E (3:13)

In those situations, I feel like we have to make a lot of life altering decisions and they sometimes happen rapidly and the delegation of all of those decisions and those vulnerabilities to AI can affect and impact lives of people. So I'm kind of like thinking from that perspective, like, how do we build that trust? How do we ensure that there is no risk specifically in those vulnerable situations or like crisis where you don't have a lot of time for having that human in the loop feedback or having those assessment templates to walk through. So in those terms, having AI to boost the velocity of like making those decisions is great, but then how do you trust that is, I still think, a big question mark there.

B (4:03)

Yeah, well, that was big because I want to go to, I want to go to Hadassah on this from a legal perspective as well, on making these decisions. But when you see the billboards today, Palantir believes that everything can be read by AI and all decisions can be made from where to send a missile to where to do humanitarian action. So it's a big question. I mean, and trust is there. Hadassah, when you guys were looking at the responsible AI lab, were you guys designing not just how to design the systems, but saying here are the guardrails that you need to put in, but here are the places where you don't go. Right. This is the quicksand.

F (4:38)

Yeah, that's such a good question. I'm really happy that you brought that up. We were actually behind the scenes before this podcast. Sabrina and some colleagues actually talking, well, please, like sharing some of these developments. And it's just insane when the question that you just asked, you know, the first thing that came to mind for me was like, wow, you know, I see AI very much as double edged sword. Right. For a long time we were talking about the amazing benefits and of course there's a lot of hype around it. But then also you're seeing the dynamics in society changing around like what AI does where, what the role of humans is. So I find it really interesting to see from the perspective of the Responsible AI Institute, when we worked there, we were helping organizations playing the right guardrails in space, right. And oftentimes organizations themselves already had some sort of an idea of what is a no go area and what is all right, I think it's like the gray zone in between that is oftentimes very hard to really kind of identify and be able to understand like what guardrails are you put in place in different contexts. It's not the same across all organizations either and not even within a specific industry. Across organizations, the right guard rules can be very different. So I think from that perspective we well really try to support those organizations in understanding where to start, what to do next, both at a technical level and then also at a kind of like higher governance, organizational governance level.

C (5:49)

So that's very interesting to talk about guardrails. And Jegassa, you mentioned the human in the loop process. And my question for Sabrina is what does that actually mean? I think we've overused the word AI, that it's lost its meaning at this point. And I think we've used the word human in the loop a lot. Something a term that was coined Maybe more than 10 years ago, at least, that brought in human centered algorithmic design, human centered AI. I just want to know when we talk about guardrails and we read these guardrails and sometimes they refer to human in the loop, what does it actually mean and what should we actually account for when we are deploying AI solutions? Over to you, Sabrina.

G (6:37)

I think reflecting on a lot of what everyone has already said before, and I think this will help in understanding what we think about when we think about human and loop is that when we think about AI nowadays we tend to think that it always means that we're off sourcing decision making. I don't think that's always the case. You know, AI is a tool, is an accelerator for whatever work that you want to move quicker towards. But that means in a certain direction or it can mean at a certain magnitude. And so there are certain use cases, even in the same use case. For example, let's say like AI helping us make medical decisions, where it can be a support instead of oh, this is the diagnostic, this is the treatment and the AI just decides to say this is how we're going to do it, or it can be a research or diagnostic informational tool, where a human, an expert, a doctor, a physician, continues to be the person who just takes in that information, has a much better understanding of something, perhaps if the system as well is very well built and then is the one who can makes that decision. And so in that way, the space of AI applications and systems is much broader than we think, even when we're thinking about a particular use case. And so what that means for a human in the loop means that what human in the loop can mean can be a very, a lot of different things depending on how much autonomy or control over this decision making in this particular context the system has means that how humans are placed into how that system works should be different. If it has more autonomy, it moves quicker and makes decisions faster at a larger scale every single day, then perhaps you need a different type of way for humans to check the decisions it makes. Maybe it checks every 10. It checks, you know, it has certain flags where if something is like a little bit outside of the normal distribution of decisions, someone has to look at it. But that that type of human little loop is very different than the type of human in the loop when a physician is the one who has to input the decision at the end of the day. And so I feel like that is one way to kind of understand when we say human in the loop, we're talking about accountability and we're talking about checks in the process.

