Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Nick Allardice on Using AI to Optimize Cash Transfers and Predict Disasters
0 Plays2 seconds ago
Nick Allardice joins the podcast to discuss how GiveDirectly uses AI to target cash transfers and predict natural disasters. Learn more about Nick's work here: https://www.nickallardice.com
Timestamps:
00:00 What is GiveDirectly?
15:04 AI for targeting cash transfers
29:39 AI for predicting natural disasters
46:04 How scalable is GiveDirectly's AI approach?
58:10 Decentralized vs. centralized data collection
1:04:30 Dream scenario for GiveDirectly
Recommended
Transcript
Introduction to GiveDirectly with Nick Allardyce
00:00:00
Speaker
Welcome to the Future of Life Institute podcast. My name is Gus Docker and I'm here with Nick Allardyce, the CEO of GiveDirectly. Nick, welcome to the podcast. Thanks for having me, Gus.
Unconditional Cash Transfers Model
00:00:11
Speaker
Maybe you could give us a brief intro to what it is that GiveDirectly does.
00:00:16
Speaker
So GiveDirectly helps, lets people send money directly to people living in extreme poverty. They're unconditional cash transfers to the extreme poor, no strings attached. And this is a fairly, I guess, somewhat revolutionary way of doing aid or charity or international development, however you want to talk about it, because one says, you know what, where we being the global north, being governments, being NGOs, are probably not best placed to decide what is going to most serve any given individual community in what they need at any given time.
Efficiency and Impact of Cash Transfers
00:00:58
Speaker
and so The unconditional direct cash transfers allow people in poverty to choose for themselves what's most important for them. The second thing about why it's like pretty revolutionary is
00:01:09
Speaker
it kind of cuts out a lot of middlemen along the way. So I can click a button here in New York and watch exactly how money flows through all the way down to an individual in a rural community in Malawi. And there are no intermediaries who are kind of taking little cuts along the way. And so it's an extremely efficient way of actually getting money into poor people's hands. And that obviously allows us to stretch the money further. And then the final, I guess, big shift that I would say is it's extremely scalable. like there There are these many
00:01:45
Speaker
development interventions highly bespoke that take huge amounts of design, planning, a lot of complexity involved. This is because it's so efficient, because it's so technology enabled, it's extremely scalable. And it I think has a really plausible chance of like dramatically accelerating the end of extreme poverty for a really large number of people as a result.
Spending Autonomy and Paternalism Concerns
00:02:05
Speaker
So that's give directly at a big picture level.
00:02:08
Speaker
So what about something like, the first first question that comes to mind would be something like, if we talk about a hypothetical future, a universal basic income where you give it to everyone, you might worry that some of that money and would be spent on wisely. And of course, there's an argument to be made that people know best what's what's most important for them to purchase. But you could also worry that some of the money would be spent on things that might not be the best.
00:02:36
Speaker
Do you track how the money is spent or what do you do to ensure that the money is actually spent well? We track pretty closely. We we kind of do follow-ups with everyone who receives money. we We both ask them what they spent it on, but we also do kind of surveys to observe what happens, not just taking people at their word. and Overwhelmingly, what we find is that the money does not go into what we would call temptation goods, alcohol, cigarettes, things like that.
00:03:03
Speaker
Overwhelmingly for people in extreme poverty, it goes into getting enough food to eat. A lot of these communities are very fit, food insecure. It goes into improving your basic living conditions, like making sure that you have a roof over your head that doesn't leak, things like that.
00:03:18
Speaker
and Then it goes into productive assets, it goes into livestock, it goes into small business starting, things like that. and so That's the observed results from now doing transfers to more than one and a half million people over the years, both the observed results that we've seen and kind of independent academic studies that like look at the same things. Temptation goods, it just doesn't happen.
00:03:41
Speaker
The thing I'll say though, sometimes that question comes from a place of... you know It comes from a good place. You want to make sure that money goes ah fur the furthest it can in supporting people who have not won life's lottery about where they're born. I think it does sometimes come from a place of paternalism as well.
00:03:59
Speaker
Having autonomy over our ability to like live life at its fullest is actually a really important value. and so even if the data didn't say ah Let's say that the data said that a kind of small percentage of the time people are spending some money to have a drink with friends or to celebrate at ah at a community level or whatever it was. like I wouldn't say that's a bad thing. I would say that's a good thing because people like you know deserve to live their full lives and make choices for themselves about what's most important to them. Yeah, that makes a lot of sense to me. Can you say something about how GiveDirectly is different from traditional aid organizations?
00:04:39
Speaker
Yeah. i mean I think first and foremost, it's that we trust people. We like we say we are not best placed, they are, to make decisions about their lives. and That's kind of built into the bones of the organization in a way that is really top to bottom fundamental.
Effectiveness of Lump-Sum Payments
00:04:57
Speaker
We can talk a little bit more later, I'm sure, about how that shows up in things like making decisions about technology and AI and various things like that because we do really consult and talk to and put recipients at the core of everything we do. That's the first thing. Another way that is different is we're extremely data driven. I think we're founded by a bunch of economists and from early on said we should be using data to work out what works and we have this hypothesis that
00:05:25
Speaker
not only was it more respectful of people's agency to give them unconditional cash, but also it was probably going to be more effective and more efficient in actually alleviating poverty. and so We've just run this extraordinary number of randomized control trials now and have kind of integrated data and tracking and third-party evaluation as well into all of our models. I would say as an organization, we're very research forward We partner with a lot of academic institutions to make sure we're getting independent studies of our programs and that data really does shift like how we behave. An example actually would be that we have the longest running trial on UBI in the world. It's like we're running that program right now. We're running a 12 year program in Kenya.
00:06:10
Speaker
where people are getting monthly monthly payments for 12 years. And so there are three arms to this trial. There is one arm where people got the same amount of money all at once, as they would as if they had like a three year UBI payment. There was another arm of the trial, which was a monthly payment for three years. And then there is a third arm of the trial, which is a monthly payment for 12 years. And so we're kind of looking at the data, what happens between each of those different kind of tests. And after the first three years, we're halfway through that trial right now. After the first three years, the data was fairly clear that a one-time lump sum payment
00:06:50
Speaker
was more effective at most of the indicators that we cared about than spacing those payments out monthly over a three year period. and So we're still waiting for the results of the 12 year trial, but for the comparison of like, okay, we could take $1,000 and split that up into small dollar payments over the course of three years, or we could just give them $1,000 upfront. Which one is actually going to have the better results? It's pretty clear to us that it's actually giving them $1,000 upfront.
00:07:17
Speaker
And the reason for that, or at least like what seems to be true from looking at data from our perspective is it allows people to make more substantial productive investments. It's like rather than having small amounts that kind of just get incorporated into your monthly or weekly budget and often end up getting spent on consumption, like it gets spent on food or things like that. Instead, people are more deliberate about how they kind of divide that chunk of money up.
