
Geoff Buteau, Director of Booz Allen Hamilton AI Practice and Saeed Uri, SVP for the Sustainable Development Impact Lab at Chemonics International join Mike Shanley to discuss AI in Federal & USAID Markets. Specifically, this...
Loading summary
A
Welcome to the Aid Market Podcast where foreign aid partners connect to learn about key funding trends and market insight. The podcast is co hosted by Aid Connect Data, the pipeline and market intel software for USAID Partnering and Connected International, the leading USAID partnering support consulting firm. Now here's your host, Mike Shanley.
B
Welcome to the Aid Market Podcast. We're excited to have our guest today. Saeed Uri, SVP for impact at Comonics said has 15 years of experience at Comonics, including focusing on AI work in the USAID market and the international development market over the last year. And Jeff Bhutto, the Director and director of AI in the AI practice at Booz Allen Hamilton, with 12 years of experience in Booz Allen Hamilton and five years of experience in the AI practice. Also thank you to Booz Allen Hamilton for co hosting this very important discussion today. As I was talking with Jeff and said in the the preparations for this typically when we look at past performance, we're looking at what can we learn from other countries, from other donors, from other organizations and apply it to usaid. This is an interesting situation. We're actually looking at what can we learn from other government applications, other federal applications that might be applicable for USAID and the USAID implementing partners. So I'm excited to get right into the conversation and start out with Jeff. Jeff, for those of us in the aid market, could you give us a little sense of what's going on with applications of AI in the general federal market?
C
Yeah, definitely. So, so thanks for having me, Mike and Saeed, I I can't wait to get started on this on this discussion in the federal market. I think what you're seeing is pretty consistent with spending. So if you're the Defense Department, you've been doing this for more than six or seven years at this point, which started with Project Maven. Big, large projects that are bringing in big, large artificial intelligence capability providers. I think as those budgets get a little bit smaller, you're seeing this spectrum of big efforts on the big money, big side. And then you're seeing some small prototype efforts where some medium and small sized departments are trying to understand what the best use of artificial intelligence is. So just to just to color that a little bit, I can give some examples. Way back when the Defense Department had conducted Project Maven, which was an effort to bring computer vision algorithms to the table just to help analysts be able to see and identify objects with with computer vision. That resulted in a lot of testing and evaluation as well. Some of that work Booz Allen did, did conduct along with a lot of partners, competitors and competimates that. So that was a great project overall. I think the Defense Department took, took a lot on when they said, okay, we're going to try and test and evaluate these algorithms so that when they go down into the field, we have a level of confidence that they're going to perform how we think they're going to perform. As you get a little bit smaller on those budgets, you're seeing these, these projects get a little bit more medium sized. Like for example, the army has done a little bit of work with bringing artificial intelligence algorithms and computer vision specifically down to the, down to the edge device so that that type of work can be conducted in low compute and low bandwidth and low power environments. I think what you're seeing in some other agencies that might have a lot more restrictions than the, than the Defense Department does, you're seeing some medium sized projects such as entity resolution, just to make sure that some of this data can be automatically, each entity with that data is automatically recognized so that when you apply artificial intelligence to it later, that data is easier to comb through. And then on the prototype side, you're seeing some of these agencies, perhaps in law enforcement or immigration and migration, you're seeing prototypes and generative artificial intelligence where it's essentially taking, giving the ability for agents and other types of analysts to use generative AI to comb through a lot of data and help in assisting making a decision. So that's a lot of that stuff is in the prototype phase right now, which generally would be my suggestion for any agency or any department or any sub agency that's looking to get started in this. If there's some kind of funding that they can produce to just produce like a small level prototype, you know that that tends to start to paint the picture of what's possible for artificial intelligence. And even more importantly, when you show it in front of a user or you try and match it to the mission, you know, you get that feedback that you wouldn't normally get with an out of the box solution.
B
Thanks, Jeff. And I would say, and say I want to get your thoughts on this in one second, but one quick follow up. When ChatGPT came out, I really feel like that was a watershed moment really when most people in the aid industry at least were able to go on there, engage with and say, oh wow, there's at least some things this does really well. Who knows what the broader application is in the federal market. Did you feel that ChatGPT's release was as much of a Watershed moment or as I noted in the intro, you've been in the AI practice for five plus years. I don't think I know any organization in the aid space that's had an AI practice. I think there's definitely been individuals and champions inside those organizations working on this for a while. But yeah, did. Was that as big of a splash in the Federal Market when ChatGPT was released?