C (8:51)

I really like the fact that you tried to recognize the context specific applications of AI, but also what I hear from you is the AI lifecycle and recognizing the role of individuals in this. So I wanted to kind of now contextualize it for the humanitarians. And you also talked about the role of experts, AI experts, in just reviewing AI outputs, for example, when it comes to decision making.

G (9:21)

Right.

C (9:22)

When it comes to just assessing the output or the outcome of an AI algorithm or model. So my next question is essentially what is the role of affected populations when it comes to the humanitarian space? Because they're not just the users of AI or the target population that they, they will be given an AI solution, but they are often part of the input.

G (9:47)

Right?

C (9:47)

They should be part of this process. So when it comes to this AI life cycle, I wonder if we could talk about the role of affected populations. But also it is still sector specific.

F (10:01)

Right.

C (10:01)

Depending on the program, depending on the area of work, we can discuss the roles of affected populations. So Gatri, over to you on this. I know that you have some experience in your background in the health sector.

D (10:14)

Thanks so much. It's a good question. And just to clarify, when you say effective population, I think I haven't heard that term before. Would it be possible just to understand what that means?

C (10:25)

Thanks for that. Because we sometimes forget to explain when we're in the sector and we recognize that there are some terminologies. Affected populations are those populations that are affected by a CRIS or they are a recipient of humanitarian services. Right. So there could be different terminologies based on their circumstances and the situations as well.

D (10:48)

Yeah, thank you, that's really helpful. And it's a great question. Where do they find a voice in the AI lifecycle? And I think I try and think of it from two perspectives. One is from an implementation perspective, where is the voice of the end user in creating these solutions? And then two, where is the voice of the end user in creating policy or guidelines for ethics and safety and things like that? And those seem to be quite separate things. In terms of input, I certainly have much more familiarity with the first one. So typically how we try and account for user input from the very start is at some level the process, in my experience of designing these AI solutions is inherently inequitable because I am coming up with the idea. And you could say, who am I to come up with the idea? I am not part of the affected population or group. But okay, maybe leaving that aside, let's say you do come up with like five ideas. Given where technology is at right now, it's quite easy to make five quick prototypes and try them out with a group of users, get feedback through a simple Google form or even through like another chatbot that collects feedback in order to kind of guide your decision on both what kind of use case you should be focusing on, where for whom, in what language, what the format, structure, content should be. So I think just gathering feedback early on is one way in which at the Maggie we try and bring in user input. Chris, I'll over to you.

B (12:24)

This part of the conversation is one of the pieces that becomes difficult to grasp because sometimes we're not thinking about the end user, the affected populations that we're talking about as a customer, but they really are. And so what would we traditionally do from a product side? We wouldn't build a product and then take it out there and get some feedback. Right. We might bring people in to talk to us first about what are the problems that they need so that whatever idea we have, it's already formulating in our head and being formulated by the end user that we're trying to address. And so there is this point though, that. Because I completely understand what you're saying, because that's how it tends to work. The innovation tends to work in the humanitarian sector and in the development sector. We kind of build something out there and say that people will use it, right? And then. And then it's deployed. But I wonder, from a responsibility standpoint and an ethical standpoint, what ground is that? Are we being unethical by doing that? Are we being irresponsible by doing that? Or I guess it's just a question because we talk about it all the time. We talk about deploying something, testing it and asking for feedback and then iterating like that. But is that responsible? And is that ethical? Should we actually be going to the populations first and designing it with them and then building something and then deploying it? Because it feels right to do it that way. But I mean, are we really wrong by not doing it that way? I don't know if that question makes sense, but because we all. We talk about responsible AI and everybody that we talk to throughout this podcast has said responsible AI is going to the population first, asking them, and then building something that addresses their needs, that's built together with them. That's the definition of responsible AI that's coming from the sector. But is that an informed, responsible AI or is that just applying the old way that we did things to new technologies? It's just for me, I feel like the humanitarian sector is defining what it means as responsible AI for itself. And is that an informed definition? I guess is the distilled version.