00:07:46
Speaker
and More of it ends up going into productive assets like livestock or small businesses and so on and so forth. and so It ends up generating more more return for the person over time to get it all at once. and so As a direct result of that, we had a planned a ah UBI program in Liberia.
00:08:04
Speaker
And as a direct result of that, we're kind of pivoting away from that. We're saying, you know what, like we're actually going to double down on these large one-time payments because that's going to have more sustainable impact for the people involved. Because they'll have the money to buy tools or supplies or something to set up small businesses, for example.
00:08:23
Speaker
Yeah, exactly right. and you know maybe Maybe they want to buy a sewing machine or a motorbike. like This would be pretty common. it's like If you want to set up a textile business where you're trying where you're kind of making clothes for people, a sewing machine might cost a bunch of money. It might cost like six months of payments. You could try and save up that or you could just spend it all up at front and then like make use of it over time.
00:08:44
Speaker
and And so that that I think we see pretty clearly in the data that if what you want to do is allow people to have a kind of step change in their ability to kind of earn income and to more kind of
Flexibility and Empowerment through Cash
00:08:57
Speaker
sustainably exit like the poverty situation that they're in, then a one-time larger payment is better than smaller payments over a longer period of time.
00:09:06
Speaker
Do we know why it is that people are smart about how to spend their ah the money they get? Do we know why it is that they might be better at spending the money than, for example, an organization deciding to buy sewing machines or motorbikes or something like that for them? is it Does it have something something to do with local knowledge or constantly changing factors and in local economies? What do we know about that?
00:09:32
Speaker
Yeah, the flexibility of cash, like the fact that it's so fungible, you can apply it to so many different problems and opportunities is kind of extraordinary.
00:09:43
Speaker
And so when you think, yeah it's it's almost like a planned economy versus market economy question. It's like if you go into an individual community and say, what's the optimal number of goats that this community should have? And how many taxi businesses should there be? And how many bakeries? How many like corner stores? One.
00:10:08
Speaker
The likelihood that any outsider is going to get that right, I think is is really low. like Let's be honest. like I wouldn't trust someone to kind of make that judgment. but The second is like the amount of effort that you would need to go to, the expense involved to actually try and get some sort of informed answer to those questions.
00:10:28
Speaker
is like kind of extraordinary. So you end up wasting all this money that is well-intentioned, but you kind of have like these consultants and you have bureaucrats and you know program design people creating these log frames and creating like doing these community consultation processes. And like all of that is very well-intentioned. like and And some good things can come out of these processes. But when you think for a second about the amount of money and time that goes into those processes, it's kind of extraordinary.
00:10:56
Speaker
And like what if we just took all the money that we were spending on those processes and just like gave it to people? Could they make better use of it? I think the answer is like pretty unequivocally yes. Now, they won't always make perfect choices. like We all make poor choices at times. There are sometimes people that I will engage with in our programs who I'm like, why did you do that again? or like I'm not sure I would have done the same. and Sometimes so with benefit of hindsight, you're like, wow, you really knew what you were doing. and Sometimes you're like, you know what? like Maybe they did make the wrong decision. I want to give one example here. There's one one story I'm thinking of which was a person in one of the communities we were in who wanted to become a musician and basically used some of their transfer to buy a bunch of musical instruments.
00:11:37
Speaker
And from outside, it might be like, you know what, like, is that the best use of this money? Like, you might you have to kind of work quite hard not to not to judge or to try and provide some guidance about that. That might not be the best thing to do. But then we go back six months later, 12 months later and They're touring around nearby villages. They're making quite a lot of money from planning at community events. They're now like this feature and the local scene, and it's this huge kind of sort source of sustainable income that would otherwise not have happened. and What's the likelihood that any one of us would have come up with that as a particular business idea for this person? Absolutely zero.
00:12:15
Speaker
you know and so i think that the People know better. There's less wastage. And then the final thing, which I think is really important, is the level of ownership is just very different. Like, if.
00:12:28
Speaker
I came to you and said, you know what, what I really think you need for your like life is to actually live in a slightly larger house and and to maybe by upgrade your audio equipment. And so as a result, I'm going to give you some better microphones. How much ownership would you have over like how to use those things on on making the most of those investments? I think it would be pretty marginal. Whereas if on the other hand, I said,
00:12:53
Speaker
here's some money. like I want you to use this to make the program that you run as like best possible. ah You're going to have so much more ownership over the outcomes that you're driving. and I think that really matters. For some of the but people that we work with, you know this this is in more money than they'll see at any other time in their life. like We're talking about kind of one to two years worth of income all at once.
00:13:16
Speaker
and They take that responsibility incredibly seriously because
Self-Sufficiency and Income Thresholds
00:13:20
Speaker
we're very clear. We will only overcome once when we kind of give that one like large lump sum. They spend a lot of time thinking about it. They're like, hey, like I've kind of just won the lottery and I i want to use that very very well. and and so They're very thoughtful and they put a lot of time thinking into it. They talk to all their friends about it and there's a lot of community conversation about it as well. and so I think that's part of what improves the decision quality as well.
00:13:47
Speaker
Do you find that there are threshold effects where once you get over a certain income or net worth or any number like that, you become more likely to be self-sufficient and can can kind of function more independently going forward?
00:14:04
Speaker
Definitely. i mean I think what we hope to do is to give people a one-time shock to the system. It's almost like you know trying to kickstart the local economy with a giant in injection of demand that then allows a kind of sustainable reinforcement of that money over time and kind of allows people to kind of increase their their wealth and to put leave poverty behind and to not require that in ongoing ways. What the exact threshold is for a person or community, like it really does differ because like you could imagine that a young person who's kind of quite entrepreneurial may not need as much of a jumpstart as an older person who's lost their partner, who's got dependents to look after. you know They're ultimately going to need different things, but communities that we work in
00:14:53
Speaker
Our whole goal is to essentially kickstart that local economy in a way that creates ah a positive flywheel and that then allows that community to be self-determining going forwards.
Incorporating AI into GiveDirectly's Operations
00:15:03
Speaker
That's wonderful. Okay, let's let's get to what I would consider the core of the interview, which is about how GIF directly is using AI in its operations. So when did GIF directly begin thinking about AI and and and ah how have you incorporated it so far, if you can give a kind of broad overview?
00:15:23
Speaker
Yeah, so the high-level overview, gift directly has always been very technology forward. i The thing that enabled us to really grow as quickly as we have and to exist in the first place is the advent of mobile money and basically kind of bank accounts on dumb phones in Africa has like allowed us to exist. And so we've always been very technology forward. And I think over time,
00:15:49
Speaker
We've basically looked at more and more how different ways we can use technology to become more effective and efficient. And and AI has played a role in a number of different ways there. So the high level, the different places that we use kind of AI right now. So one would be on targeting.
00:16:07
Speaker
Who should receive money? Now, that may seem like a somewhat straightforward question to ask, but it's actually very complicated to answer. And so we use AI in a range of different contexts, both disaster-related context as well as poverty-related context to basically say, who really needs this?
