:27]
B
Yeah, we take a default saturation approach to the communities that we focus on. And so to give you a tactical example, let's say a hurricane hits a community or say like Mozambique or Jamaica or these are both places that we responded in the last six months. Our approach to responding in that context will be to take poverty data, which we have some combination. We have access to government data, but then we also use satellite data to kind of forecast and predict poverty levels and deprivation levels across different communities. Then we overlay that with disaster data to basically say, okay, we can see based on pre and post satellite imagery that kind of destruction has been particularly high in these areas. And so taking the combination of poverty and vulnerability data and disaster data, we then overlay those to say which communities are both most vulnerable and have been most affected by this, by this disaster.
- [13:27]
A
And just very quickly, Nick, I want to let you finish that point. But mobile phone data is also part of this mix as well. Right. And future phones can still tell us quite a lot.
- [13:34]
B
Exactly. We kind of build pretty in depth partnerships with mobile network operators that allows us to forecast poverty levels based on people's usage of their mobile phones. And that's all Anonymized. And that's all like, you know, we spent a lot of time working closely with the mnos to make sure that that is kind of done in a kind of ethical and secure way. But that does allow us to get very pretty granular on which populations are most vulnerable and have been most affected by a crisis. Once we have identified those communities, we try to reach everyone in them. And the reason that we do that is, number one, as you say, we want to make sure that no one's left behind. And so often when you're trying to take this person by person approach, the people who are least able to participate in these processes, maybe they're elderly, maybe they're disabled, maybe they're kind of vulnerable in other ways. They're going to be the ones who are going to actually be hardest to reach in those contexts because they're least able to kind of access the, maybe the kind of points of contact or anything like that. So we take a saturation approach. We try and reach every single person in the community. We accept that there are going to be some people in the community who may not have had the same levels of vulnerability as everyone else, but we think it's worth it for a couple of reasons. One, it means that no one's left behind. Two, when you step back and look at what is the ROI of precise targeting and you say, how much money are you willing to spend in order to kind of make sure to minimize false positives? And I would hazard that the vast majority of time, all of the effort that is spent by the kind of humanitarian community trying to get as precise as possible about the exact beneficiary lists, the amount of cost, both in money and time to actually making that happen is kind of vastly outweighed by the amount of money that you could be getting into people's hands faster. And so we've kind of very much taken this approach of keep it simple, move quickly, reach whole communities, we think that overwhelmingly that's going to have the most impact for people.
- [15:53]
A
Hi, I'm Kate Warren, Executive Vice President and Executive Editor at devex. At devex, we don't just cover the biggest moments in global development. We create space to understand who and what are driving the headlines. Alongside gatherings like the World bank and imf, Spring and annual meetings, the World Health assembly, the UN General assembly and beyond, we host Devex Impact House, where our journalism comes off the page page and onto the stage. We bring together a curated group of leaders for live interviews, intimate roundtables, hands on workshops and candid conversations you won't hear in the official meetings. It's where tough questions get asked, the spin gets stripped away, and meaningful connections happen. If you'd like to join us or stay in the loop on all of our events online and in person, please visit devex.com events I think that's very interesting. And one thing that really struck me in talking with one of your colleagues is sort of the dignity angle here as well. So when you are talking about all the effort spent to get as precise as possible, essentially what that can look like. And again, this is Leith Baker on your team, who I was talking with, is going door to door trying to get a sense of how poor are you really? I mean, is essentially the question, and what does that feel like on the other end versus, you know, we're going to go with the risk that this reaches someone who doesn't need it as much as someone else, but could definitely still use it. When you look at how vulnerable these communities are.
- [17:24]
B
Totally, totally. I mean, we often what this looks like, right, is when you are going door to door, you're. You're making these proxy decisions like how much salt is in the house, what type of roof did this person have before? That is always going to be imperfect. You're always going to make mistakes. And the incentives of both sides of that transaction are so just like awful when you think about it, when you think about it from the recipient's kind of perspective and experience. And so, yeah, I think we are really kind of strong believers not only in this kind of saturation approach for efficiency and being able to spend more money on recipients rather than on us, like doing precise targeting, but also for dignity that, like, this allows more people to participate with higher dignity.