F (14:20)

I'm not sure if I'm going to make this much easier if I'm actually going to give you an answer or make things more complicated. But just a thought that kind of popped up into my mind while you were saying this is that the context really matters here as well. What I mean by that is that when we're looking at just say B2B context, an AI product is being built to support sales teams make more and better sales. Let's just use that example. It's a lot easier for a product team and product developers to reach out to their target audience to get feedback on how they want to use that product and then embed responsible AI and trustworthy AI features within that process and better understand the for instance, around privacy and confidentiality, right? Like what type of information can you share? Can you not. Like, how would you want us to train the model, et cetera, like, things like that. When you're looking at the humanitarian context, I think the dynamics of developing and then ultimately implementing technology is slightly different. I think possibly also the speed at which that needs to happen is very different. And I think just. Again, I'm not sure if this is really helping come to an answer, but I just wanted to say that I think that the context and the dynamics in which that technology is being developed and ultimately deployed make it very hard to involve affected parties from the outset. So again, not an answer, but just a consideration that make it, I think, particularly hard to find effective ways to involve that group of population like you would in any standard kind of process of product development, right?

B (15:45)

Absolutely, Tsukiasa.

E (15:48)

I was exactly on the same train of thought that it depends very much on the context of the application being built. Yes, there can be a lot of transformative shift when we stop seeing a lot of populations as subjects of AI systems and kind of like start recognizing them as partners in the creation of those systems and governance. But again, it might not be as responsible as we might expect it to be. So I feel like that is where technology can help. We can simulate a lot of circumstances and synthesize a lot of situations, which we would anticipate by taking into account a lot of experts knowledge and local knowledge. I feel like that is where we can reach a middle ground where we're not directly going to the population, maybe a subset of them, and trying to synthesize or simulate a lot of circumstances and situations and seeing how our model or system would react in those. Before we go full out with the launch or the general awareness of that.

G (16:53)

I wanted to add from a perspective, just say, I have not built a system before. I don't understand this, perhaps an industry perspective. So please push back if you think that this is not grounded in reality. But when I hear this conversation, I think it makes me think of something we'll probably touch on soon, which is the speed of the innovation and how much that speed needs to meet the need of this particular problem or to serve populations effectively and serve those in need versus the quality of that help and how context specific it is. Why I think about that is I think that it goes back to where the technology is already going. Like the way that machine learning itself has developed into a focus on deep learning. Deep learning and this idea that we can take a very flexible architecture, throw the world's data into it, and therefore it knows all these things and then it can then understand from a generalized perspective how to work on specific problems is a very interesting way that machine learning technologists have thought about working on very specific problems. And I think that's where, like when we're talking about humanitarian problems, which are very unique, which talk about marginalized populations that have their own unique problems, this is where a lot of inherent tension is, is because when we think about things like few shot learning or no shot learning or trying to look at very specific underrepresented diseases and serving those populations, the idea is let's build a general understanding of this region and then let's take the few data points that we do have and just put it into the machine and let's see what the machine can figure out. And that is one way to go quickly. But there are problems with that, right? Because you're going from, you know, usually a very obviously represented population and trying to apply it to a marginalized or underrepresented population. And so the way that I think about this is that there's like a spectrum here. You choose, there's a trade off, you choose the speed, you choose a foundation of general knowledge and you try to go from there. Or you decide, like we're talking about here, to take a different response and say, let's work from the population, let's not decide to go straight to, you know, a giant model that knows everything and go from there. Let's see if we can formulate the problem and understand if we need to build a completely different model. Maybe it doesn't have to be a deep learning model, it doesn't have to be a giant model. Maybe it just has to be of a specific traditional machine learning or AI model.

