Loading summary
Transcript45 lines
- [00:03]
Chris Hoffman
Welcome to Humanitarian Frontiers in AI, the podcast series where innovation meets impact. In each episode, we dive deep into how artificial intelligence is reshaping the future of humanitarian work. From enhancing crisis response to making a delivery smarter and more effective, AI is opening new doors in the way we support communities in need. In this series, hosts Chris Hoffman and Nassim Motelabi bring you thought leaders from academia and the tech industry to discuss not only the vast opportunities AI offers, but also the ethical considerations and risks we all must navigate. Join them on this journey as they explore AI's potential to transform lives and address humanity's most pressing challenges.
- [00:49]
Nassim Motelabi
Hey, Naseem, welcome back. This is our second episode. How are you doing?
- [00:53]
Lindsey Moore
Hi, Chris. Glad to be back.
- [00:56]
Nassim Motelabi
It's awesome. I'm still reeling from our last conversation. I had so much fun with that group of folks and so much learning there and everything. How have you been since the last episode?
- [01:06]
Lindsey Moore
Pretty busy. There's a lot going on around AI for humanitarians. There's a lot of interest. So happy to continue that conversation here and learn a little bit more about other organizations. And we have Yuri and Lindsey with us. We can speak about strategy and how AI is shaping how organizations strategize their humanitarian response efforts.
- [01:29]
Nassim Motelabi
Absolutely, absolutely. It's super great to have them both here. Yurian Lar is the director of Digital Transformation at the ifrc. Lindsey Moore is here with us, the CEO and founder of Develop Metrics. So this is going to be a really great conversation, I think, following on from our last strategy conversation. I think this is going to be a little bit different and provide some different insight as well from their perspective. So let me get started with you, Yurian. So when you're looking at what's going on at the IFRC and is artificial intelligence reshaping the way that you're thinking about strategy now? Is it influencing the way that you're thinking about it at the IFRC and also with the national societies? And is there an example you want to give us on the impact that you might already be feeling right now?
- [02:14]
Yurian Lar
Yeah. No thanks, Chris. And it's a pleasure to be here. So maybe first to say about our IFRC digital transformation strategy, overall objective is to increase the speed and the quality, the relevance and the scale of our humanitarian services to people. So it's important that anything we do with data or digital, including the artificial intelligence, is really to do and achieve that change. And certainly for us, AI has been around, but of course is becoming more prominent for us globally in our day to day work. But basically we identified three Pillars where we use AI. The first one is to improve productivity. The second is around knowledge and sharing knowledge better internally and externally. And the third is about delivering our humanitarian services to people. So maybe quick example of each three. So improving productivity, it's all about the workplace, tools, meetings being summarized, data being analyzed, translation. But also when it comes to knowledge management is going through big chunks of reports, make them available and easier, hopefully. So with some sort of chatbot functionality through our internal and external platforms, and when it comes to humanitarian applications by our national Red Cross and Red Crescent societies, it can be in areas of social media, listening, improving blood collection, but also very advanced and novel mental health chatbots that are being developed. So I hope that gives a bit of perspective of what we do and how we use AI.
- [03:52]
Nassim Motelabi
Well, that's awesome. I mean, I can imagine, Lindsey, that resonates quite directly with a lot of the stuff that you're doing as well, right?
- [04:00]
Yeah, for sure. I mean, we're seeing a lot of clients that we're involved with right now, particularly usaid, delving into AI a lot for strategy development. So really looking at all the evidence that wasn't humanly possible before to actually do evidence based decision making in the humanitarian field, which has been a huge advancement that we've seen recently. And also a lot on the word innovation has come up. Again, it always rears its head with a different meaning, mostly on kind of cost effectiveness. So really reducing costs, using AI to cut down some of the more burdensome tasks. And so, yeah, we're seeing a lot of the same types of work.
- [04:40]
Lindsey Moore
Sounds great. I'm super excited to kind of get into some of the challenges and opportunities you've seen moving forward in your organizations. We have this overall vision of how AI can contribute to the work we do in the humanitarian sector. But of course there's also challenges, either organizational challenges, processes that we face, or systems that are not necessarily set up to facilitate AI adoption. But also there are ethical considerations. Right. What are the challenges we're facing around the humanitarian principles and the rights of affected populations and the data that we're working with, the protection of their rights, the protection of their data, but also just generally in terms of our workforce. Right. Yuri, and you mentioned how the two major use cases, at least you identified around knowledge management, knowledge access, and the other one was productivity that includes the wider humanitarian organizations and their workforce. So with that also comes a level of responsibility. So I'm really curious to know what are these opportunities that you identified, but the challenges that you've seen so Far or you expect in the near future as you are navigating this area in your role?