00:16:25
Speaker
And so we might integrate with cell phone providers or governments to basically analyze mobile phone usage in order to, with machine learning models, so in order to say which people are kind of poorest in the community and therefore need support most. And so we actually did a big partnership with the government of Togo during COVID to use machine learning, machine learning models to identify the most vulnerable people in society and get cash to them fastest.
00:16:51
Speaker
Yeah, so how how does that work? Because how do you identify poverty just given a mobile phone usage? Yeah, so I think this is this is a super interesting area of research. And we've kind of partnered with academic institutions on this as well. How this would be done historically, or what the kind of normal mode of doing this might be, is you might have government ah registries, you might have census data that often, frankly, is like imperfect or out of date.
00:17:21
Speaker
because those things happen very infrequently. So that's like one source of potential data, targeting data. Another source of potential targeting data is literally people in the field, like going door to door, basically using proxies of some kind and being like, this house has a thatch roof, and that house has an iron roof. Therefore, we think that thatch roof is 70% more likely to be below a certain property line. And so there's this very manual data collection method.
00:17:45
Speaker
and This new method is essentially saying we can use machine learning-based models on cell phone usage data to give everyone a likelihood score about being below a certain like poverty threshold. and It will look at things like How often is the phone used? What is it used for? What times of day is it being used for? How often is it recharged with like credit? And you can kind of take all of those inputs and then basically predict. And and with a large enough data set, you can compare it to reality, right? You can kind of hold some part of the data back and you can say, how much better is this at predicting relative poverty within the the audience compared to the other methods that we have? And so in the example of of Togo that I gave,
00:18:31
Speaker
We actually looked at the data and I actually noted this AI-based method was actually substantially more accurate than the alternatives, accurately identifying who really needed the support. and That's like one of the principles that we use in deploying these kind of technologies is that we'll always be benchmarking against the best alternatives.
00:18:49
Speaker
and so we only want to be moving forwards with them if we can say this is measurably increasing the quality or speed with which we're able to kind of do this work and hopefully once we're able to prove that then it's also happens to be more scalable and and efficient but yeah that's kind of how we would think about that.
00:19:07
Speaker
It's very good that you're benchmarking against alternatives. I think that's ah something that happens often in the business world is that you will they will not benchmark against alternatives. and But it's AI and so it's interesting and new and therefore it's good. But that's that's great to hear. This is using AI for good. But if I had my phone data used to find out but how a wealthier poor I was, that that would make me slightly and uncertain. And so what are you doing to to ensure that ah mobile phone users in these areas that you maintain privacy for these people?
00:19:40
Speaker
Yeah, totally. So as we've done more and more of this, we've spent some time putting together our own like ethical use of AI kind of policy and approaches. And what I would say is there's there's some good frameworks out there in the world to draw on. And it is, I guess, our experience has been that too often they're not super practical. There's there's a lot of philosophizing about what the right kind of ethical frameworks for these things are. And we would love to kind of see it grounded in reality a lot more in terms of how you actually do it. So how we have chosen to do it.
00:20:10
Speaker
is set of principles i would encourage folks to read it's all on available on our website but really important for us to be talking to our recipients directly about what is it that. You care about here what are you concerned about what are you excited about and so on and so we actually just earlier this year in february of this year we did a series of focus groups across different context in which we work.
00:20:32
Speaker
to identify what is it that people think about us using these models? What are their concerns? What are they excited for? And what are they not? And so part of doing that involves us having to make explainable everything that we're doing. So there's an explainability piece, which is like, how do you just make what you're doing accessible to the kind of general public, but also the people who are either benefiting or not from the models that you're doing? And so we had to go through a lot of iterations to be like, how do you take the idea of kind of machine learning based mobile phone, cell usage targeting, how do you make that understandable for a rural community in Malawi that has up until this point only had cell phone penetration of 20%?
00:21:17
Speaker
And that's a hard challenge. But like we went through a bunch of sho iterations on that. And so we've come up with ways of explaining that now that makes it actually much more and understandable to people. So that's the first thing. The second thing is you talk to them about, OK, what is it that they actually care about? And what are they concerned about? And we had a couple of takeaways from that process.
00:21:36
Speaker
The first is what they care most about is speed and accessibility of the cache. It kind of came through overwhelmingly strongly that if this enabled us to move faster or be more accessible for them, then they were like enthusiastic because that is the kind of primary concern. And sometimes we're operating in natural disaster contexts or in places where there's conflict and speed really, really, really matters. And so that just kind of came through overwhelmingly clearly.
00:22:06
Speaker
The second thing was, they had some concerns about privacy, but it was specifically in the context of neighbors and people they knew. so They actually expressed like reasonably high kind of trust in us um when we explained, okay, we do encryption, we have all these securities in place to kind of make sure that that your data is protected. But what they talked about in a very practical way was, the thing that we're concerned about is my neighbors knowing about what my cell phone usage is And so that's the kind of thing that we want to put priority on when it comes to protecting privacy. And obviously that's something we had kind of thought about, but it is helpful for us to know where it's most important for us to double down and where it's most important for us to proactively communicate and to solicit feedback and things like that.
00:22:53
Speaker
There's a few other kind of takeaways, but there's some of the big ones. and you know In general, this process of ah making it understandable, directly consulting, and then feeding back into the day-to-day usage of models is you know part of the muscle that we've kind of built over time and I think is incredibly important.
00:23:11
Speaker
the The third one that also comes up, I just remembered an important one, was inclusion and equitable access is like a really big deal for us. and That also came through from ah recipients. If using this type of targeting would result in more people being able to more fairly get access to funding, then that's something that everyone was really enthusiastic about. If on the other hand, there would be parts of the community that were excluded,
00:23:40
Speaker
or who would not otherwise have access to things then that was actually a point of significant concern. And
Community Economic Impact of Cash Transfers
00:23:46
Speaker
I think this is actually really important principle for when you're trying to leverage AI for good is that like data is often yeah unclean imperfect and and so we we do have now these like,
00:23:57
Speaker
models for kind of ensuring enrollment or ensuring we're reaching people who we otherwise wouldn't have data access for by sending out field teams and by kind of making sure that there are like people to manually engage with those who might otherwise be left behind because it may be more efficient for us to just go, oh, there's only some people that we can get data for. likeke Let's just focus on doing doing it for them.
00:24:22
Speaker
Yeah, that's that's also a way that so a project like this could fail if you're optimizing for whoever has the phone and whoever is presenting you with the data. and and But of course, you're you're looking beyond that. Can you infer anything about the people who do not have cell phones from the people who do?