D (19:27)

That's a really interesting point. When you said few short learning, what came to mind for me is we've been doing these bots in local languages like Malawi and Chichewa or Ndebele and Zimbabwe. And these models, like, even the top of class models, they just don't do well in these languages. So we've been using few shot learning or few shot prompting as one technique to try and improve the performance of a model without doing the whole fine tuning process. But I think what Chris's question made me think of was, and Chris, let me know if I'm interpreting this correctly, but I think your question is getting to the heart of this question that's on my mind a lot, which is who decides what is responsible and who decides what is safe and who decides what is accurate? Who is the decision maker here? And to everyone's point, it's so hard to like think like, like it's almost like we're proposing some kind of democratic structure that takes everyone's opinions into account in an equitable manner in order to come to some kind of framework that we can all agree to and then apply to the AI solutions that we make. And that to me sounds ideal. And also I don't know how we're going to get there, but I agree with the principle for sure.

E (20:42)

I think I should have lowered my hand because I felt like I was going to say exactly what Gayathri just referred to, that having a framework where we expect people from technical backgrounds and policy backgrounds to come together, but then again it becomes very much context dependent, like what kind of application, what kind of tiered risk are we aiming at? So the framework would very much vary depending on how risky the application is, what kind of like effective population is it impacting. And from like a technical standpoint, I also feel like having that diversified data collection approaches that combine a lot of like qualitative as well as quantitative contextual information gathered through a lot of like participatory research might help. And from like model building perspective, like developing the model, we need a lot of cultural competence and we need teams that are diverse and they come from like different parts of the world, different backgrounds, maybe even sociologists and local experts alongside these technical developers. Because that is when along with developing the model, we need a lot of context specific testing regimes that would help evaluate the performance under maybe actual conditions, the systems would be up, deployed and not just ideal environments that we very much often simulate.

C (22:05)

Gayathri, you mentioned recognizing our positioning when we're talking about AI, and it reminded me of my thinking around AI these days as a commodity. It's becoming something like oil or dollar or gold or whatever that we trade and put a trademark on. And then it took me also to the thinking about what you mentioned around responsible AI practices, because who is fueling that and where is it coming from? And then I went into my own thinking rabbit hole of colonial perspectives around technology and how much of this drive around commodization of AI and then commodization of responsible AI is fueled by these colonial perspectives, which is I think a fascinating topic. But at the same time I don't know if we will ever have the answer too in terms of resolving the issues around it, except a little bit of hope. Is this participatory action thinking or really understanding what the users, what the affected populations would want from AI? Because have we ever asked them, are you concerned about AI and the risks of AI? Because I have a feeling that if we ask them about AI, they will say, what is it going to do for me? That's the first thing that they're going to ask us. And I think if we can answer that, we'll be golden. What can I do for the humanitarians and humanitarian organizations and emergency response? That's a different question. And it may be be a bit easier because the humanitarians are part of an ecosystem largely set up by the global community to help the affected populations from developing countries or those who are prone to crisis. So I don't know if this makes sense, the map that I created, but essentially my question is how do we engage with the users, as Chris mentioned, if we want to step away from this thinking around affected population. But let's say the users, when it comes to the humanitarian space, let's ask them, what do they expect from AI? Something that is almost at this point an unknown thing that could do a lot of things, but what would they want from it? And then how can we actually respond to that need? The needs assessment you mentioned, Sabrina? I think it's very important, but also we're doing things backwards. The technology is there, we don't know the need. And it will take a lot of resource management, a lot of effort to involve the affected populations and the users to understand what would they need and what we can do for them. We kind of feel the opportunity, we see the opportunity, but we don't exactly know how it would help them, really. Right. So what are your ideas on that? Where do we start? I want action at this point. You know, like sometimes we come from all this thinking, but I'm really excited to learn more about what the future has for us when it comes to this. Gayatri, I think you have a comment. I'm excited.