- [05:58]
Yurian Lar
Yeah, what we see in general is that there's a huge amount of colleagues that are really interested in using AI, so they go around. So one important step we took was we need to develop these guidelines and giving guidance to colleagues how to use it and reflect against our fundamental principles to make sure we indeed do not do harm to people we want to serve or to our staff or volunteers as well. So this is still, I would say, in an experimental phase in the sense that how do you do that? Application of your guiding principles, your fundamental principles, how do you do that and apply that in a day to day practice with AI? So that is ongoing and it will be, let's say, a continuous journey. What we ask people to do is to review whatever is produced through AI. So always check the outputs generated and be very mindful of bias and anything sensitive. We also want to make sure that people have a safe space, that confidential information is not going outside to external parties with to their training models, because then we cannot control it. So we want to offer as well, and we do offer specific solutions in house that they can use without being at risk of security breaches. So as with anything, it is really also important we help colleagues remain really responsible in their attitudes and how they use AI, because we see that a lot of it is used. There's a lot of freebies out there. So when we have meetings on teams, we always check is there any sort of AI bot listening into our meeting. And if that is the case, we most of the time we delete or cancel them out of the meeting. But we then also have an alternative that we can do the recording, that we can provide a transcript, that we can provide summaries also generated through AI. So has to be that alternative that we offer. And in terms of bias, it is a real risk. And as an example, when we have a leadership meeting, we ask colleagues to produce pictures and of course pictures in the red cross and red crescent setting. So we have both these logos. And no matter how some of my colleagues and myself tried on the web to produce a picture where we are sort of meeting in a diverse setting, there was no way we could get the red crescent logo on the picture. It just doesn't exist. While the entire web is full of red crescents, it's full with Iversi logos, but no water, it didn't show up. So that is basically clear, the bias is significant. And the cross was all over the picture. So it was Possible to create diversity of the people around the table by prompting extra questions. But the red question didn't show up, however many prompts we were entering. So that was an interesting experience of bias.
- [08:49]
Lindsey Moore
Right? So interesting. This reminds me of some of the work that we're doing in WFP as well. Part of it, or a big part of it is kind of capacity building and I'm going to say that word, but it's really about sharing the knowledge around best practices. Right. And sharing how can you actually like what you mentioned around transcriptions and the bots and even the logos. You know, it takes a little bit of responsibility on individuals to actually take this on. Right. And adopt best practices around AI. And we realize like sometimes we may need a framework around it, we may need to have some guidelines around this, but it takes time. Right. Because you need to identify first what are the issues, what do we actually need to do. And I think some corporations are also struggling with the same. You see different governance themes coming out from Microsoft, from Google and others because they're also learning by just releasing these products and these tools with that. I wanted to actually turn the table to. Or the mic, I don't know, a table or mic to Lindsey to say, I think you have a very interesting position because you worked with an international organization before and now you're in a very technical position. And I think you have the perspective on what does it take for an organization to adopt a tool and a technology and where do they need support? So my question for you was how do you actually support those organizations that you're providing a service to in terms of building the capacity around adopting your solution? But also something that you mentioned is experimentation. It takes a lot of experimentation, it takes a lot of resources to even test and validate whether a solution is appropriate for an organization or not. And I think you are very familiar with that resource intensive validation process. So generally, how do you support organizations and what are your thoughts around this? Let's call it capacity building, right?
- [10:52]
Nassim Motelabi
No, that's a great question. And I mean the reality of it is that in the development sector, AI has not scaled yet. And we've seen this in a lot talking to a lot of organizations. There's a lot of pilots, there's a lot of trials out there. But the way the development sector is using AI, they're not taking it as their own yet. There's a lot of generalized tools, teams, GPT things that they're playing with. But development sector hasn't made their own mark on it and hasn't really made AI theirs yet. And I think there's a lot of valid reasons why, why that's the case. As part of our job we have to be more precautionary because the actions that we take influence a lot of people. And as you mentioned, there's very serious ethical bias considerations. So I don't fault the sector for not being at scale yet. I think it will get there. But in the meantime, like you mentioned, we have to be supportive with a lot of pilots, a lot of people getting their feet wet. And what I tell our clients and the donors that we work with is you just have to start playing with it. You have to just try and have that. There's always, at least in the us at usa, they are always are talking about how there is an appetite for risk. And I really hold them to that with AI because there is going to be failures, it is going to be messy, things are going to happen, but you have to start. If you don't start then you're going to fall behind. But I have seen the clients that are curious, they already are the experts in their field, right. We work with for example, if pre looking at how women empowerment and agriculture has progressed over the years, looking at all the literature and just by starting to engage with it, they are now the experts in how to work with AI in that subject matter. So there's so much rich technical expertise that can be brought to bear in the AI world if people start playing with it. And so I think that's really the attitude and it takes also an internal champion or more and somebody who really wants to do that change management because you can build the AI, you can make it in the right colors of everyone's logo, give it to them, but unless people are really internally wanting to use it and know how it's not going to work. So it's a multi step process as you as I think you've experienced as well in the seam that's still really in its infancy but that's also exciting as well because it just shows how much space there is to grow.