00:24:39
Speaker
I don't know the answer to that, to be honest. you know In general, the way that we choose communities to operate in, I mean, we have a bunch of different programs, but for our flagship program, we will choose communities based on We want 85% of the population to be below the extreme poverty line, but then we will send money to everyone in that community. and The reason that we do that is, one, it's actually just like much more efficient. and If you're not below the poverty line, then you're only marginally above it. and so
00:25:11
Speaker
There's people still need help. But two, there's actually a pretty giant multiplier effect. By operating at a community level, you kind of stimulate the entire economy. And so there's like substantial positive spillovers from the fact that there's just more cash in the economy more generally. And maybe you could ah maybe you could give examples of that of those spillover effects. So how how does that work in practice?
00:25:31
Speaker
Yeah, so think about it. If you if you own a small business, and ah let's say you're a taxi driver, like your motorbike, or maybe you're you're kind of sewing clothing, or maybe just selling farm goods at a local market, if suddenly everyone in the community has a year or two's worth of income all at once, the demand that is going to come to your business is going to be significantly higher. And so that money will recirculate through the economy significantly.
00:25:59
Speaker
And so what we find is there's the value of ah sending the money directly, just like let's say that's for every dollar that goes in. You see $2.50 of economic activity generated as that money continues to circulate. And so that's actually where a lot of the positive impact actually happens. Like, yes, it's about the direct money that gets sent, but even more so, it's about the money that then circulates repeatedly within the local community in a more sustainable way.
00:26:26
Speaker
Would it change anything for GIF directly if the poorest people on earth had smartphones instead of dumb phones? um Maybe that's a naive question, maybe it's not happening anytime soon, but with a smartphone you have access to so much more information, you can input and output much more frequently and efficiently and so on. Would that change anything?
00:26:46
Speaker
It's a really good question. And it's something that we're actively exploring. So some of the places where we operate, some people already do have smartphones. So sometimes we are working with urban refugees, people who've kind of fled conflict and who've migrated to say Nairobi, slum slum in Nairobi or something like that. And it's more likely in those types of urban contexts that people will have access to kind of very cheap smartphones.
00:27:13
Speaker
In more rural ones, we're actually planning a pilot next year at the moment where we give people the option for very cheap smartphones along instead of very cheap dumb phone. We're going to see what happens and see what that increasing access to information could unlock. You know you could imagine you know other applications of AI in this space that are about you know, business coaching and other things like that that that might be unlocked. And I think the short answer is we don't know yet what the kind of level of of impact of that might be, but it's something we're very keen to study.
00:27:50
Speaker
Do you think there would be any change in how the money is transferred at a low? So I don't know how much overhead you're incurring when you're sending cash, but I i would expect this to be a metric that you're trying to optimize for or minimize. And so maybe maybe having a smartphone would open up ah different ways of of sending cash that are not available on dumb phones. That's a guess. I don't know. but You tell me.
00:28:13
Speaker
Yeah, so at the moment for our flagship program, we we target 85 cents in the dollar, like reaching the recipient, which is I think actually quite different to how other organizations would calculate overhead. Many organizations, they will say overhead is anything we spent on the central central organization, and then anything that's not overhead is the doing the program.
00:28:38
Speaker
What that hides is doing the program usually costs a lot of money. and so If you're buying a Land Rover and sending people to the community to kind of consult with them or to build the well or whatever it is, that will be calculated as not overhead, but it actually still is not necessarily reaching the recipients. and so For us, when we say 85 cents in the dollar should reach it's to the recipients, that's basically everything end to end. like it's It's all going to land in that person's hand.
00:29:06
Speaker
other opportunities unlocked by smartphones? Maybe. I mean i think that that some of the expenses involved would include you know withdrawal fees from the the kind of telcos that are doing the mobile money payments. and That's kind of often worth a couple of percentage points. and so You could imagine that with smartphones and apps and maybe some some types of kind of different digital payments, you might be able to kind of squeeze that a bit.
00:29:31
Speaker
On the whole, I wouldn't expect that there's like dramatic improvements that we could find. It would probably be more in the like small percentage points, would be my guess. So you mentioned AI for poverty detection or for targeting these cash transfers.
00:29:48
Speaker
But you also use AI for helping with and of more emergency situations, natural disasters. well How is AI useful in in those situations? Just a few months ago, maybe six weeks ago in Nigeria, there was like huge floods that happened that led to kind of significant displacement and affected a huge number of people's homes. And so what we did in advance of that is so we actually worked with, we actually partnered with Google on essentially flood forecasting. And so we're able to say, we think that there is going to be a flood event in this community in the next two to three days.
00:30:29
Speaker
And so rather than wait for the flood to hit and then send people support afterwards, what if what if we actually sent them some money in advance? And that allowed them to make their homes more resilient. It allowed them to move their livestock. It allowed them to maybe kind of move to higher ground or stay with family or or whatever it was. And so our hypothesis is that you know and and an ounce of prevention is worth a pound of cure.
00:30:57
Speaker
And we can use AI to essentially forecast when they're going to be adverse like events, climate events, flooding, that kind of thing that are going to significantly affect people. And then we can get in before it happens and as a result, reach more people, do it more efficiently and allow people to kind of suffer less through that process.
00:31:16
Speaker
So, we've done that in Nigeria, we've done that in Bangladesh, and it's a program called Anticipatory Action. It's like anticipate the action and then kind of take it in advance. um And we're actually running an experiment on this in Bangladesh at the moment where we're basically working with some third-party academics to study for people who received the cash in advance of the floods compared to people who received the cash after the floods.
00:31:43
Speaker
what happens? like What are the difference in outcomes? And if we can show through data that our working hypothesis that getting cash in advance is is far more effective, if we can show that that's true, then that's something that would be really exciting to scale up as well.
00:31:58
Speaker
That's one of the main ways that we're doing it. yeah i mean Just to stay on that point for a bit, it it seems intuitively true that it would be good to give the money and ah in advance. But you you would be have you'd have to trust your predictive model a lot, I would think. How good is modern machine learning at predicting these events?
00:32:18
Speaker
It's a really good question. I mean i would say that they're good without being perfect and it's getting better over time. and one of One of my favorite, actually, examples is the first time we tried to do this and give directly one of our kind of values as an organization has been transparency since day one. We want to be super transparent about like the things that work and the things that don't.
00:32:37
Speaker
and A couple of years ago, the first time we did one of these and anticipatory action programs was was in Mozambique. We had teed it all up. We had all of the models. We were ready to go. and Then the triggers came through that said, there's going to be flooding in these communities. and We sent cash there in advance.
00:32:55
Speaker
and Those communities didn't get hit by floods. It was other communities nearby. So it's it's it's it's doubly negative, right? Because not only you haven't targeted the right community then, but but you've also kind of sent the money to... Yeah, if you send the money to the wrong place, you both have you both have ah money where it shouldn't be, but also you lack money where it should be. So it's it's it's a very high-stakes situation, it seems.
00:33:19
Speaker
I think that's i think ah I would say yes and no to that. like Yes in the sense that you know if we had run that propp perfectly, I think the money would have been targeted to those other communities that were ah that were affected. No in the sense that this is a highly flood-prone region where the vast majority of people are living significantly below the extreme poverty line.
