AI for Good: clean water access in Africa

Chandler McCann tells Daniel and Chris about how DataRobot engaged in a project to develop sustainable water solutions with the Global Water Challenge (GWC). They analyzed over 500,000 data points to predict future water point breaks. This enabled African governments to make data-driven decisions related to budgeting, preventative maintenance, and policy in order to promote and protect people’s access to safe water for drinking and washing. From this effort sprang DataRobot’s larger AI for Good https://www.datarobot.com/ai-for-good initiative. Sponsors: DigitalOcean https://do.co/changelog – DigitalOcean’s developer cloud makes it simple to launch in the cloud and scale up as you grow. They have an intuitive control panel, predictable pricing, team accounts, worldwide availability with a 99.99% uptime SLA, and 24/7/365 world-class support to back that up. Get your $100 credit at do.co/changelog https://do.co/changelog. The Brave Browser https://brave.com/cha913 – Browse the web up to 8x faster than Chrome and Safari, block ads and trackers by default, and reward your favorite creators with the built-in Basic Attention Token https://basicattentiontoken.org. Download Brave https://brave.com/cha913 for free and give tipping a try right here on changelog.com https://changelog.com. AI Classroom https://datadan.io – An immersive, 3 day virtual training in AI with Practical AI co-host Daniel Whitenack. Get 10% off using the code PRACTICALAI10. To learn more and purchase tickets go to datadan.io https://datadan.io/. Fastly https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorship – Our bandwidth partner. Fastly powers fast, secure, and scalable digital experiences. Move beyond your content delivery network to their powerful edge cloud platform. Learn more at fastly.com https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorship. Featuring: Chandler McCann – LinkedIn https://www.linkedin.com/in/chandlermccannChris Benson – Website https://chrisbenson.com, GitHub https://github.com/chrisbenson, LinkedIn https://www.linkedin.com/in/chrisbenson, X https://x.com/chrisbensonDaniel Whitenack – Website https://www.datadan.io/, GitHub https://github.com/dwhitena, X https://x.com/dwhitenaShow Notes: Worldwide Water Access: Tapping into a Well of Data https://www.datarobot.com/blog/ai-for-good-with-datarobot-saving-the-world-one-use-case-at-a-timeAI for Good with DataRobot: Saving the World One Use Case at a Time https://www.datarobot.com/blog/ai-for-good-with-datarobot-saving-the-world-one-use-case-at-a-timeAI for Good: Powered by DataRobot https://www.datarobot.com/ai-for-goodDataRobot https://www.datarobot.comUpcoming Events: Register for upcoming webinars here https://practicalai.fm/webinars!

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Being with your change log is provided by Fastly. Learn more at fastly.com. We move fast and fix things here at Change Law because of Rollbar. Check them out at rollbar.com. And we're hosted on Leno cloud servers. Head to leno.com slash change log.

• 0:17Speaker 00

  This episode is brought to you by Digital Ocean. Digital Ocean's developer cloud makes it simple to launch in the cloud and scale up as you grow. They have an intuitive control panel, predictable pricing, TM accounts, worldwide availability with a 99.99 uptime SLA and 24/7 365 world-class support to back that up. Digital Ocean makes it easy to deploy, scale, store, secure, and monitor your cloud environments. Head to do.co slash change log to get started with a $100 credit. Again, do.co slash change log.

• 0:59Speaker 01

  Welcome to Practical AI, a weekly podcast that makes artificial intelligence practical, productive, and accessible to everyone. This is where conversations around AI, machine learning, and data science happen. Join the community and Slack with us around various topics of the show at changelog.com/community and follow us on Twitter. We are at Practical AI FM. Okay, take it away, guys.

• 1:24Chris Benson?

  Welcome to another episode of the Practical AI Podcast. This is Chris Benson speaking. I'm a Principal AI Strategist at Lockheed Martin. And with me as always is my co-host Daniel Whitenack, who is a data scientist with SIL International. How's it going today, Daniel?

• 1:40Daniel Whitenack?

  It's going great. I'm already been a busy Monday with a lot of prep for training stuff that I'm doing and, you know, also training people and training models. That's been my day so far. So that's a pretty good day, I guess.

• 1:54Chris Benson?

  I'm not surprised. You are a ferociously busy person as I have known with between your classes and your day job and the podcast and and I know your wife has a business which you help out in. So

• 2:06Daniel Whitenack?