- [13:11]
Yeah. And I mean so talking about that it's not just one industry, right. In terms of we need to have an AI tech company come and help us. Right. There's a lot of folks that need to come and it's academia, it's really a collaborative effort to get around these things, to be able to get it right. As we're saying, I hearken to the discussions on inclusion.
- [13:30]
Yurian Lar
Right.
- [13:30]
Nassim Motelabi
And that's a whole nother piece of AI on Who can access it, who understands it, how can they then start to engage with it? And so I guess from your perspective, because you've been doing it longer than anybody, Lindsey, of the four of us sitting here, well, what have you seen? What is that, right, mix of collaboration that organizations have taken that made it work better? You talked about champions, right? We know that the champion piece is there. We know that breaking down the cultural barriers of internal organizational politics, et cetera, around it's really important. Those are the internal pieces. But what is that collaboration mix that you've seen that's tended to work?
- [14:06]
Yeah, I think those where it works really well are those people who understand that AI is not some kind of magic answer. Right. It's a collaborative process. And often those people who understand basic research methods understand, okay, it depends on the question that you ask and the data that you're looking at and the framework, the knowledge graph that you create, the answer that you get. And so people who aren't expecting the world from AI, because you know how it's been so overhyped, those are the ones that are willing to kind of get a response and say, this isn't exactly what I'm looking for. How can we change it and work to get it to where they want it to be? We see a lot of success in that type of mindset. People who just want quick answers that sound like they wrote it. We find that sometimes they can get frustrated because the first attempt often doesn't work. And so it does need some fine tuning and it can just be a couple hours of work getting it right. But I think there's been so much hype. Sometimes the expectation and the reality are too far apart within different teams. So I think that kind of healthy wanting to collaborate. A lot of organizations also are intimidated by it from the technical aspect, and so they don't want to engage in that process. But you don't have to be an engineer or a data scientist to engage in AI and to think through the methodology. So people who apply their own knowledge to this new technology seem to be extremely successful and are able to actually bring new ideas and new applications. So I think it's really a mindset in the end. And also I find that organizations that are newer tend to be a little bit more nimble. And so we have a lot of success with kind of companies that have recently acquired other companies and they're coming together and thinking, how can we bring our company together as a whole and unite all of our knowledge bases? That kind of bottom up thinking also tends to be effective, but every organization has such a different character. And then the last thing I'll add to that, Chris, which is a huge piece, is security. Because before we start with any client, we always have to get into security. And right now I advise gsa, which is the US Government's procurement agency, on AI procurement because it is such a thorny issue. And so you have to make sure you have the security fundamentals down before engaging in any project. So that's another one. You have to have a security team that will work with you for sure.
- [16:32]
Congratulations on that. By the way, I just saw that you posted, that was super cool. And again, I guess I want to point out and say thank you, Lindsay, because I know it's really early where you are, it's a holiday season, there's lots of travel. So thank you for doing that. I want to bring the question over to you, Nassim, and then over to Yurian, but it's the same thing from the academic perspective, right? You're coming from the academic realm and on a collaboration level, how does that look from academia and now sitting in WFP where you are, what lens does that bring in? And what could organizations learn from including academia in these discussions that they're having?
- [17:08]
Lindsey Moore
Super interesting question, especially that Lindsey just talked about procurement. I think these two don't often go hand in hand together. When we see how organizations adopt technologies in our sector perhaps and many other sectors is sometimes it's an off the shelf solution, it's ready to go, you deploy it. And so in wfb we have a big research unit, right? We have dedicated teams that conduct research and because we understand the critical understanding of not knowing what to expect, but the fact that it's important to actually explore and contribute to open data, to open research and also work with the data that is available to you to meet your needs and let's say predictive modeling, forecasting, all of these are very interesting cases, but they haven't still proven themselves as a solution, right. As something that could be scaled or reliable in some context. I'm very curious about Urine's perspective on this, generally in anticipatory action use cases, but when it comes to research, I think that's where the collaborative environment between academics and organizations come in. We have a lot of data. We have to apply certain processes to anonymize them or understand what data can be used for research and work with them to actually explore our needs and identify the best solutions forward. And I mentioned procurement because I think when we have this mindset Of I want to have a solution that is already proven valuable and valid in the industry, so I can easily deploy it. Given my lack of resources, given the sensitivity of the context that I work with, this duality is a very interesting dynamic. Right. So yeah, and in the organization in wfb, I'm always constantly trying to kind of bring that experimentation being that research perspective, by the same time provide value through the solutions that are already there, like low hanging fruits per se. But I'm going to end by saying that the low hanging fruits currently with LLMs and certain AI use cases, it's not so low hanging. Right. It's a big fruit, it's not a low hanging fruit. So I think with that we need to actually have a research perspective and mindset in order to effectively deploy these solutions. So now we see this kind of like, oh, where does this lie? Does this lie with procurement or does this lie with partnerships? Or does this, you know, like how can we create the capacity to monitor, to maintain these solutions, but also adapt and change as these solutions change as well? But also we're curious to know what your hints on this.