D (25:32)

Thank you. Yeah, that's so helpful. And my first thought to your very good question is, do the people, the affected groups that we want to work with, do they understand what AI is? Do we fully understand what AI is? Like I to want work on LLMs and make chatbots day in and day out, Do I fully understand what AI is? And just as a first step, I would think the goal would be how do you communicate what AI is in order to achieve everything else that you said.

B (25:59)

That is a big one, Right. So because we're talking the different pieces that you guys have mentioned, the cultural aspect of things, right? And not only using it in language, but translating these big concepts into these culturally appropriate situations is really tough. And then there was the conference on AI in Africa that happened in Kigali, I think it was last week. And there was a lot that came out from that where a lot of people were talking about homegrown, from the African continent, right? Homegrown AI tools and things coming up. And there was a lot of talk around this decolonialization on that side, Nassim there at that conference, and amongst many of the speakers. But I think there's this juxtaposition that we get into on speed of rolling out, right? Because we're talking about humanitarian. When we think humanitarian, if we talked about what we call a sudden onset crisis, a sudden onset disaster, right, that you need to have a tool rolled out. One of the things that many organizations come to me and talk to me about is we want a tool that we can easily deploy to a number of different contexts. And that to me, this conversation happened twice today with two different clients of mine. And I said, guys, yes, the framework can be the same, but the model needs to be trained again on that context. Right? The language changes. Therefore, the way that we interact digitally through a chatbot or in any way, shape or form, the way that the answers are given has to be, again, culturally appropriately trained to that context. So, yes, I can have it on a Miro board and say that this can work in a lot of countries, but that doesn't mean that anything behind the Miro board, all the back end, is going to be any. You know, everything has to be new. And so that then brings up this issue of cost. And I wanted to talk a little bit about cost for a second, because doing something responsibly, which would be potentially appropriatizing these things to happen in every country, that's very costly. And can humanitarian organizations use it? Because for you working in the private sector, you have potentially a lot of money behind you, a lot of opportunities behind you to support you to do that. Gayatri, you're in the social impact sector. You've got less money to do those same things. So you're trying to adapt your tools to do it. But maybe to Sabrina and to Hadassah on this, when you were talking to people, when you were at the Responsible AI Institute about implementing Responsible AI, how were you explaining to them this juxtaposition now between cost, speed and the contextualization of the tools? Because it's like if everybody wants something that works globally for them, they've got to understand that that's probably seriously irresponsible because it'll never be contextually. Right. But I want to turn it over to you guys and get your thoughts on. When you were researching these things, did you ever come across this type of issue.

G (28:42)

The first thing to your point, Chris, I would say, is cost is less of an empirical metric than we think. When you're talking about framing, when we're talking about, well, okay, it is empirical, but it's not empirical in the way that we, as different people positioned in an organization or positioned around a problem, think about what cost means, because cost is about, well, how do you justify the cost, really? How do you understand it relative to what the problem is, what the benefit or the risks are related to the investment that you put into something? And what we saw with organizations that we worked with that I think applies in these situations is understanding how these different people surrounding this problem think about the problem differently. So, for example, a lot of the people, a lot in these organizations, especially private companies, when they think about this, the money is flowing from the top. This is not too different from what we're talking about here in these situations, right? The people who sponsor these things, the way you get the budget and the buy in is these executives who say, okay, well, this is worth. And this is not worth it. But then when you talk to the people, like even people who are the ones that will be using the chatbots that will help them with their work, they have a different understanding of it. They say, well, yeah, it's worth it, but it's worth it. But I can understand that from, well, is it actually going to help me? Is it going to do these things for me? And oftentimes in these organizations, we see people come out as champions in one way or the other. You know, sometimes it's even a role. It's a responsible champion. And they say, well, for this to work well in the situations that we're talking about, I understand the problem because I'm working with it, or I work with the people who work with it. And so when we're talking about cost and we're talking about this in a humanitarian perspective, we'd like to think that the customers are those executives at the top, or it's the humanitarian organizations at the top who hold the money bags. But the actual customer, the way that we need to understand and frame the problem is that those who are using the system, those who will be affected by the system, are the ones who decide. They say, this is worth it for me. And that's not in a money context. That's in a. It helps me in this way or it harms me in this way. And so I guess that's what I'll contribute to that.