C
Absolutely. I mean I think that the, the, the wide range of applicability that I think anyone could really see when they initially use that technology. I mean I, I, I think it, I think it just inspired a lot of people with regard to some of these federal mission use cases. And I think, you know, I've got colleagues that have talked about this on social media and in interviews with aws. There's this, there's this broad level capability that generative AI is bringing to the table where you can ask questions and you can respond and you get a response that's like somewhat accurate to what you're looking for and you can point it to different data sources. Some of my colleagues have mentioned that the next phase is going to be how close can you get that very broad applicable generative AI capability and pull that into a specific mission. You know, how can you refine the LLM? How can you pull in just only a certain amount of data sources to make sure the LLM is, you know, make sure that the vector databases are there so that the LLM can go and efficiently and precisely go and grab the data that you're asking for in a, in a mission context with different types of data in different forms. And you're starting to get into that multimodal aspect of artificial intelligence and generative AI, which is merging the computer vision and the semantics field to be able to get you that like mission specific type of data which I think at this point, you know, you've got many of many firms are working on this and many federal agencies are doing great and leaning forward and the executive order right now has asked all the federal agencies to document all of their use cases. So you're going to start to see a little bit more of the sharing of information. And so there, you know, one agency is going to say, oh well look at what this agency is doing, we should give that a go. And you know, you're start, you're going to get this like inverted kind of pyramid of specificity when it comes to generative AI.
D
Great, thanks.
B
Jeff Saeed, let's go to you and let's check in on the USAID market, the international development and aid market. I know there's really been a focus of you and your team. Could you talk a little bit about what you see as the current state of AI applications in the aid market and feel free to share a little bit about you and your team's work.
E
Yeah, sure. Thanks Mike. And thanks for again for hosting this wonderful podcast. I'm a first time caller but longtime listener. Wonderful to hear for your comments.
B
Great to have you.
E
Yeah. And I think me personally, my first interaction with AI as a possible solution for some of the challenges we see in the aid space was maybe back in 2017 when I was working with some partners on Syria and we were exploring the possibility of using machine learning to forecast where potential airstrikes were happening in Syria. And this is kind of using that type of technology based on the data that we were able to collect inside of Syria to then alert civilians of incoming airstrikes, which was an interesting use case for us that has then developed into other use cases using similar technology on our Some of our work that's looking at countering poaching in different parts of the world, for example, is taking a similar approach where we're using large data sets and feeding them into learning models to try to forecast issues of poaching and where there might be poaching in the future. And how can rangers then prepare themselves? Because as you know, you don't have enough rangers to cover these huge parks that are out there. And another example that we know of as well and through our supply chain work is really looking at how we can leverage our access to huge amounts of data to then forecast supply and demand needs at the last mile. So where clients and communities are receiving medication to ensure that that medication is always available for folks around the world. More recently, and as we've seen kind of the chatgpt surprise kind of the.
D
Examples have shifted quite significantly. I think right now what we're hearing from USAID and from our projects around the world and how they're exploring using generative AI is really around helping with pulling insights out of data, qualitative data. As you know, Mike, USA projects produce hundreds of reports. They're constantly interviewing stakeholders. There's lots of other organizations that are working in similar spaces. And I just know from my own experience, you know, you don't have enough years in your lifetime to read everything the UN is pushing out on a specific country. So I've seen our teammates around the world use different gen AI solutions to pull out those insights very quickly and to help inform our Programming in a way that allows us to accelerate impact on the ground by better understanding what others are doing, by better understanding the challenges and opportunities that are identified in different reports. We're really able to put the power in the hands of our, of our teammates in those countries to then design interventions that take all that information into account. So where you might have a discovery phase of weeks or months now with the help of tools like Chat, you're able to do that in days, if not shorter amounts of time, which is really incredible. And I really can't state how significant this is, especially when you take into consideration the fact that a lot of our teammates, their first language is not English. And so you really leveled the playing field as well by providing them this tool where they now have access to a lot more information and capabilities than they did before. We're also seeing USAID look internally. USAID is also a huge collector of information and reports. They have a huge trove of lessons learned that I know that folks access through the Development Experience Clearinghouse. USAID is trying to understand what insights and learning it can pull out of that, pull out of those reports to then inform future programming. Where are they most effective? Where is assistance achieving the results that it wants to achieve in a way that was, was not possible, you know, before, before the release of ChatGPT within.