- [19:59]
Yurian Lar
Yeah, so a few things first about the question of scaling and the partnership. So we developed for the Philippines a predictive model to anticipate where the impact of typhoons are expected to be highest. So we could then take preemptive actions or early actions and to put people out of harm's way or their assets out of harm's way. So this was a journey of like four years to develop because we needed the partnerships at the local level with the local meteorological institute, with the local disaster management authorities, not just nationally, but then also sub nationally. And why we needed those partnerships because we needed to understand the decision making processes, what would people want to have in terms of information to then actually take meaningful actions. So thinking that through is very important, but it's also meant it took a long time. It was also about local partnerships. So if we would take that more globally, we need long term partnerships that want to join us on this longer complex journey, which includes all this research as well, but also deep design work to really understand what is the problem we're solving and what is the solution actually that makes people be able to actually take the right actions on the ground. So thinking through that, we discovered more or less the same with what we call automated damage assessment. This is where we've worked with Microsoft's AI for good. They verified some of the models, then they have also worked further and created some of these models to actually do these automated Damage assessments. And again, what is the main thing is maybe not the technology, but actually how do we bring that information to the people on the ground so they can make the right actions at the right time? And that information changes also over time. So in the beginning, you might have benefit from a global perspective, where is the impact highest? Like which province or even in the province, what sub level, what municipality level? But then you go further in the process quickly already, and you need other types of information to be brought to these responders. So it's very important to be able to go through that. Now, the point of scaling, I mean, this predictive model for early warning, early action, as we call it, we haven't been able to bring that to other places. But a large part of the issue is because the data in those other places is the challenge. It's not AI, it's not more hours of a data scientist to produce the algorithm. It's really about the absence of the right volumes and the right quality of data that we need to actually make these models scale to other places. Plus, of course, these aspects around how you need that local ownership of these models as well. There's examples that scale easier, I would say. But yeah, some of the challenges around predictive models and early warning, early action, which is extremely prominent in our industry, I mean, in the humanitarian sector, this is some of the challenges around partnerships.
- [23:04]
Nassim Motelabi
Yeah, Lindsey, on that, a data question, not to dive too deeply and not to get into too many technical terms, but this idea of bias and synthetic data and all of these things. So how do you fill in some of these data gaps? Like, I remember when I was with World Vision, we were doing also a predictive model, and we had to balance the indicators that we were trying to get data on based on what available data is. That was before people were really diving more deeply into the availability of LLMs to add to that, but two questions. A, using AI to fill out those data gaps, is that actually a possibility of being able to bring in what potentially it could be based on the maps, based on some historical different types of information that's coming in? A, and then B, this idea of synthetic data. Right. And does that give us an opportunity to make quicker changes in terms of our growth and scale of these issues? Because what Yurian, I think, was bringing out was that in general, we're still trying to use traditional processes on a new technology. Right. And how do we shorten the timeframe from four years to two years to one year to six months? Is it possible? When is it possible? What should Organizations be thinking about around that.
- [24:20]
Yeah, I mean, I don't like my answer, but I think it is that there's not a shortcut in this kind of work. I mean I really for me, the value of a highly thought out, well labeled data set, it's not replaceable with synthetic data. We experiment a lot with it. If we're in a crunch and we just really need to, we will. But we see that it increases bias significantly when you start using synthetic data. I mean it depends on the use case. Right. There are some use cases where you can tolerate some level of bias, especially if you know what the bias is going to be. But often we find that you need high quality and a high number of well labeled data to get an answer that you can trust. When you're talking about people's lives like you are with international development, if you're wanting to learn about management processes or other things, it's fine to use synthetic data. So I think it really depends on the question. And we, we have a data scientist just working on synthetic data. Right. Because that would be. We want to really cut down the amount of time we need to get to our answers. We don't love hiring teams of academics and debating what does the word resilience mean for hours on end so that we have the right label data. But you have to kind of like your said, sometimes the highest value of the AI projects are actually the human work that goes on behind it to get behind a shared understanding of a concept or a term. And what does it really mean? Because one of the big pitfalls we see is that you can't train a machine something that humans don't agree on themselves. Right. So we've gotten into some cases where there's an approach that donor wants to train the AI on, but then they don't have internal agreement on what that approach actually means. And so then imagine using synthetic data there you're just going to get yourself into a huge tailspin. And so we will use it in certain cases. But as of now, and the field is evolving every single day, so I'm hopeful. But right now we've seen that I wouldn't recommend it if you're worried about bias.