00:33:40
Speaker
And so I was like, okay, so the worst case scenario here is that a bunch of people who are facing really significant challenges and who are regularly affected by floods have received some money. And I can feel really good about that. Even while I still say, hopefully we can get this program to be even more kind of efficient and targeted over time and kind of reduce that error rate so that it gets even even more targeted. So you know I would say that this is ah not a perfect science yet.
00:34:08
Speaker
but it is one that alongside everything else in AI is just getting better every day. You mentioned that GiveDirectly runs these studies with outside academics. Studies are expensive, right? Isn't it the case that just running a study to get the information is is very expensive and time-consuming and you have to go through a long approval process and so on? This might seem like ah a funny question, but do you have how do you evaluate whether to do a study? How do you evaluate what that information is is worth to you that you would potentially get from the study versus the cost of the study itself?
00:34:43
Speaker
That's a great question because yeah they are pretty intensive process processes. I think about it on a couple of levels. The first is maybe I have a starting disposition that It is incredibly important to stay humble and learning first in pretty much everything you do. I don't think anyone has the right answer at any given time. and One of my big critiques of the rest of the sector and other people doing this work is that there's a lot of fill opinions and not enough data. and so I think starting with a disposition towards kind of measurement and learning is, I think, just an important starting point.
00:35:20
Speaker
But then once you have that starting point, there's still trade-offs and how do you make those? I think the first is, what's the counterfactual use of the money? And in most of these circumstances, we're actually not the ones providing the money. Most of the time, it's this kind of academic institutions that basically a bunch of social science researchers and economists who have these like incredibly incredibly ambitious plans to understand how we can do better in development interventions and otherwise, but who lack testing partners with the operational capability to do it so do it at sufficient scale. And so we can unlock a lot of research simply by orienting our ah ourselves around being a good partner for those types of institutions because they actually have the resources and the capability already to do it, but they actually don't have enough organizations that are good enough at working with them on it or that are at the right scale to do it effectively.
00:36:15
Speaker
so There's a question of counterfactual money, but i I do think that we need to set a pretty high bar on, are we learning new things? yeah it's It's not enough to just say, let's just run another trial on this like because of expense involved, because of effort involved, and because it you know it does often involve holding back control groups as an example.
00:36:38
Speaker
where there's groups of people who will not receive money for some extended period of time so that we can actually see what the difference in outcomes are. And I think it's important to take that kind of responsibility incredibly seriously and to have a really high bar for the learning outcomes that we're kind of going for when you're making those decisions. In general, though, I would say that it's only in the last 10, 15 years that there's been a really increased focus on figuring out what actually works. And so there's an extraordinary amount of unknowns still. And so it's rare that I would say that there's not significant things to learn.
Optimizing Cash Transfers with AI and Technology
00:37:19
Speaker
To give you one example of ah of a recent thing that we studied or looked at was if just the timing of transfers could make a significant difference for people receiving them. And so these are very agricultural communities.
00:37:32
Speaker
And so the agricultural season is a huge driver of the economy locally. And so we wondered if we could time the transfers to be consistent when it is most useful to have extra capital to invest in the kind of agricultural productivity of the local community. Would that increase outcomes for that whole community?
00:37:54
Speaker
And the answer seems to be yes, in a fairly meaningful way, that if you have time transfers to be maximally useful with agricultural season, then it can make the dollars stretch 10% or 20% further than they otherwise might have. And so like these kind of what what might seem like pretty small or narrow questions actually can have fairly significant outcomes for how far the dollar can go and in helping people.
00:38:18
Speaker
Is there an explanation for that? Did you in that same study find out why it is the timing matters, even holding the the dollar amount or the cash amount constant? and The hypothesis would be that the more of the dollars go towards productive investments that have immediate return.
00:38:38
Speaker
And so kind of similar to the point that I was making earlier about UBI actually, where if at the moment you're deciding how much fertilizer am I going to buy? What are the tools that I have in order to increase my agricultural yield? If at that moment in time you have a full bucket of money with which to decide how to use it, then the likelihood that you spend more of it more intelligently or or smart more smartly to kind of maximize your your agricultural yield goes up.
00:39:05
Speaker
I think you had other examples in mind when we were talking about using AI to intervene during disasters or perhaps right before disasters. Yeah, I think the another big one is is conflict, actually. So when you look at where are the world's extreme poor, close to 50% of the world's extreme poor are in what we would call fragile contexts. Fragile contexts are countries that are affected by conflict.
00:39:32
Speaker
These are places like the DRC, the Democratic Republic of Congo, or the Central African Republic. They are often incredibly climate-prone as well, like a lot of climate disasters happening in these places. and so Every year, almost 30 million people are kind of pushed back into extreme poverty as a result of disasters or shocks.
00:39:52
Speaker
And so one of the things that we're looking at using AI for and actually starting to work on this year is using AI to identify people that are affected by conflict and and remotely enroll them so that you don't need to have large numbers of like boots on the ground in incredibly insecure regions to say, hey, here's a community of people who have been affected by violence in the local area who've had to flee their homes, for example.
00:40:20
Speaker
and who might be in desperate need of support in this very moment, but there's kind of no easy way to reach them. And so using satellite imagery to identify displaced populations or using satellite imagery to identify communities that are affected by disaster that we otherwise wouldn't be able to know about um is another way that we're essentially using AI to increase our ability to reach people who otherwise wouldn't be reached.
00:40:49
Speaker
And how how good do you find that models work in in those areas? We're more advanced with communities that are affected by natural disasters. So let me give an example there. In the last few months, a couple of big hurricanes have hit the US. And we have work that exists in the US as well as in countries like DRC or Malawi and so on.
00:41:11
Speaker
and so In the US, what we were able to do when Hurricane Milton came through and like really kind of hit North Carolina pretty hard is use ah satellite imagery to identify which communities are kind of hardest hit as a result of this disaster and overlay that with ah poverty data.
00:41:30
Speaker
to basically identify which communities were both incredibly hard hit as well as incredibly poor. And then we work with a partner propel as an app that basically works that that people use to access their social security benefits or different benefits in the US.
00:41:47
Speaker
And so then we could partner with them to essentially just send a notification on their smartphone and say, you're eligible. You're you're in this community. This community is incredibly hard hit by this kind of natural disaster. You know you can enroll, ah receive $1,000 in the next couple of days to help you through this moment in crisis.
00:42:03
Speaker
and doing that process in that way that I just described radically increases the speed with which we are able to get resources into hands of people affected by disasters. like It can take months, like literally months for disaster relief organizations to get cash payments to people because of the challenge in identifying who needs it and then like enrolling them and going through Does that mean that it's it's wouldn't the money be worth much less to to the people in need at that point? Absolutely. Speed matters incredibly, incredibly, like every second matters in moments like these. And so radically increasing the speed with which you can make these decisions and then deliver money to the people involved also significantly increases the the efficacy of the money as well.