E
Chemonix, we're kind of running on those, those two same pathways as well. We're thinking through how can we augment our work with the different gen AI solutions that are out there. And we're working very closely with Microsoft on a few solutions that are out just to kind of provide productivity tools to our staff and teammates around the world that will really help achieve significant efficiencies.
D
You know, I can't tell you how many times we've had crunch time over writing weekly reports or annual reports or.
E
Drafting analysis of, of specific information that.
D
We'Re collecting that would eat through a day and really kind of take your attention away from some of the more critical tasks that you should be doing. And so we're really seeing what productivity gains could we achieve that would free up our teammates time to then focus on that more important work, pulling together a PowerPoint presentation, which is something that would take quite a lot of time. Now with tools like Copilot's M365 suite.
E
You'Re able to do in a much.
D
Shorter amount of time.
E
On the other end, we're also thinking.
D
Through like, how could this be transformational? All of the different activities that we had planned before Chatgpt was launched. We're now taking another look at them. How can we think about building the capacity of our staff differently now that we have tools that are out there that can essentially serve as your own assistant? How do you think about training differently when you have access to your own personal assistant that can answer questions from all of USAID and Commonix's information? So we're really thinking through what role can Chemonix play with its global presence, with its experience, with its relationship with USAID and other donors? How can it play that convening factor, that convening role to see what's possible and really transform the way the sector works and really put the power in the hands of the communities that we're all aiming to serve.
E
So it's all really exciting for us.
D
And we're very much in the think big phase, but kind of working off of what Jeff shared. We're also starting very, very small. It's a novel technology. We all know the risks that are out there. We want to make sure that we're approaching this in a very ethical way and taking into consideration the different biases these tools might bring and kind of starting small with minimum viable products that will then allow us to learn and scale as quickly as possible. So what Jeff shared is something that we've definitely taken to heart and it's why Comonex established Sustainable Development Impact Lab as a way of trying to kind of find an idea that's worth testing and then test it out and learning as quickly as possible.
B
Well, Saeed, I appreciate that and definitely, especially at AID Connect, we're very excited about the applications for local partners on both the business development side and as you mentioned, the project implementation, the report writing, what are ways that it can make that, you know, reviewing hundreds, thousands of reports no longer the barrier to entry and really facilitate putting together even better reporting. If you're spending less time drafting it, you could do more time doing data collection and you can use AI for data collection. So there's a lot of great things there. You mentioned the I guess you two quick follow up questions. I'd like to get Jeff's thoughts on those both in how you're applying these in an ethical way. How are you even framing or approaching the conversation? And then also I guess going with that is partnering again. We talked a little bit in the prep here about how are you going about finding the right partners with the right AI expertise. And yeah, yeah, maybe there isn't that traditional past performance or expertise among traditional aid partners.
E
Yeah, that's A fantastic question and something that at Commonix we actually spent quite a bit of time focusing on up front. And what's been great about this AI experiment is that a lot of organizations, whether private sector learning institutions or the public sector, have really been approaching this in an open source manner. Everyone is recognizing that this is a novel technology, that it has risks and opportunities, and they're all sharing what they're experiencing and learning. For example, I know Stanford University pushed out a report recently that highlights different risks that AI tools bring and how you can mitigate some of those risks. And so when we were looking at what's possible with AI, we actually started with building some principles around how we were going to engage in the space. And we really learned from others. One of the things that I like to emphasize is that we're not starting this, this race, we're not kind of the first ones there. And we want to be able to contribute back in terms of what we're learning. And so we did develop principles that included input from across the company, whether it is from the cybersecurity side, whether it's from the legal side, whether it's from our colleagues and the ethics team, which are really leaders in this space in terms of thinking about ethical work and ethical programming and the very sensitive environments that we work in. And we took all of that information and kind of distilled them into principles that would be easy to share across the company and then kind of adopted based on specific needs of projects or units or departments. So giving teams room to then adopt those and adopt them to their specific needs, but ensuring that they still fell under the broader umbrella of these, of these principles that we've identified, which frankly include like sharing your learning and sharing your successes and your failures, because we want to be there with you. And we really see that as a responsibility of anyone that's engaging with this technology.