- [26:11]
Yurian Lar
Absolutely.
- [26:13]
Lindsey Moore
I just wanted to actually ask both of you, how do you think AI can actually help us advance in our data practices in an organization? Because, for example, Yurin, you mentioned you identified certain data gaps through a process of defining the project around the AI analytics that you were defining. Right. And I think that's a great opportunity to actually for us to see where our gaps are and perhaps even assess if it's worth the investment to fill in those data gaps. And maybe, I mean, synthetic data is one opportunity to fill in hopefully at Sunday some of these data gaps. But ultimately I can even think of all this data management work that we may want to do, data cleaning, data processes around, you know, what type of, let's say, metadata would you require? Right. Like as simple as that. We've recognized the importance of metadata in the work that we do with LLMs. Right. So I think this is a great opportunity to even use AI as a way to inform data governance and data practices in an organization. So what are your thoughts around this? Maybe urine you could go first because this is something that I would like to learn myself as well.
- [27:30]
Yurian Lar
Yeah, no, for me, AI and data management go really hand in hand, obviously. But the AI has also given us the opportunity to put more investment and place more attention to data management in the organization. So we used to have an enormous fragmented data landscape with huge amounts of teams that were really advanced in using data in their day to day work in different areas. But we didn't have one data platform, one data management approach. We didn't even have a single data governance for the entire organization. So that became of course, an objective to help us deliver more valuable data. And AI is just giving that push to the organization that we need to get that sorted and fix it. So that is now a journey that we started and it coincided actually with the deployment, development, deployment of an ERP for the organization. And that gives us now also one data technology layer that we can then start to leverage for all the different teams across the organization. And by bringing more data together, obviously we can benefit more from AI in generating the right information we need for making decisions and make those decisions faster. So the two are so intertwined. But I'm really exploiting the AI and the attention it has and the potential it gives to strengthen our data management, full stop.
- [29:02]
Nassim Motelabi
Yeah, I think what Yuran says is what I hear everyone say whenever we start with a new client. They're almost embarrassed by the state of their data systems and they think they're the only one. Oh, our data is in multiple drives. I don't even know where this is. It's horrible. And everyone, even the, you know, donors, you would think have the most act together. Nobody has all their data with good metadata stored in one place. It doesn't happen and, or structured. But AI exactly creates that impetus to, we have to get all of this together and we need a data Policy and we need an AI policy. So often when you start wanting to do a practical solution, we see that that's an opportunity where teams have to get all that thinking in place. And so I think it's great urine that you're recognizing that opportunity and leveraging it because there's so much internal work that needs to be done. And I also think the creation of a knowledge graph is so important because an AI can learn from it. But also the organization then has a really clear understanding of what their shared goals are and their shared objectives are. And we've actually seen conquest concrete improvements in organizations outcomes when they have a shared understanding of their objectives and their goals. And the cool thing about it is that once you create these kind of knowledge graph and AI is working within it, you find that organizational knowledge stays within the organization because that shared understanding and because of that contribution. When you know a key person who, the only person who understands, you know, extension services and Molly leaves, you still have all that knowledge curated in there and so you don't have to recreate that wheel every time you want to do a new project. And like you said Nassim, then you can also really quickly get to what are the evidence gaps rather than re researching the same things over and over again, which is what we really find happens. You can see very clearly where are those gaps that are worth investing resources in to learn something new and to actually be innovative. And I think that's really exciting.
- [30:57]
Lindsey Moore
And I have a follow up question to that because what I've recognized is we need people who are experts in the domain that they are working in that have some knowledge of the processes it takes to clean the data to generate this knowledge graph and also to deploy AI on top of that. Right. So this is a very interesting cross sectional, let's say in place to be. And we need that, you know, some teams may have this capacity, but majority like we look at the landscape of an organization, there is not much of that and it takes perhaps even a year or more to have a team with expertise in a certain domain to actually start adopting AI processes. Right. And when I say AI processes I mean what it takes in terms of a human capacity to start contributing to the process of AI adoption. So my question for both of you again is how do we build that capacity and what are you doing already in that regard? Right. Because I think I had the luxury of working with some teams that already had some expertise or we hired some expertise that works closely with us to kind of define that pipeline. But without it, it's a very difficult task. Right. Maybe Lindsay, you could go for some maybe.