00:42:56
Speaker
Do you collaborate with local authorities or other organizations in in and organizing how you do you you do your work or how you help? like I could imagine that's trying to coordinate donations or cash transfers.
00:43:12
Speaker
with other orgs or local authorities would make it much more complex. It would be annoying work, I think. But i could also I could also see some conflicts arising if you haven't coordinated with others. and Perhaps someone gets help twice and and someone another person doesn't get any help at all, something like that. How do you how do you handle this without without and of without slowing yourself down too much?
00:43:35
Speaker
Yeah, it really so it does vary a lot based on the context, but as we have scaled in some of the different countries that we're in, it's become increasingly important to do the type of coordination work that you're describing, particularly with governments actually, because if we if we if we look at somewhere like Malawi where Earlier this year in Malawi, we we've kind of always done a community by community approach where we kind of say a single village, we're going to saturate everyone in that village. We're very interested in answering the question, how what does it look like when we start to do this at greater and and greater scale? What happens when we saturate a whole sub-district?
00:44:16
Speaker
a whole district. What happens if you saturated a whole country? like Could we, with a single injection of cash, like catapult an entire country out of extreme poverty? like That's a question that we're trying to answer. and so We're starting by going from the community to the sub-district level. and so Earlier this year, we redid every single adult in a sub-district.
00:44:36
Speaker
So 75,000 people all received transfer cash transfers in the same period of time. And then we're running a ah study to basically look at the effects of inflation and we're studying to look at the effects of like what it does to the local economy and various things. And you can imagine that if you're the governor of that district or if you're in the kind of local political leadership, if we were doing that without any sort of close coordination, you'd be like, what the hell is going on here?
00:45:02
Speaker
So we work very closely with governments in particular in the countries that we work with to make sure that we're kind of enhancing and part of their overall like poverty reduction strategy. And so in Malawi, in Rwanda, cash transfers and some of the work that we're doing is explicitly included into the national government's plan.
00:45:23
Speaker
And so that level of coordination happens quite closely. In disaster situations, it it is it is a more complex beast because you you want to speed is of the essence and you want to move fast. And I think one of our criteria is basically we want to be we want to go places that we have differentiated value to add.
00:45:41
Speaker
like And so we'll try to identify communities, we'll try to identify solutions that no one else is touching in order to try and both hi make sure that we're adding differentiated value and to reduce the cost of coordination overhead and so on and so forth, because you can really tie yourself in knots doing that kind of thing.
Scalability and Fraud Prevention with AI
00:46:04
Speaker
Maybe you could speak about the scalability of GiveDirectly's approach, and but then also the scalability of the the AI experiments or the AI solutions that you're using.
00:46:15
Speaker
One of the things that it attracted me to the whole cash transfer model is that there are just very, very few interventions in the world that can credibly scale to effectively and efficiently deploy billions of dollars in ways that very, very meaningfully help people.
00:46:38
Speaker
and cash transfers is one of those and so we could say like i can actually say with like some confidence that with our existing systems and organization that we could.
00:46:50
Speaker
deploy $1.5 billion dollars over the next few years with high confidence about what impact that's going to have, that it will be extremely efficiently used, that it will counterdpot catapult a really significant number of people out of extreme poverty. and That wouldn't require building like some monster infrastructure to do that because so much of it is enabled via technology. and so That, I think, is just like more generally talking about cash transfers as a model.
00:47:14
Speaker
Now, AI specifically, I think, does play into that, which is that when you're talking about deploying tens or hundreds of millions of dollars into community, you can be a bit more manual and targeted in how you identify who should you be who should be receiving it.
00:47:33
Speaker
as well as like how do you put the infrastructure in place to actually support troubleshooting that comes and all the problems and resolutions that need to happen when you're kind of building a program at that scale. I think AI is one of the big enablers.
00:47:48
Speaker
to think about how you can, with scale, get even more efficient with that kind of deployed resources in targeting. Yes, we've already talked about targeting, but also in you know fraud prevention, we do a lot of work on fraud prevention. We kind of spend a lot of time to make sure that the money actually reaches the people who who need it and isn't diverted by people along the way.
00:48:11
Speaker
We use a bunch of ML models in that, but that becomes a lot more scalable with AI as well. So so with with fraud prevention, this is something that credit card companies, existing organizations are spending a lot of resources on. Do you find that you can reuse some of their tools, some of their approaches, or do you have to do it differently because you're working with different populations, perhaps?
00:48:31
Speaker
I wish that we could leverage more from others on this. The nature of the problem that we're trying to solve is is a little bit different, unfortunately, than kind of credit scoring. you know There are some providers that we use to kind of help flag abnormalities in our data and to kind of send signals that we should investigate and so on, but a lot of it does have to be homegrown home'm grown as well because the populations that we're working with are not not not the normal populations that kind of fraud prevention machine learning companies are focused on.
00:48:59
Speaker
Yeah. So what is the big worry with fraud prevention for GiveDirectly? Is it middlemen cutting out or taking the money for themselves? Or or is it more low-level fraud? what is What do you worry about most?
00:49:12
Speaker
Yeah, so the there would be a couple of categories. so The first category would be imposters. This is people who move to a community in order to access the money that we would be deploying. and so We want to make sure that we're essentially not creating incentives for people to kind of follow us around and enroll and get money that they otherwise shouldn't have access to. So we have a set of systems in place to try and ensure that the people receiving the money you know only from that community and who are fully eligible and are not trying to kind of access money that they shouldn't otherwise have access to.
00:49:49
Speaker
and We actually do use some some AI in that, for example, when we enroll people, we register them, we like kind of tag that geospatially, and then we kind of see when they withdraw the money, are they in the same place, and you know things like that to kind of make sure that there's some good protections in place. so That's one. The second is diversion. This isn't about middlemen, but this is actually about are there bad actors that may actually be involved in low levels of our organization who try and divert the money to their own accounts ah or something like that. And so we do have teams that are enrolling recipients and that are trying to like make sure that we get everyone who's eligible. And so you could imagine that there are people who try to kind of manipulate that system in order to make sure that the money doesn't go to the recipients and instead goes to their personal banking house somewhere.
00:50:38
Speaker
And so we have a bunch of systems in place to try and prevent that, both automated and manual. There's automated systems, which is about recognizing abnormalities and various things like that. But then we also have this team that's called the internal audit team, which is fairly unusual, I think, for organizations like us because it's entirely firewalled. It's like a team of private investigators that no one knows exists.
00:51:01
Speaker
Or it's like, everyone knows that it exists, but no one knows who's on it. And so this this group actually, they have they have separate offices. They you know have code names. They're not allowed to represent that they kind of work forgive directly in any public forum whatsoever. And their entire reason for being is to basically do to run checks and balances on our systems. And essentially, but it's an internal red team is basically what it is. And and so that's another way that we kind of protect against that. That's highly advanced, it seems.
00:51:30
Speaker
Do you think GiveDirectly should give dumb phones to people who might not be able to afford them? Of course, GiveDirectly's mission is to do cash cash transfer, but maybe another way of scaling that would be to give the device you need to receive the cash.