- [18:11]
A
I think that's really interesting. I also wanted to go back to something you mentioned earlier. You mentioned this RCT in Bangladesh, and one of the themes I want to get into in this podcast series is not just what AI programs are being rolled out and what impacts they're having, but how those programs are being tested and what the learnings are and kind of the evaluation side of things. I know that's always been core to how GiveDirectly operates, but can you talk us through the why of that RCT and also kind of the why now? Because I would think that especially in a time of donor dollars drying up, the learnings from an RCT like that might not just inform GiveDirectly's work, but say the government's work and how they spend those precious dollars. So I Would love to hear more about your thoughts on the importance of testing.
- [19:06]
B
Yeah. GiveDirectly was founded by four economists and we really, I think, emerged in reaction to, and then grew because of a lack of measurement in the sector more broadly. Feeling that there's a. There's a lot of money and effort being spent on things that seemed good in theory, but that when you really dug in, in practice, we're having null outcomes at best and sometimes like even worse. And I think we just have to do better. Like it is just, like it is just unacceptable. Especially at a time when donor dollars are contracting to spend money on time and time on things that aren't evidence based. Now, that doesn't mean you can't try new things. It's just really important that you kind of measure them when you do. And so we've since our founding had this huge emphasis on measuring our own work in order to improve our own work, but then also publishing that and doing that in public in a way that can inform and hopefully inspire others in the sector to kind of follow suit and to learn from some of our learnings in. And as part of that, you need to be judicious in what you choose to test because like, RCTs are expensive and they take time to set up. There are real costs and real trade offs associated with doing them. But when we're moving hundreds of millions or billions of dollars through systems that are kind of making a default assumption that it is kind of fine for aid to reach communities two months after a crisis, and then there's a lot of effort going into this new model of anticipatory action, which is like, hey, we can actually reach communities in advance. Like, how much more impactful is that? We've got to be able to put a number on it. And so that's been a huge emphasis for us as a, as an organization. And our kind of aspiration has been to build our own capability to be the best organization in the world at partnering with researchers. We don't do the research ourselves in general. We'll partner with third party institutions to kind of run rigorous academic research on our programs and then build a learning loop such that we're constantly reinforcing that and kind of putting it back into the work that we do.
- [21:30]
A
And do you anticipate? I mean, in the same way that I think GiveDirectly's goal has not just been to do the work it does with cash, but to influence others to think hard about cash, as he's done with donors in The UK and the US I would think that also with this anticipatory cash work, there's what givedirectly can do and then there's the influence you can have. Do you see that governments could start to move in this direction or how? I mean there are real constraints in them doing that too. So where do you think that that's maybe a more difficult leap?
- [22:05]
B
Yeah, it's very. We see a lot of hunger from governments actually. And the barriers are often, not always, but often not will or interest, it's technical capability or infrastructure. And so our work on using machine learning to identify vulnerable populations and things like that, this was first catalyzed back in 2021 as part of our Covid response. And we were working pretty closely with the Togo government to say, you know, huge Covid crisis, lots of shutdowns, the normal methods of going door to door to identify, you know, poor communities and poor households just was not possible for public health reasons. What are our alternatives? And so it was this kind of three way partnership, but between the government of Togo, the major telco there, and us, that was about that. We kind of ran this big experiment to say could we get as good as the kind of gold standard household door to door measurement of identifying vulnerability just through mobile phone data? And the answer was we could get there and we could do it cheaper. And so that really kind of catalyzed a set of this work for us. Our goal with this infrastructure that we're building for ourselves, the ability to reach people within five days of crisis anywhere in the world, that's infrastructure that's not just for us. We want to prove that it works for us. We want to build it, we want to be using it regularly, but we intend to make it available for others. And we already see a lot of actually inbound demand for that we are working with. We're doing a kind of response in Mozambique at the moment to a recent hurricane. There's and the government has basically had a huge amount of enthusiasm to say like, you know, we would love to be able to reach these populations quickly, we would love to be able to get emergency support to them. Sometimes it's as part of, you know, the kind of social protection ministry, sometimes it's like part of their kind of disaster response management but they don't have the infrastructure to do it. And so if we can get those tools in place, then we think that it's going to be kind of hugely influential, not just for us, but for the sector and for governments as well.