- [32:15]
Nassim Motelabi
Yeah, I think, I mean, that's so true. And I think when Chris asked, you know, what does it take for an organization to be kind of successful? That's really it is that willing to get your feet wet. Because, you know, I know this sector loves training, but no amount of training is going to prepare anyone to really understand that cross sectional mix that you're talking about. And that skill set needs to be developed and it needs to be developed within the international development sector, the humanitarian sector themselves. We have to teach ourselves. It is really a self learning moment. And the only way you can do that is by doing it, engaging it. And those are the people who learn. You can study it, sure, that helps. But until you're actually faced with how are you going to apply it in your current situation, you're not going to get that skill set. And I think people who have the willingness to confront that huge breach of what they don't know and start playing are the ones who are going to be the innovative people who actually understand how it works. But I think the only way to get there is by doing it.
- [33:13]
Yurian Lar
Yeah. On our side. So I think we had, as I said before, many teams working with data or with IM professionals in those teams and they've been delivering the value for data to many parts of the house for a couple of years. So there was a huge amount of experience already, but there was not one central corporate, let's say organizational wide team responsible for the data for the. So that is what I've started to develop is one the objective and the second is then a central data management team which works with the other teams. So that was the first thing. So find the right manager for that role and with the right expertise that can actually then also build a team with the different types of experts that we need these days. So what we didn't have as an ifrc, we didn't have a data architect, we didn't have data engineers, we didn't have data platform manager or data platform engineer or architect. We had data analysts or data scientists in the house also. Not so many, but some of these foundational roles were not covered. So we are bringing these now in to the organization. And again we leverage the erp, the deployment actually to create this core capability. And with that we can actually make much more benefit from the existing experts that are already in the house in the different teams. So that is a bit the approach of bringing the right skills in and the right people into the Organization and then it becomes about finding the right partnerships with external parties, whether that's academia or whether it is with companies. But we need that first, that core capability in house also to manage those partnerships. Right. Because if we want to make value out of it, we need to be clearly able to explain what is the problem, we want to solve it and then leverage that partnership and actually getting it solved in the house and not external of the house. But that requires again that capability inside first. So that's been the strategy that we've been pursuing for the last year and a half to two. We're slowly getting there, but it's indeed a long journey. But it's exciting to actually get onto that path.
- [35:33]
Lindsey Moore
I have to say that's super. I think I relate to that. That's super relevant. What I'm hearing is we have a technical debt, I think, in a lot of organizations and we need to build those foundational technical skill sets within our organizations, but also incorporate it and mix it with some of the subject matter experts that we have in the organization across our different divisions who can contribute to developing the knowledge, cleaning the data, working on defining even our needs. Right. Because I sometimes I feel like we haven't done the right needs analysis even in the organization or like what do we want to actually do with AI. So I definitely appreciate this. Chris, over to you.
- [36:15]
Nassim Motelabi
No, it's great. I know that the listeners can't see Lindsey, but just her nodding of the head throughout this whole part has been great to watch because she's like, yes, yes, yes, exactly, exactly. HR to me is an enormous issue. And we know that the human resources, the questions we get from all the other folks that we speak to, how do I have the money to pay for a position that needs to compete with a tech company that needs the same position? And how do we balance that cost? And then the other side of it is the infrastructure. We tend to forget. The reason why this has always been piecemeal is because there's never been enough money put into the fact that this infrastructure, whether it's a data warehouse, a data lake, and then the staff to manage that and understand all of those pieces, it has never been there. As I think, Uran, you pointed out, you have to leverage something to get something else. Organizations, more for the listeners. I think organizations need to understand that this is not as easy as logging into ChatGPT. Right. There's a lot of stuff behind this that is required and there's a lot of investment. And so that goes to my question, which is around this return on investment. So we're all talking about the idea of what it can bring us, what it can do. And we also talk about the investment requirements that are actually there. And not every organization has deep pockets. Right. And especially we look at the predictions for the next few years, which are those pockets will be shrinking potentially, et cetera. And so how do people talk about this return on investment? Are you talking about it, are you looking at it, how are you trying to apply it? So maybe Lindsey, first to you, just how have organizations, when you've worked with them, are you able to showcase a return on investment or are they actually putting something on top of it that says okay, now we can actually see where we've gained efficiencies based on the investment, those types of things. And maybe after you Jurian, you can take it over.