D
And we're hoping to kind of take that to the next level as we start to share what we're learning outside of Chemonics as well.
E
And then on the other end, where.
D
We'Re learning, you know, it's a funny question to ask since tools can also help you learn very quickly on what's possible and what's out there.
E
But we're also being very proactive and.
D
Engaging with different events that are happening. The UN recently held AI for Good conference that we attended, and we really go to these events to learn and network and see what's possible and identify potential partnerships that are out there. We also look at non traditional events that Sustainable development consultant firms like Commonix might attend. For example, this year we attended south by Southwest where we wanted to understand how other sectors were leveraging generative AI and other applied technologies.
E
Because as an impact lab, we really.
D
Want to think outside of the box and learn from others and take those opportunities that others have identified and bring them to the sustainable development field. And we see that as a big part of our responsibility as a connector. Great.
B
Thank you, Saeed. Jeff, what are you seeing in the federal market? Are there for various reasons, Some of the agencies won't want to share what they're working on as much as in the aid market. But in terms of ethical standards, are you seeing each organization sort of developing their own? Are, is there any consensus building around any standards in the federal market? I'd be curious how, how that conversation, where that conversation's at and, and being framed in the federal market.
C
Yeah, we, we've, we've given this a lot of effort and thought over the last handful of years. You know, obviously folks are aware of the executive order that came out about artificial intelligence and some of the requirements that are placed on, on, on federal agencies. There's a NIST risk management framework for artificial intelligence as well that serves as kind of like a layer or two down. Again, we're talking some pretty broad, but the framework that NIST has, has provided has been incredibly useful I think for even down to the developer level on how they should be considering some checks and balances on the artificial intelligence. And then there's been the OMB guidance that turned into official language recently where it does say what some of these federal agencies have to do. So you've got these three large, you know, kind of big federal government directives, laws, frameworks that people that managers and leaders within the federal government are going to have to adhere to. And you know, Booz Allen ourselves, we're, we're adhering to those standards as well by providing use case inventories on every single piece of AI that we're doing. There's a, there's a technical side to this as well, which I would say is like a little bit more specific and like pinpointed with the mission. So those, those artificial intelligence models, they, if it's a computer vision model, there's a lot of test and evaluation technology out there that's pretty, pretty advanced right now. There are a few companies that you could bring on to a mission to do tne for your computer vision algorithm. You know, there's ways to do this. Mathematically it's, it's a Discipline that's, that's been maturing over the last four plus years at this point. There's also a lot, there's also some test and evaluation tooling for, for generative artificial intelligence capability that is coming out of some of the cloud service providers that are having, offering unique ways of providing the, the differences or the, the closeness or the precision between a prompt and a response. So all of this is through the frame, is through the idea that if you're going to deploy an artificial intelligence algorithm anywhere, there's inherent risk involved. And so on the mission level, on the, on the kind of ground level where the rubber meets the road, you're going to want to have some kind of test and evaluation rigor to apply to that. Now what I will say is that's like very specific and technical in nature. You know, these, these are quantitative metrics that are being produced about the performance of the model. And then you have these like large government, you know, directives, executive orders, et cetera. So in that kind of middle layer where you would expect, you know, this is the exact performance of the model that that is required, that that type of language is not very highly publicized at this, at this point. There are standards that I think are tolerable by every different agency as they deploy some of these prototypes and they try and understand what the risk in that truly is. But I would say that that's where we are in the evolution of deploying responsible artificial intelligence. You know, we're leaning on the, on the test evaluation side and we're going, going by a lot of the higher level standards.