00:51:45
Speaker
Yeah. And so we actually do. Okay. You already do. Okay. So when we go into a community, you know, maybe there might be like 10 or 20% of that community who already has a dumb phone of some kind. And if so, then they might get a full transfer of say $550. If you don't have a phone, then you can get a transfer of $540 and you'll get a ah kind of dumb phone alongside that and we'll actually get you set up on a SIM card for the first time and like make sure that all of that all of that is working.
00:52:13
Speaker
it It actually can go even further in Liberia, people don't just need cell phones, they also need cell coverage. and And in many of these remote places that we work, cell coverage is also a problem. So in Liberia actually, we're in active conversations with telecommunications companies.
00:52:32
Speaker
And they were basically saying, hey, can you tell us where you plan to go? Because if we know that you're going into a community, we know that there will be enough demand from the local community for cell phone coverage and for like withdrawing money and so on and so forth. And so we want to plan for that. and So if you tell give us enough notice, we will actually go through and we'll build cell towers wherever you're going to go in order to make sure that people have access because it's also in their commercial interest.
Data Quality and Strategic Investments
00:53:00
Speaker
And so we try and get all of the layers of the technology stack. because You can need cell coverage, then people need cell phones, and then they need the little bank accounts and so on. and But yeah, we do need to do that because in a lot of places, none of this infrastructure exists. Do you also think about ah this giving of of dumb phones from a data collection angle? Do you think about where you can gain the most information by giving giving out cell phones?
00:53:25
Speaker
Every single person that we that we give cash to, we will do follow-up surveys with and and we'll basically ask them what their experience was like. you know We try to run like a customer service organization in a sense. they're They're our customers and we need to improve based on what they tell us. so We ask them a bunch of questions about how the experience was. We all also ask them what they did with the money and all the different things.
00:53:49
Speaker
What is an example of an extreme upside from from giving say five hundred or a thousand dollars in cash if you do these interviews you must have. ah Many kind of good stories about how people are using this money do you have one one in mind that's like a a substantial outlier in terms of how well things can go.
00:54:07
Speaker
substantial outlier. That's a great question. I mean, I have a lot of positive stories. I'm trying to think of a specific outlier one though, because no one's ever asked me that question before. I mean, I think you do see you see a lot of people starting businesses. you see a lot of people I remember one woman that i that I met with a couple of months ago in Malawi.
00:54:28
Speaker
who, you know, what was interesting about her is many people, when they get the transfer, the first one of the first things they do is they decide to improve their home.
00:54:40
Speaker
They often are sleeping in thatch grooves that leak, that wake up their children, that you know have mosquitoes breeding in them, that you know all sorts of like really horrible things. One of the first things they choose to do is they say, I'm going to replace that with just an iron sheet, which allows me to get a good night's sleep, which means that my children won't get woken up by um rain, which means that i Mosquitoes won't be breeding in the thatch roof as well, and that's often one of the first decisions they made. and I met this woman who she was very notable because she hadn't made any improvements to her house. She basically had instead chosen to, she had three business interests. ah She had a grocery store, she'd bought some land and she'd bought some cattle and some ah farm animals, and then she was hiring local laborers to basically kind of work the land as well.
00:55:31
Speaker
And so you had these three different business trends and she's like, I calculated it all out. And if I put all of my money into kind of growing these businesses, then in a very short period of time, I'm going to be able to improve my home. But, you know, i I didn't want to, I wanted to kind of generate the cash flow first. And so she was doing great. She was bringing quite a lot of money across all of the different things. I don't know if it's an extreme outlier. I'll have to go and look up. I want to have a good answer to that question in the future because I don't, I don't have a real enough the top of my head.
00:55:58
Speaker
But it does' i mean it doesn't have to be an extreme outlier to be a great ah kind of event. So that's your story is also good. If we think about where data is is is kind of abundant and where data is sparse, i you know machine learning models often work well when you have an abundance of data from a variety of angles, so on. I can imagine that data is very sparse from the sources you're drawing from.
00:56:23
Speaker
And perhaps we've touched a bit on this already, but what are your strategies for for getting more data in these areas? Data is a huge challenge and I think both amount of data and quality of data.
00:56:36
Speaker
It usually involves a couple of angles. but One is we try to collect a lot ourselves, but two, we do a lot of partnering. We partner with both local institutions as well as international ones, and often that involves helping them identify which parts of that data are usable and which parts aren't and how to clean it and improve the quality over time. you know Even working with someone like a Google, the data can still be incredibly hard to access and incomplete. you know like Google has to task its platform with like satellite image and to kind of investigate the satellite imagery, to kind of investigate whether the quality is high enough to actually use. That's still a pretty manual process and that's with Google and and let alone the
00:57:22
Speaker
you know dozens or hundreds of local organizations that we often partner with in order to help identify who the right populations are to be targeting. and so I think this is number one of the biggest constraints honestly in the entire system is the amount and quality of data. and I would say in general, there also aren't many organizations operating in this type these types of environments that I would say are incredibly data fluent. There are some that are really good, but like on the whole, you have these like scrappy nonprofits like operating on a shoestring. Some digital forms, but also a lot of just manual stuff is incredibly cash-driven economies where you don't actually have a lot of digital infrastructure. and I do think that's like a really significant constraint in the entire system.
00:58:10
Speaker
Do you think that the future of of kind of data collection is more centralized or more decentralized? Do you think, think for example, satellite imagery, which I would call a a more centralized solution, is better or more scalable than having people with, with say, smartphones collect data on the on the ground?
00:58:29
Speaker
If I had to put my money on which of those options is going to increase the scale and quality of data fastest, it it would be that it's the central approaches. It will be that the combination of satellite imagery with national government poverty data with machine learning models and like telco data that that is kind of an unlock far faster than going community by community and surveying and all all of that kind of stuff because it's just so incredibly manual. And whenever it's that manual, you also introduce a lot of quality issues. And so I don't know, I kind of would like to believe that the decentralized approach would have a ah greater chance of catching up faster. But I think my
00:59:17
Speaker
kind of practical experience tells me that it's probably more going to be the more centralized approaches. When you've trained the model on some some data, say the satellite imagery, do you find that the quality of of the model's predictions degrade over time? Do you find that this the the situation on the on the ground is is changing and so a model that you that used to work well might not work so well and now?
00:59:40
Speaker
Yeah, I mean, the I think the short answer is yes, it is extremely rare for a single algorithm or set of inputs to have kind of durable value that doesn't degrade over time without constant fine tuning and updating.
01:00:02
Speaker
And this is you know before I was give directly I was working I was in technology and I was working more on consumer product type stuff and we would say this all the time that you know you you would.
01:00:13
Speaker
kind of have a new model or a new algorithm or new set of data and and it would work for a year and then it would just like everything is constantly degrading because context changes. you know it's ah and and You can never incorporate the full complexity of the world as a system into model in a way that allows it to update.