- [38:12]
Yeah, thanks for bringing up the financing part because it's true, it's very under invested. The budgets that we're seeing, all these RFPs coming up. Create us a chatbot for $15,000 that reflects our own organizational knowledge. That's really what we're seeing, a lot of those coming out and then they're going to get answers, they're going to get products that don't meet their needs and then they're going to get angry at AI. Right. So I think that having budget behind these is a really important point that we also, you're saying probably are going to see dwindling. And so it's an interesting time where you have to make this investment and you really have to prove roi. As you mentioned, and I'm seeing the word innovation thrown around a lot in especially the US government right now. Meaning reduce costs, you find ways to reduce costs innovatively. But yeah, thankfully, because we were working in AI in donors, before GPT or before people knew what AI was, we were able to prove a lot of use cases before a lot of kind of the security crackdowns happened. And so we were auto generating documents I think about three or four years ago and we saw already then and it's been even better now that for example the Tajikistan mission at usaid and I forget the exact number but I think they saved around 1200 hours and I think a million, just under a million dollars of staff time in auto generating documents that took a couple days because the thing that I find the ROI where I really see the ROI is a lot of that work. That's a lot of bureaucracy where you have reporting work, right. So you have to go into all these reports, find the sentence that's talking about women's empowerment in each report and then report back on how all the projects are doing. Women's empowerment, you know, kind of these like questions. For example. Another example is USAID's ITR, the Innovation, Technology and research hub had a question from Congress. What is every digital intervention you're doing to counter repression and how are they going to figure that out? They have billions of documents. When I worked at usaid, I would get these requests all the time. I would bother or call every mission, I would bother them, they would go through their drive, find something, send it to me, I would compile it all together. You don't have to do that anymore. Now the AI can just go and find that exact sentence, pull it out, recompile it. That saves so much time and money that just there you can see a huge return on investment. And nobody wants to do that work either. So it really frees people up for kind of higher level thinking. And so there, I think you really, really can prove ROI quite quickly.
- [40:49]
Yeah, for sure. Great examples as well. Thanks a lot, Lindsey. And you're in.
- [40:54]
Yurian Lar
Yeah, so I have seen very, very good examples, let's say how returns of investments were shown. So the American Red Cross around blood donation and blood collection. They use AI already for some time and they showed that they've really cut cost by millions and they've also become much faster in blood collection. So the blood collection drives and then also the diversity or the quality of the products, the blood products have also improved. So they were very well able to show how AI have actually helped them in that part of their work. I also know colleagues in British Red Cross, for example, around automation. They also made a big case of proving the recovery of savings over time. But there we saw that it was three years, potentially four years. So then that was the perspective to front cost and then benefit two, three years down the line. Which is obviously not everyone's favorite model, but sometimes it is what it takes. And then we also have cases where we are not yet good enough in really developing the metrics and miss the baselines also then to measure against later. So thinking through before you make the investment, what type of ROI you want to make and how you can measure it, what baseline data you need, that you gather that data before is really important decision making or being aware that it does add something that is not quantifiable. I mean, in some cases being faster doesn't, you know, how can you express that in a financial term? So what is your R and what's your, your eye? Let's say, I mean it's very important to define that well enough for every of these use cases. And there was one more point. What I wanted to say is about cost savings. So it's a tough one because we are already tight in budget. And what is technology saving for the commercial sector? The tech just increases the revenue potentially. I mean the objective is revenue or it's improving the margin. But for us, what is the saving? The saving in principle, most of the time, unless it's around the logistics, but it's about the people. So we reduce the number of staff. That's a long process. If we want to change and use the technology to reduce the headcount. The question is where is the time saving taking place? With which particular positions? Where in the globe are these positions? How can we then actually transform those positions, realize the savings and pay for the technology? So far it's basically adding cost to the tech stack. And it's not most of the time we don't see yet that we are converting certain types of cost back into the tech budget.
- [43:52]
Nassim Motelabi
I love how you say that we have to define what is the I, but for our clients we're not seeing that it's about reducing people because what we're finding is people are so overburdened as it is that there's too much work to be done. And so what the return on investment a lot of the time is, is being able to do their work more efficiently to get better outcomes. So really how can you use the evidence, use the AI to better allocate resources so that they have higher returns so you're not inefficiently doing the same intervention that hasn't been working because you'd missed that one study that was written three years ago. And so a lot of the time it's just taking the burden off the existing staff so that they can actually operate on, kind of do the more high level thinking and have more effective projects so that the return on investment would hopefully go to the clients of development projects.
- [44:42]
Yurian Lar
Yeah, I just want to, I mean, absolutely. So that's for example, with these predictive models, that's where we can use the insights to actually prioritize the assistance in the right place at the right time for the right people. So certainly a lot of it is about that. But we need to be aware that the budgets are sort of at some point there fixed. It's hard to scale up your tech investments because of some of those limitations. But it's very, I mean I'm very excited about what is Ahead of us in this field.