B
And Jeff, what are you seeing or anticipating you're going to look at over the next couple of years? And really two tracks. One is the development of the technology itself, which is different than applying the technology to actual federal government needs and use cases. Are you seeing, do you think, is this as, is there a plateau here in the technology? And now everyone's taking a breath. Okay, genitive AI had this big step with ChatGPT or is that just the public facing side of it? Yeah, like we talked about driverless cars again in the prep 15 years ago. 10, 15 years ago, I wasn't sure if today anyone even be driving a car. Everyone's still driving their cars. There's some very niche uses of driverless cars. Is generative AI the same thing? 10, 15 years from now, everyone will be using it as their copilot, as Microsoft uses. So, so yeah, Jeff, technology, what are you looking for? And then also just as it's, as it's being applied to federal use cases. Love to hear your thoughts on that.
C
Yeah, I mean I think you're going to see, just as Saeed had pointed to and even some of your previous guest, I think you had the chief innovation officers of AID a few sessions back. You know, that there's going to be a natural draw, I think toward the enterprise tooling which as Saeed mentioned is reducing some of the burden of the, of the tactical day to day tasks of you know, agents, operators, staff members, you know, and so they can focus a little bit more on some of the human required tasks that their agency requires them to do that that side of it. With regard to development, there's still going to be a challenge when it comes to the amount of compute and power and access to some of these cloud service providers to be able to run these LLMs. So that, that part of it, I think from a development perspective is certainly a challenge that whether it's the, whether USAID is, is, is going to do some, some subsidizing of that compute or local governments are going to do subsidizing of that compute or there's some other creative way perhaps with some high performance computing to reduce some of that compute requirement for, for LLMs. I think we still have to cross that threshold and technical capability. But you know, I say that in terms of generative AI and development. What is I think interesting about that and what could be a challenge or potentially a temporary blocker to realizing the full capability of it is there's a lot of mature artificial intelligence right now, for example, in computer vision where that technology can be placed in an environment that doesn't have a lot of power, connectivity or compute that, you know, there are, I think you had some colleagues on your show before that were mentioning some work about Kansas State and doing crop analytics. You know, we've done some prototypes and crop analytics and done some research on bringing drones and making small and medium sized landholder farmers be able to understand their crops a little bit better and deploy labor a little bit more efficiently. I think that's the kind of artificial intelligence that I, I think myself and a lot of my colleagues at Booz Allen are seeing that is going to be able to be a little bit more transferable to the development space, especially when it comes to some of the mission use cases. You know, there are, there's technology out there right now where you can put a box that trains a computer vision model in the field and you're able to deploy that model onto a drone or onto a, another device that is going to ingest data, send it somewhere, bring it back, retrain that model all in the same place with limited power, compute and connectivity. So there are these like, you know, mission focused, mature artificial intelligence capabilities that I think are primed for some of the development use cases.
B
Saeed, I'd love to hear your thoughts on that. And then as we're getting here to wrapping up what you think the so what is for both USAID and USAID implementing partners?
E
I think the answer actually to both those questions is similar. And really the way that we're thinking about this is around what opportunities are there to leapfrog some of these challenges that we're seeing in the countries that we're working in. Generative AI, we believe, is a foundational technology that has really transformed how you could work in this sector. And we believe that Chemonix, other implementing partners and USAID have an exceptional role to play in kind of democratizing access to that technology and thinking through how it could be leveraged by communities around the world in the most efficient and effective way possible. So really it's an exciting opportunity for us to think differently about how we go about working in this space. Our role as a company and as an organization and USAID's role as an organization around the world and thinking through what does this mean for the communities now that they have access to this powerful tool? How do we make it accessible? How do we provide the framework so that it's used safely and that they understand the risks, but then continue to emphasize the importance of the human aspect to all of this. And we really do view it from the human plus AI perspective. The challenges that the world is facing are incredibly complex and I'm not sure where we are right now is going to help us resolve all of those challenges without critical understanding from humans. And I think we're very far away from that point and so really freeing up some mental space for our teammates around the world and for communities around the world to really be leveraging these tools in a way that allows them to focus on what's most important is really exciting to us and is really making us think very differently about how we go about doing business around the world.
B
Thanks, Saeed. Jeff, same question to you. What would your. So what, what's the takeaway you'd like to leave with again, USAID and their implementing partners on as they think about AI applications and, and how to use that to improve development?