- [24:24]
A
Yeah, that's so interesting. And I wonder if you can kind of tell me what success looks like there, like your dream scenario. My understanding is, you know, right now it's sort of disaster strikes. People arrive on the scene and try to figure out how to help versus what you're describing is more like before disaster strikes, actually people already have the support we need. It just completely reverses that dynamic. So in terms of what that would really look like, again, in talking with Leith Baker, my understanding is just to kind of wrap my head around it. He described it as like a portal of sorts that has information on, you know, where those in need are, how to reach them. Is that how you envision it? Like essentially a portal that partners can plug into?
- [25:14]
B
Yeah, exactly right. It's a, it's. Think of it as a dashboard that is as simple as possible for us or another NGO or a government and ministry to say we can identify the target population and all we have to do is like wire money from one place to another and we can have confidence, we can kind of see every step of the way how that money is flowing down to the community. We can see it at a recipient level and we can see what data is being drawn on and we can kind of set our own guidance for what transfer sizes we want to use and all of those different things. So yeah, that's very much the vision here. To give you an example of what I think is possible, we do a lot of work in the drc. In places where there's a lot of rebel activity, there's a lot of conflict. And I think that the existing humanitarian infrastructure, I think on Average often takes 100 plus days, sometimes 130 days in order to reach people with cash or in kind support post displacement. So we're talking about communities here who are displaced by violence. There's been an attack of some kind. They fled their villages and it takes somewhere between 100 and 130 days to kind of get support to those people, if it ever reaches them at all. We have been operating in this environment for a while and we've kind of built a set of partners and technical capabilities that allow us to get to identity, to use cell phone data to identify displaced populations within 24 hours of those populations kind of fleeing violence, reach out to them remotely. Sometimes this is via SMS messages, but we don't rely on that as well because often literacy rates are really low. And so we'll have call centers that call people and rad messages that broadcast how people can kind of get involved. We have them go through super simple enrollment and verification mechanisms that take like a couple of minutes at max and and we can get cash to them within 24 hours of a displacement event versus the 100 to 130 days that is happening more, more generally. And it take took time for us to build that infrastructure, but now that it's in place, it's extraordinarily efficient, it's extraordinarily effective, it's extraordinarily fast. And my goal is that we have that infrastructure everywhere and that we are not the only ones who benefit from it. Other organizations, governments and so on can draw on that infrastructure as well.
- [27:52]
A
Absolutely. And I think another important point to bring up, which I know GiveDirectly has been sort of amplifying, is when you talk about this MoonShot build a 5 day global emergency cash system, it's not just about responding to what we would think of as one off crises, because increasingly, and we talked about earlier, increasingly Give directly is working in these contexts. Well, increasingly that's going to be much of the world when we think about climate shocks, conflict, instability. So it's going to be an around the world round the clock situation, I would think.
- [28:29]
B
Totally. I think you and your audience would be very familiar with the stat that by 2030 the vast majority of the world's extreme poor are going to be in fragile contexts that are disproportionately affected by climate shocks, by conflict and things like that. And so that's a big part of why we, as an organization that has historically been very focused just purely on development, has started leaning much more heavily into this work. Because if we want to take seriously our mission of reaching people in extreme vulnerability and poverty and catalyzing them out, and we've got to be operating in these environments that are subject to these types of crises.
- [29:04]
A
Absolutely. I want to hear from you as CEO of GiveDirectly, what this looks like operationally, especially because so many organizations in our space know that when it comes to seizing the opportunities with AI and mitigating the risks with AI, it's going to have to change how they hire, how they fundraise, how they operate day to day. So can you take us through that? I mean, I know you mentioned the work in Togo as an example of kind of some of the earlier work, and of course it's exploded from there. But what has this meant for you as CEO to kind of lean into AI and how has that changed GiveDirectly's operations, team structure, fundraising? I kind of want to pull back the Curtain for our audience a little bit, yeah.