F (30:55)

I think it's really interesting, especially kind of like the return on, you know, investments made in AI, it's so different across the board. Right? Because when we talk about return on investment in the humanitarian context, it's something completely different than for instance, what we were talking about when we were talking to large enterprise level organizations that were using AI to streamline their operations for a large part and to a sense essentially achieve operational efficiency as an organization. Right. And so what organizations just do the same very, in a very generalized way and bluntly is like they're aiming to make money and to satisfy their customers. So I think that again, it always comes down to context. I think it's, it's a word that I, I say the most. But even there, it's really interesting. As a society, as all of us, you know, thinking about the development of AI, it seems like oftentimes we're kind of running around without really knowing what we're exactly doing and why. What is the actual reason that we're talking about AI. And I think at this point in time it's not that we're all talking about the same thing. I think to the contrary, right. When we in this, like, I'm assuming at least like the six of us right now are talking about or seven are talking about AI, we're mainly thinking about how it may benefit people and society at large. But there are many different ways in which AI can be perceived, in which we can decide whether yes or no to develop, to deploy it. And so I think that plays a big role in understanding costs as well. Right. Like what is ultimately the threshold that you want to meet or pass and for what reason. So those are some thoughts that I think are important to weigh in.

E (32:29)

Yeah, thanks for putting it forward. I think like if responsible AI and humanity as a whole is taking a step forward, I think there is a lot of like research specifically in AI being done as well so that the cost can be lowered. And there are techniques being developed so that we can adapt these models for specific context, significantly improving their effectiveness, having like context specific understanding via techniques like fine tuning, making the models adaptable in a much more resource efficient manner, that they require less computational power than training the models from scratch. So techniques that make it feasible for humanitarian organizations, maybe with limited technical resources, or focusing on a lot of, I'd say open source models that require not only smaller, more efficient models to be deployed on maybe edge devices in low connectivity environments, while also maintaining a lot of acceptable performance with targeted tasks. I feel like those technical advancements focusing on data scarcity, data security and a lot of specialized applications which will maybe improve multilingual capabilities and focus on the effective population that we've been talking about, can also be a step forward where we don't have to worry a lot about cost or resources that might require for them to be more personalized.

D (33:56)

Thanks. My quick thoughts on cost from an NGO perspective. I think when we're creating these tools, one thing that we try and think about is cost effectiveness and what a cost effective interaction or what an outcome looks like and what kind of outcome. And even actually before you think about cost effectiveness, because one could do whole studies around cost effectiveness, but even in terms of outcome or output of a given interaction or a number of interactions, what were the outcomes and then at what cost? Even coming to those metrics or an understanding of those things is hard and takes time and again depends on the sector. Your answers would be very different if you had an education focused challenge, chatbot versus a healthcare chatbot versus like an agriculture chatbot. What your outcomes are or the impact you hope to have and then how much that cost. It's a question at a very, very practical level. I'll also say that there are the costs in terms of a technical perspective that of course I have far less insight into, but there are also costs from an implementation perspective. So just to throw this, because it happened last night, earlier this morning we had a deployment of a large language model based chatbot to about 100 users and was starting this morning in Zimbabwe. And until last night we had this complex workflow with lots of different LLMs being used at different points and things like that. And essentially at 8pm it broke. And so then we had to kind of redo it and then really quickly make a very simple chatbot that served the same functions. And the reason that we had to do that really quickly overnight was because our partner organizations are spending money to bring frontline workers and other people into a room to try out this chatbot. So there are multiple different costs involved. And when you think about speed and things like that, yes, it is the cost that you're going to incur when you use a particular key, like a particular LLM. But it's also all of these other costs that you're thinking about. At a very practical level, yes, like.