- [45:14]
Nassim Motelabi
Well, awesome. Well, that's a great word to use, excitement, because this has been an amazing hour with you guys spending the time talking. So as we get into the excitement, the question to both of you is, what are you excited most about? The second question is, what are you scared most about when it comes to AI in the sector? So maybe Lindsay will give you a chance to think about that in Urian as well. But, Lindsay, first to you. What are you excited about and what are you fearful about?
- [45:42]
Yeah, I'm excited about the new models coming out. You know, right now we're still really relying on the normal Bayesian statistical methods and probability, but as we see kind of AI go more towards AGI and look at, like, more physical and social understanding. So when AI can mimic human intuition and causality and interact with the natural world, as we get to that level, I think it's going to explode. Everything I say excites me, scares me in equal measure. So I don't think I can answer that question equally, but it's going to be incredible as we see it evolve in that way. And then I think the biggest fear that I have, though, is really around the inequality that it will produce in terms of the people who are working with these models, mostly Western men who are building these and profiting from these. What is that going to do to people? And I know Nassim has a lot to say on this topic as well, but what is that going to do to people in developing countries who are not engaging, whose data is not being used into the model or is being used exploitably in the models? There's. I could go on. That's a whole other podcast, but I think an important one as well.
- [46:53]
Yurian Lar
Yeah, for sure, for sure. Yeah, I'm very excited because I. There's so many use cases or so many scenarios where we can use AI, as I said, in the early warning, early action domain or in blood donation, blood management is social media listening for me, especially where we are in an era where information and disinformation and misinformation, harmful information is becoming so much more around us. This is an important one where we can actually become much faster in understanding what is being shared and talked about, what is relevant for people to where we need to interact and where we need to act. So I'm very excited about that. Then everything around the data so that we are really getting our data quality improved. I'm very excited about that because that will also generate new opportunities we haven't seen now how we can Actually, maybe anticipate, well, using, let's say, anticipatory action to generate already the first steps in supply chain, putting already the orders out there before actually disaster has struck. Because we know from the past what we needed when there was a particular type of disaster. So how can we already advance some of our work in time to be just quicker on the spot with the right stuff? So I'm very excited about those things. Concern mostly it's about the divide. Yes. In all forms we can have is that we need to find ways to bring these technologies really to all the places that we work and where the vulnerabilities and the risks are highest to people. That's where our work has to concentrate. So we need to be really investing disproportionately, I would say, into the communities where our work is most needed.
- [48:43]
Nassim Motelabi
Well, Nassim, we're coming to the end of our second podcast, podcast number two. It's been great to have you here, Lindsay and Yuria. Nassim, I'll give the last thoughts to you, the takeaways, what you think and where do you think we're headed?
- [48:57]
Lindsey Moore
Super interesting conversation. So thanks to our guests for being here with us. I actually just wanted to maybe tie the conversation around your last question, Chris, because we are afraid of the inequalities associated with AI, but actually I wanted to also answer this question myself because actually I'm excited about the equality that it could generate because for the first time we see how knowledge can be accessible through AI because of the language, because of the LLM models. And I think I want to see that grow in our field, especially because we can contribute to that. I'm sure that WFP and other humanitarian organizations such as IFRC have a lot of field operations and field presence. And I think with that comes access to communities, comes being able to get the voices of the communities, be part of the conversation with AI and part of the data that we would probably want to utilize, but of course, responsibly. So I really learned a lot around the conversation around impact, how to measure it, and where do we actually go next. Right. We are being very realistic. There's a lot of groundwork to be done in terms of capacity building and our data systems and also collaborating. And I think AI, to me, is a beautiful space where actually has brought people together. And I think this is the most valuable part of this journey. Right. So, yeah, thank you all and it was a great conversation.
- [50:28]
Nassim Motelabi
Yeah, I really, really appreciate it, guys. Thank you so much. And we look forward to everyone tuning in next week for our next episode. And thank you to Innovation Norway, who is making this all possible so that we can really get the word out on frontiers in humanitarian AI. So thank you. Lindsey. Thank you. Yurian. Thanks, Naseem. It's great being here with you all. Have a great week.
- [50:49]
Thank you.
- [50:49]
Lindsey Moore
Thank you, thank you.
- [50:52]
Chris Hoffman
Thank you for joining us on humanitarian frontiers in AI. We hope today's conversation gave you new insights into how AI is transforming humanitarian efforts and the steps we need to take to ensure it's done ethically and effectively. If you enjoyed this episode, be sure to subscribe and stay tuned for more discussions with leaders and innovators at the intersection of technology and humanitarian work. Together, we're exploring how AI can bring real change to communities in need. Keep pushing the frontiers of possibility.