- [29:51]
B
I mean, the first is definitely talent. And I think Give directly has always had a bit of a bias for kind of fairly technical talent, but it is even more true now than it ever was. We, I think, compared to many, hire pretty significantly from the kind of private sector. We have a lot of people who come from mobile network operators and telcos. We have a lot of people who come from tech companies and things like that. And so baseline technical capability is expectation at the organization. It's something that we've kind of factored into our interview processes. It's something that we, you know, I know that a lot of organizations are kind of grappling with. How do you, how do you handle the fact that AI has made work tasks like, much harder to kind of deal with as part of hiring processes? Because everyone's just using ChatGPT, kind of do the prompts. And we're like, we encourage it actually, we encourage the use of AI in the kind of interview process with us. But we want to see your prompting, we want to see your work, we want to see how you are leveraging these tools. And so we've kind of made some of those shifts. The second thing is I'm a really big believer that the most interesting stuff for the use of technology, but also AI and this type of work is often going to come bottom up, not top down. And so for me, that's meant deliberately creating space and encouragement for the whole organization to experiment and to plaque and to move quickly in the way that they do that. And so some of the specific actions that we've taken, one, it's just like, yeah, provision enterprise versions of like all the major tools to basically everyone in the organization as baseline, streamline kind of approval, such that if anyone wants to kind of experiment with new tools, that that's a decision that can get made and unblocked within the day. And there's not like long processes that kind of slow everyone down. But then we've done like, we've done orgwide hackathons where we literally say, you know, everyone in the organization, we're going to, we're going to clear these two days, every meeting is going to be cancelled. And the encouragement is that you do something with AI to accelerate your work that could be automating a workflow that could be building something new. And so we did one of these last year and the only requirement is that you demo something at the end of the two days. And so we have like everyone in the organization, like demoing some of the things that they built and what they're using. And then we pick the best ones, we celebrate those on the term hall, we have prizes, we kind of put a whole vibe around it. And so that I think has been, I think really important because the space is moving so fast that my ability or my executive team's ability to identify like what are the best ways for us to use this and what's it going to take for us to seize it, I think are really limited. And what we actually need to do is create a culture that embraces it and empower our teams to actually take advantage of it and to propose solutions. And so that has been a big area of focus as well.
- [32:59]
A
I wonder, you know, even as you are encouraging the organization to embrace AI for all the reasons you laid out and in all the ways you just described, what are the risks that concern you and the areas where you think givedirectly needs to tread very carefully and how do you ensure that the team is thoughtful about that as well, in addition to leaning into the opportunity?
- [33:20]
B
Yeah, there's kind of a few different use cases. Right. So there's internal tooling, which is kind of a productivity thing. And so in that case we want to have some boundaries around data sharing and around kind of privacy protection. And so we've done some pretty intensity intensive. Just like let's make sure that that's never something that is a kind of problem and that is being taken quite seriously. But apart from that, when it comes to internal tools and internal tooling and workflow automation and things like that, I actually think that the kind of risks are fairly minimal and we should be pretty encouraging of experimentation. When it's more recipient facing or kind of beneficiary affecting, I think the stakes are much higher and we need to have a much higher bar for what it takes for us to do this type of work. And so anything that involves supporting decision making around targeting of communities, for example, that is something that we will like rigorously test against what the existing standards are. And so the only we only started using mobile network data to predict vulnerability. Once we had run a series of experiments that established it was performing as well as the kind of standard practices. And we kind of had the data to back that up and got increasingly confident about it. As we deploy it in more communities and as we use it in different ways, we are kind of constantly running these focus groups with recipient communities to understand what's important to them. When we are balancing different trade offs, what are the most important things to those communities. And it's really interesting talking to folks in extreme poverty about what the different trade offs might be and how we can build that into the principles that we use to guide the work. Guide the work. And so to give a couple of examples here, when we were doing one of these, I think it was in Malawi, the message that we heard kind of loud and clear from folks that had experienced kind of crisis was if it helps you reach us faster, we are so insupportant, you know, like it really like speed is very important to them. And so they're like, yeah, that is a top righting priority. But when we talk to people about privacy, you know, it is valuable for them to understand and trust us as an institution about like what different data gets used and how that's involved. And we have to get really good at explaining even how all of this works. We have to go through a lot of iterations to be like, how do even help these communities understand what's happening in the background here. But once we do, we're able to hear that the value that those communities place on the saturation model that we were describing. A lot of the risks that people see or were concerned about were about kind of disproportionate treatment within communities. And so the fact that we were kind of going to everyone in a community was like a significant mitigant for folks in how these types of tools were being used to interact, enable decision making. And they all wanted like really high confidence that as it related to their neighbors and their friends and family, that it wasn't going to be possible for other people to kind of understand or predict or kind of make judgments about them as an individual based on some of the things that we were doing. And so privacy within communities is obviously something that's very important to people as well.