C (35:56)

In the humanitarian context, we can measure costs through how many people is reached right. Or the time of the intervention and humanitarian response. And just generally measuring cost has been always the question question in the humanitarian sector. But what interests me in some of the data that I see is to have a good return on Investment, sometimes you have to be able to really scale your AI solution. And that sometimes is in contrast when you want to contextualize your AI model or when you have to deploy in a smaller scale or for a very particular task. So we see that sometimes the return on investment that businesses or corporations advocate for, for scalability for on investment may not directly translate to the humanitarian sector, necessarily, given the discussion that we had today. So I guess Chris L. Asked the last question. It was a very fascinating conversation because we had very technical experts in the responsible AI field. So my question, I want to be a little bit more positive today and ask, what is the direction that you would hope AI investments would take us, given that we see different countries and different organizations and corporations really starting to pitch in in the area of AI, maybe not having the ROIs or return on investments yet, but kind of foreseeing a return in the future. So maybe we'll start from YASA and then we'll go around.

E (37:32)

I wanted to understand the question a little bit more when you mentioned where can AI take us? What are we focusing on from, like a humanitarian organization perspective or like focusing on responsible AI particularly? Yeah, if you could like, rephrase the question, that would be great.

C (37:48)

It's for your individual perspective. It doesn't matter. We all have very different backgrounds and very different hopes for AI, so we would like to capture that. Right. Like, what do you expect AI take us in the near future?

E (38:03)

There is a notion of that AI is a threat to humanity. And I am of a different viewpoint. I feel like AI is nothing but an enabler of humanity. It helps us boost and free up a lot of time focusing on valuable tasks that absolutely need human insight and not oversight, particularly where you are having that. We talked about human in the loop where we need humans particularly. So I feel like if we think of AI as not a replacement but as like a coworker or someone who is working alongside us or maybe like a digital twin. That is where I see AI moving towards. But there's this flip side where we also want to focus on, like, in whose hands are we handing over this technology? Because a slight slack in these intelligent systems, as we all know, can create havoc. So we have to tread with a very focused viewpoint.

C (39:02)

Sounds good. And Sabrina, over to you.

G (39:05)

I hope you don't mind if I answer this a bit indirectly, but what I think I'm very hopeful for, for AI being used in humanitarian contexts is that I think it's an opportunity to meet the challenges that come with AI as a way to Continue to improve on the process of empowering people in humanitarian contexts. What I mean by that is that we've talked a little bit today about kind of the colonial history of some of these things, the power imbalances that emerge in a lot of these areas. And the great thing about AI to what we talked about before is that we can build small prototypes, we can. We can move really quickly, we can try a bunch of different solutions, and we can do that not by just trying to build different prototypes of the technology itself, but let's think about different ways of how we can build it amongst each other. So something I was talking about before was, what if we had the person who led the building of an AI solution for a particular context, be a local champion? That person is the project manager. That person is the one who defines the success metrics and builds out the problem space and articulates all those things. And the person who sits at the table and lays out their tools is the AI engineer. That engineer is not the expert in that situation. And if we could pilot that, we could pilot those ways of working really quickly with technology that enables us to try a bunch of things. I think that the humanitarian space is the space where. Where if there's any space in the world, where there's an opportunity for us to rework the way that we think about building solutions that are meant to actually help people.

C (40:53)

I resonate with that a lot because I'm trying to always advocate for new processes and that we don't spend enough time talking about the rigid processes that we work with. And it's less about the technology sometimes, and we have to get out of our bubbles to rethink about how we do AI. And I think it's an opportunity to also rethink how we do technology at large, not just AI. So over to you, Hadassah.