C
Yeah, yeah, there's, you know, the, the federal government, as I had mentioned, has, has been at this for I, I don't want to sell anybody short at least five years overall, you know, five plus years. And there's been, you know, these are really difficult public sector use cases that, that these federal leaders are trying to tackle and they're trying to apply artificial intelligence in a way that's, that's prudent, you know, taking in the reward and the risk factors. You know, a lot of this work was done initially with the federal government under government purpose rights, which means that a lot of the technology that's been built over the last call it five years or so can be transferable and an organization in the US government like USAID or Export Import bank or whatever in the development space can use it. Folks can pick up a phone call, ask about some existing technology and use some of that technology as a starting place. There's plenty of providers out there that can provide some really excellent high end software and artificial intelligence capability for kind of enterprise, enterprise level, big, big jobs and, and that can be really great for certain federal agencies, some of the smaller agencies with less of a budget. You know, these prototypes are, are pretty critical I think for not only just to demonstrate the viability of the technology which I think would, would probably be pretty successful, really just about getting a lot of the, a lot of some of the managers who do have a, a risk aversion to it for good reason get to understand, okay, this is the exact risk that we're looking at and this is the reward. And if, and look if it doesn't fit for the mission use cases for aid or any of the commitments that they've made to certain regions of the world, then there's no pressure to use it. Because things I think are the standards that some of these implementers have in place at the moment are great and they're working. But if there is an opportunity for artificial intelligence through some of these prototypes that yeah, I think that's a really great option, that's somewhat low risk.
B
And Jeff, I think that sounds great. The partners, both the large implementers like Chemonix, the local organizations, they know what the need is. I think that's the important thing is stay focused on the need, not how can we use ChatGPT on this program, but what's the development impact need? What's the results you want to come from this? Now let's assess this variety of tools for that. Jeff, thank you. What's the best way to contact you and your team?
C
Yeah, sure. I mean the, the website at Booz Allen Booz Allen.com has got plenty of different use cases. AI.bah.com is another. AI.bah.com is a, is a great website as well. And then there's a location in the, in downtown D.C. that Booz Allen has. It's called the Helix. There's lots of examples of artificial intelligence and how we have brought that to the table across some federal agencies. There's, there's a decent amount of defense, there's a decent amount in health, more specifically for the va. So, you know, there's a lot of different ways that folks from the development space, I think, could come to the Helix, check it out. One can make an appointment and sort of have those conversations with some of the technical experts that we have and maybe make that leap from. Okay, well, this can work here. Perhaps it can work here as well.
B
Thanks, Jeff. Saeed, what's the best way for partners to get in touch with you and your team?
E
Yeah, we try to stay active on LinkedIn so folks can be more than welcome to add me on LinkedIn. It's a forum that we hope to use even more as we continue to.
D
Explore in this space.
E
And we have about 6,000 colleagues around the globe. I feel like I can't go anywhere without running to somebody that works for Commonix or that worked for Commonix at one point or another. So feel free to reach out and connect with them. That's one of the most amazing things about Chemonix is just how many current and former colleagues we have and how that just creates so much opportunity for positive change around the world.
B
Absolutely. Well said. And Jeff, thank you. Be sure to check out our recent episode with the CEO of Comonics, Jamie Butcher, as well as, as Jeff mentioned, a episode from earlier this year with with the current and former Chief Innovation Officers at usaid. And this certainly will be a topic we'll continue to highlight on this show. So Saeed and Jeff really appreciate your time, your expertise, and the important work you and your teams do every day. So thank you very much for joining the Aid Market podcast today.
E
Thank you. That was great. Thank you.
B
Have a great day. Thank you.
C
Thanks for having us. Appreciate it.
A
Thank you for tuning in to the Aid Market podcast. If you enjoyed today's show, be sure to subscribe wherever you get your podcasts and connect with Mike Shanley on LinkedIn to stay updated on the latest USAID funding trends.