- [36:43]
A
Nick one example I want to make sure that we get to is actually some of the work you're doing in Malawi, which I find very interesting when it comes to pairing cash transfers with agricultural advice. And this kind of gets into, in our last few minutes, just any big messages you have for the sector as they look at the AI opportunity. My understanding with that work is, is this idea that AI generated advice, AI powered advice is only really going to create value if people have the resources to act on it. And that's where GiveDirectly sees an opportunity. So can you quickly tell us about that work and how do you see that as connected to maybe a larger message you might have for the sector on finding valuable use cases for AI
- [37:30]
B
the AI, I think, has the potential to democratize access to information and decision making in ways that can be transformative for those in extreme poverty, whether it's healthcare, whether it's agriculture, whether it's education. I think the types of use cases are, I think, really promising and they're kind of meaningless if people don't have the actual resources or ability to make different choices. And so putting people in a position to seize the opportunity of AI. I think that there's one world where AI creates unfathomable wealth, largely in the global north, makes a bunch of white collar marketing workers more productive, maybe causes some labor disruption, but that overwhelmingly both the benefits and the costs are kind of concentrated in rich countries. And I think there's another world where we help communities get to kind of basic flaws of opportunity where once they have a small amount of resources, combine that with all of the knowledge of human history and like the best kind of decision making support in the world, and it can be a real catalyst for people to kind of have durable improvements in their quality of life. And so we're really trying to push towards that second world because I don't think that it is at all inevitable that that's what will happen. And we need to make sure that these kind of AI use cases that support people in the highest levels of need are at the forefront of how we work. I think in Malawi, what that has specifically looked like is supporting smallholder kind of agriculture work with decision support, being able to kind of take photos of crops and ask for fertilizer advice and things like that. I don't yet know if that's going to be the use case. That is really a big unlock. Like I would say, like when we talk to people, they say that they value it. But when I look at the usage data, I'm like, yeah, I don't know, like they're not yet kind of utilizing it at a level in which it gives me confidence that we're having the kind of level of impact that we want long term. But I think there's enough there to be promising.
- [39:47]
A
It sounds like maybe an example of a lower risk experiment to see exactly
- [39:53]
B
one of the really interesting qualitative pieces of advice that we, or qualitative bits of feedback that we've gotten about how people in extreme poverty make use of these tools. We hear from a lot of people that they really value the independent advice, that they kind of have fear about talking to people in their communities about how to invest money. They fear being exploited, they fear being judged. And so on and so forth. And so having a kind of neutral arbiter who they don't have to worry about, to kind of talk things through with is something that in these pilots, recipients are saying that they really value. And I think that's a really interesting insight for what this can help unlock in the places that we work.
- [40:37]
A
Nick, I know we're about out of time, but just a final question. If you had one piece of advice for another CEO in this space, kind of looking at AI and trying to figure out their next steps, what would your one piece of advice be?
- [40:50]
B
Lean in. I think, like, you know, this is really a kind of unimaginable opportunity for us to be more efficient, more effective, and support more people more effectively as a result. And I think that that's a really high responsibility. I think it's going to be very easy for the sector to be like one of the slowest adopters in the world on this kind of stuff. And I think that it would be a total tragedy. That doesn't mean there aren't really important problems to solve in the way that we do it and really thoughtful ways to kind of manage and mitigate the risks and all of those things. But we have such a high responsibility to make sure every dollar goes as far as it can and to make sure that people are getting the support that they need in the context of a system that is just abjectly failing right now, that we have to take advantage of these types of kind of technological revolutions to reimagine what's possible.
- [41:55]
A
Well, Nick, thank you so much for taking us into how GiveDirectly is doing, just that. We appreciate your time.
- [42:01]
B
Thanks for having me.
- [42:02]
A
So much of the conversation around AI and global development focuses on what might be possible. What this conversation makes clear is that many of those possibilities are already here and they're forcing real choices about how fast aid should move, about who gets included and who gets left out, and about whether systems are designed for efficiency or for dignity. As Nick described, the technology itself is only part of the story. The harder challenge is how institutions adapt, how they rethink incentives operations, and ultimately what they're optimizing for. If you found this conversation useful, you can subscribe to the debex podcast and sign up for the Devex newswire for more conversations like this. Thanks for listening.