F (41:19)

Yeah, I'm going to be very boring and honestly say that I very much echo what both you, Nassim and Sabrina were saying, in the sense that the first thing that popped in my mind as Sabrina started talking is something that I take as my point of departure in these discussions, and that is that it's people process and then tech. And so from that perspective, I fully agree. I think that the way we think about AI in humanitarian context, but more generally, hopefully that's kind of my hope will change. Change to the extent that we don't think of AI as being the point of departure for every discussion that we have, every way we think about how we can improve, the way we either interact as people or that we solve problems, but that we get a better understanding, first of all, of the fact that there are other pieces of the puzzle that were there before that now we think AI is automatically going to solve. I think it's very far fetched to say, oh well, do we really need AI at all? There is a lot of proof for the fact that it can drive processes forward in ways that we weren't able to before. But just, you know, reiterating the point that Sabrina was making, right, we need to better understand where those other puzzle pieces fit in before we talk about that tech piece necessarily or proceed with that conversation because there is a. And I was going to end it positively, but like nonetheless, you know, there is a danger of us overthinking the tech side and then completely neglecting the fact that, that we are supposed to interact with it in some way or another, being the recipient or part of like development or in any, any way with, you know, whatsoever. We're all in some way or another affected parties. So anyway, to come to the point, I think that there's a lot of opportunity for us to shape that conversation now. Even though there's a lot of thought around how, where, you know, we're, it's becoming too late, it's still too early. It's like there are many different opinions on this, but I think there is like an opportunity to change, shape that discussion as opposed to saying like do we need the tech? Don't we? But how does it fit within our own conversations right now as people and as part of a larger process? Over to you, Gayatri. I think you're there. The last but not least, I think.

D (43:19)

I really like the theme that Sabrina showed in terms of the local VM and then the engineer besides them with the local VM kind of directing what particular tool or solution would look like that the definitely sounds like a really good one path forward for the future. And then another thing that I think would also be helpful for the adoption of AI in a more equitable manner would be trying to address the problem of a lack of an equitable access to mobile technology for girls and others around the world. And just kind of keeping that in mind while we develop these solutions because we can make for example, the most gender intentional tool, but if not that many people can access it of the group that we want, then that's probably not helpful. So I would say I completely agree with that visual of like the local VM and then also trying to think about other barriers that specific groups of users might face. And then try and keep those in mind when creating solutions.

B (44:21)

It's been an amazing discussion today. I really and truly, just listening to all of you talk and your perspectives has been. Been really helpful. And I think it's going to be helpful to the listener because as we were talking about the end user, the end user for this is our listeners, right? It's fun for us to be able to talk together, but I think this is going to be a really informational and helpful conversation because the conversation is moving very quickly on the outside. And I think that there's a force that then pushes that to the inside. And a lot like what you were talking about Hadassah, around people and process and that AI AI is now at the tip of our tongue on everything. AI must be the answer for everything. And we Forget that just three years ago, 99% of the people that are using any AI terminology had never even heard of it before. So it is a human condition to do that. Just like we needed the mobile phone. When the mobile phones came out and you had the one odd uncle that was like, you got to call me on the landline because you can't reach me on my. I don't have a mobile phone. There's always going to be that. But now this conversation has really moved so quickly and I feel like just taking a step back, going back to what the basketball coach would say would be the fundamentals of how we engage with people and how we work with people and use this as an enabler is such a great point that all of you have brought up. And I just want to thank you all. I want to thank you, Nassim. I know that it's been a busy time and I know that you took some time out. So coming here and sitting with me again, I know it's probably painful having to sit here with me all the time, but I really do appreciate you doing that. You're the most amazing co host and it's so great to have you here. And so, hey, for many it is the Easter holiday. So if you are celebrating Easter, happy Easter to you. And we've just finished Ramadan, which is amazing. So that has been a great thing for many. And lots of other amazing holidays are coming for people. So I wish you all happy holidays over this time. And yeah, enjoy spring, enjoy the flowers. Nasim. We get to have one more episode together, just you and me wrapping it up. So that's going to be fun. I can't wait to do that with you. But listen, Hadassah Chigyasa Gayatri and Sabrina. I want to thank you all for joining us today. It's been a fun time.

C (46:25)

Thank you.

F (46:25)

Thanks so much.

G (46:26)

Thanks so much.

D (46:27)

Thank you.

A (46:28)

All right, 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.