Release Date: October 1, 2024
Guests:
This episode explores the advancement and practical application of artificial intelligence (AI) in both the broader U.S. federal market and within USAID and the international development sector. Host Mike Shanley leads an insightful discussion with AI leaders from Chemonics and Booz Allen Hamilton, focusing on key trends, case studies, best practices for piloting and scaling AI, ethical and partnering considerations, and predictions for what's next. The discussion aims to help federal and aid sector professionals understand how to approach AI integration, with an emphasis on actionable insights, ethical frameworks, and lessons learned from both large and small-scale implementations.
[01:46]
Quote:
“...you’re seeing this spectrum of big efforts on the big money, big side, and then ... small prototype efforts where some medium and small sized departments are trying to understand what the best use of artificial intelligence is.”
— Jeff Bhutto [01:59]
[05:07]
Quote:
“...the wide range of applicability that I think anyone could really see when they initially use that technology... just inspired a lot of people with regard to some of these federal mission use cases.”
— Jeff Bhutto [05:55]
[08:12]
Quote:
“...you don’t have enough years in your lifetime to read everything the UN is pushing out on a specific country. So I’ve seen our teammates around the world use different gen AI solutions to pull out those insights very quickly...”
— Saeed Uri [10:07]
Productivity Tools:
[14:19]
Quote:
“...we’re very much in the think big phase, but... we’re also starting very, very small… starting small with minimum viable products that will then allow us to learn and scale as quickly as possible.”
— Saeed Uri [14:21]
[16:08]
Quote:
“...we actually started with building some principles around how we were going to engage in the space. And we really learned from others ... sharing your learning and sharing your successes and your failures, because we want to be there with you.”
— Saeed Uri [16:08]
Quote:
“...the framework that NIST has provided has been incredibly useful ... even down to the developer level...”
— Jeff Bhutto [19:42]
[24:16]
Quote:
“...I think that’s the kind of artificial intelligence that... is going to be able to be a little bit more transferable to the development space, especially... Mission use cases.”
— Jeff Bhutto [26:31]
For USAID, Implementers, and Partners:
Quote:
“Generative AI, we believe, is a foundational technology ... Chemonics, other implementing partners and USAID have an exceptional role to play in democratizing access ... and thinking through how it could be leveraged by communities around the world in the most efficient and effective way possible.”
— Saeed Uri [27:36]
Quote:
“...If there is an opportunity for artificial intelligence through some of these prototypes that, yeah, I think that's a really great option, that's somewhat low risk.”
— Jeff Bhutto [31:35]
On ChatGPT’s federal shockwave:
“I think it just inspired a lot of people... there’s this broad level capability that generative AI is bringing.”
— Jeff Bhutto [05:55]
On “local” AI adoption in development:
“We're really able to put the power in the hands of our teammates in those countries ... you really leveled the playing field.”
— Saeed Uri [10:44]
On ethical responsibility:
“We really see that as a responsibility of anyone that's engaging with this technology.”
— Saeed Uri [17:05]
On the need-led approach:
“Stay focused on the need, not ‘how can we use ChatGPT on this program,’ but what's the development impact need?”
— Mike Shanley [32:02]
| Segment | Timestamp | |------------------------------------------------------|-------------| | Federal AI landscape & examples | 01:46–05:07 | | ChatGPT as a watershed moment | 05:07–07:51 | | Chemonics’ early AI examples (Syria, anti-poaching) | 08:12–10:07 | | Generative AI for insight extraction/reporting | 10:07–13:29 | | Productivity tools in practice | 12:27–13:27 | | AI roadmap: Think big, start small | 14:19–15:04 | | Ethics, principles, partnering | 16:08–19:15 | | Federal ethical frameworks (NIST, OMB, exec orders) | 19:42–23:15 | | Tech/tool plateau & sustainable AI | 24:16–27:22 | | Leapfrogging, democratizing access, so what? | 27:36–29:25 | | Prototype-first, transferable solutions | 29:41–32:02 |
The episode underscores AI’s transition from hype to practical, scalable solutions within both the federal and international development sectors. The panelists advocate for small-scale prototyping, cross-sector learning, ethical guardrails, and anchoring AI to real-world development needs. As AI tools democratize both insight and productivity, both risk and opportunity are heightened—making active collaboration, vigilance to ethics, and human-plus-AI approaches more critical than ever.
This is essential listening for anyone planning to navigate U.S. government or USAID-funded projects in the AI era.