01:00:36
Speaker
and so you know Any organization that's looking to do good with AI or institution or or otherwise, you know it's tempting to believe that technology is this thing that you can just invest in significantly upfront and then just like let it do its thing. and It's like infinitely scalable and doesn't require constant maintenance and tuning. and i just like That's just not my experience.
Challenging Traditional Beliefs in Aid
01:01:00
Speaker
i think I think any engineer who's been working on ah on a large code base can relate to that.
01:01:06
Speaker
software to grades over time, including machine learning software. What are the biggest challenges facing GIF directly? The first big challenge is most people want to believe that they know better than people who are receiving money. The entire industry that I work on is kind of predicated me on on the idea that we know better than other people like what they should do and that it's smart people smart consultants in a room coming up with like complex economy models and whatever that are actually going to help people get out of poverty. and so Actually, the first big barrier is
01:01:45
Speaker
spreading the good word that unconditional direct cash transfers is not only more empowering and respectful of people's agency, it's also just like more effective. and so That has come a long way. Fifteen years ago when GiveDirectly was founded, this was an extremely contrarian idea that people did not even want to give the light of day to. It was seen as extremely risky. and Now,
01:02:13
Speaker
just a few weeks ago, USAID, which I would not characterize as the most kind of innovative, risk-forward organization in the world, just came out with a new position paper that says, we've looked at all the data and ah direct unconditional direct cash transfers are amongst the most effective things and most efficient things that anyone can do. and We're going to be doing a lot more of them.
01:02:35
Speaker
and that is like That's actually a huge deal because 10 years ago, they were like you couldn't even call them unconditional direct cash transfers. you had to like They were too risk averse and they wouldn't even consider that as an option. The biggest challenge is like help people understand this is not only more respectful of people's agency, but it's also more effective. That's point one. I think point two,
01:02:59
Speaker
is we are looking to operating in increasingly hard to operate contexts, and that requires different solutions. The example I was giving earlier of operating in the DRC, where almost 50% of the world's extreme poor are in fragile contexts. By 2030, it will be the vast majority of the world's extreme poor will be in these fragile contexts that are conflict-affected. We need to kind of innovate on how to reach those people efficiently and effectively and safely in order to be able to reach those. because i think the the The macro story of extreme poverty in the world is that it's on the whole getting better and there's a very real risk of a very large number of people being left behind.
Reaching Conflict-Affected Areas and Large-Scale Impact
01:03:45
Speaker
Extreme poverty as a proportion of the world's population has been in kind of significant decline since 1990, largely driven by economic growth in China and India.
01:03:56
Speaker
What that obscures is that the number of extreme poor in Africa, specifically sub-Saharan Africa, has increased during that period of time. and Often, it is in these contexts that are conflict-affected, and in order to kind of properly operate in those places, we need to find ways of reaching people that are safe, that are fast,
01:04:17
Speaker
and that allow them to kind of adapt to their local conditions. And I think, frankly, that we don't have silver bullets for those situations at the moment. And so that is like an incredibly important area of focus for us.
01:04:31
Speaker
Okay, as ah as a final question here, if you look over the the next decade and give directly is extremely ah successful, what does that look like? We have this bold experiment, which is basically we think that we could jumpstart a whole country out of extreme poverty. What would it look like to actually do cash not just at a local level, not just at a district level, but at a whole country level?
01:04:57
Speaker
and So we're working our way up to that. As I said, we started at the community level. We've gone to sub-district. We're moving to district level right now. and All of that is to essentially build a playbook for taking billions of dollars and then using that as kind of a one-time jumpstart to catalyze an entire country out of poverty. I don't want to oversimplify this in the sense that like cash is never going to be a silver bullet by itself. You need good governance. You need infrastructure in any country for this kind of thing to be possible.
01:05:27
Speaker
government needs to be investing in healthcare, care e etc et cetera, et cetera. It is 100% true that like more than $80 billion, $100 billion a year is spent on international development in various forms, taking just a small portion of that and saying, okay, let's kind of do this grand experiment to jumpstart and kickstart a whole economy all at once.
01:05:47
Speaker
and try to do that in a way which is therefore more sustainable. It's not something that it just requires these ongoing payments forever. That's our big bet. Can't say that it's going to work. But we think that it's it's absolutely kind of worth aspiring for and something that we
Vision for Scaling and Technology Integration
01:06:04
Speaker
should try. And so that's something that we're kind of building the economic models around, building the infrastructure to be able to deliver, and starting to work in a way that we think can credibly deploy the billions of dollars into those types of interventions. And so that's that's point one. Would that require yeah larger cash transfers or are we talking about the same and amount of cash transferred but to a larger number of people? Because it seems like incredible that that that you would be able to to to kind of change the fate of an entire country with a cash transfer that is that is quite limited ah per person.
01:06:43
Speaker
Yeah. so It would be roughly about the same amount of money per person, but to everyone. so now there's a lot of that's That's significantly oversimplifying it. and you know there's There's people who might need ongoing payments and support from an equivalent of social security. there's the The more targeted you can get in who's receiving the money, the less money you need to do overall. so I don't want to oversimplify it, but on the whole, it's about the same amount of money that we have been kind of doing at the village or district level. It is very significant though, in the sub-district that we just did in Malawi, us kind of coming in with these payments, increased local GDP by almost 40%. That is like orders of magnitude higher than, for example, the government cash stimulus as part of the financial financial crisis in 2008, significantly higher as an overall stimulus to the local economy.
01:07:35
Speaker
And the wild thing is that we were able to do that and we're running this like experiment on inflation and we saw zero inflation just by increasing the local GDP by almost 40%. how how can that How can that be the case? Yeah, how can that be true? i mean Basically, these economies have an extraordinary amount of what's called slack. Slack in the economy, which is essentially productive assets, people and businesses and infrastructure that just isn't being utilized fully because the there isn't the demand in the local economy. And so when you get this huge injection of of cash, people are able to take it up and absorb it because, say, a mechanic who might have
01:08:13
Speaker
previously had two customers a week to service their motorbikes is kind of not doing much in between that period like in between those two jobs. and so when When they start getting 10 jobs a week, like it can actually absorb that without needing to to kind of raise prices or anything like that. so What we're seeing empirically is when this money comes in, you don't see any movement on inflation. and That gives us a lot of optimism for being able to kind of scale this even further, but it is a really important thing. so That's why we're continuing to study it as well.
01:08:48
Speaker
so Big picture, 10 years out, like that's the that's the big task is to to kind of ah go after this this big bold bet. and you know Alongside that, we're building these capabilities to reach people in crisis, stop people from falling back as a result of being hit by a natural disaster, reach people who are affected by conflict, and be able to kind of scale those solution solutions more and more efficiently using technology, using AI, but also you know checking those things and and making sure that we're doing it well along the way. Fantastic. Nick, thanks for talking with me. It's been great. I really enjoyed it. Thanks, Gus.