  Yeah, it's actually over the weekend. We were rearranging stuff in her factory there to make sure when some essential people come back that they're six feet away and all that good stuff. So yeah, it's been a range of things over these weeks, which makes things interesting. That's for sure.

• 2:24Chris Benson?

  Gotcha. Well, I'm here in Atlanta, Georgia, and we have officially opened up from sheltering in place, but I am I'm more cautious than that. And I expect we're going to keep doing it for quite some time. But I'm looking with envy at neighbors who are having parties and stuff at this point. So I'm afraid to go over, but we'll see. So hopefully everyone stays well. So, you know, I wanted to talk a little bit today about AI for good topics and adjacent.

• 2:51Daniel Whitenack?

  Good timing. This is great stuff.

• 2:53Chris Benson?

  I know. And we have a pretty awesome guest for that. With us today is Chandler McCann, who is the General Manager of Data Robot for Federal. And Chandler, welcome to the show.

• 3:04Speaker 04

  Hey, thanks, Chris and Daniel. Thanks for having me. Great to be here.

• 3:08Chris Benson?

  I was wondering if you could give us a little bit of background about yourself and how you came to be General Manager at Data Robot and some your background before we dive into the topic today.

• 3:19Speaker 04

  Sure, sure. Would love to. So about myself, you know, like a lot of data scientists, I'm a recovering engineer. So my undergrad was in material science engineering and spent a few years in the flash and DRAM manufacturing space as an engineer. And then from there kind of flowed into statistical consulting, largely focused in the Department of Defense. And that's really where I really kind of fell in love with data science and got exposed to more modern machine learning techniques. And after that, we pursued my masters at Berkeley and during that time became an employee of, at that point, a fairly young Data Robot. And by the time I did a robot, Began as a customer-facing data scientist and over the course of the years have evolved

• 4:04Speaker 04

  to lead various teams, including the AI for Good program, which I still oversee and now are federal practice today.

• 4:13Daniel Whitenack?

  And I know, so just in terms of Data Robot and the world you live in, I know when I was starting out in data science, it was definitely very much the Wild West. There wasn't a lot of, you know, platforms or systems to manage your data science work and other things. And I remember distinctly kind of going to conferences frequently, but then there was like a certain year where I remember there's all these companies sort of popping up in this space. Data Robot was one of those early ones that I remember popping up and just kind of being consistently present in the in the data science world. Maybe just mention kind of the premise behind how Data Robot got started and what you do.

• 4:52Speaker 04

  Sure. Yeah. So Data Robot, and I actually found Data Robot through one of those conferences at Strata in 2016. That's a funny story, but you know, Data Robot was born from our co-founders, Jeremy H. and Tom Diggodoy, who were early CAGLers in addition to working in the insurance space. But you know, during their CAGL days, they realized that they could benefit from automation. They could automate a lot of the tasks that data scientists were doing, not to replace them, but to just streamline the workflow. So, you know, following some funding, Data Robot and automated machine learning were born. So today, you know, that was back in 2012. Today, Data Robot is a full end-to-end enterprise AI platform. And

• 5:35Speaker 04

  that means really going from the whole value chain of data, from I have an idea and a, you know, raw unstructured data set, all the way through model building and monitoring and deployment and even consumption into automated applications. Data Robot provides a very high level of automation all the way across the spectrum. So it's been a lot of fun to be part of that ride.

• 5:58Daniel Whitenack?

  And what do you see? I'm kind of curious as to what you see about the receptivity of data scientists and AI people towards automation, maybe now as opposed to like, like when you joined in 2016, it seems like, you know, there has been a shift that I've sort of observed, but I was wondering, you know, what are your conversations about automation? How do they typically go down when you're talking to a data science team or to even a software engineering and data science team combined? What's the feeling about automation these days and what should be automated, maybe what shouldn't be automated?

• 6:35Speaker 04

  Yeah, that's a great question. I think to your point, it's definitely shifted. You know, I think there was more resistance earlier 2016, 2017 phase where there weren't that many full-on enterprise auto ml platforms. But I don't think the world's ever really looked back from automation, right? If you can't get anything from iPods to digital cameras, it's hard to reverse that. And I think data scientists are kind of coming to grips with this is not something that's replacing me, but it's something that's augmenting my workflow. So while, you know, Data Robot certainly has a complete GUI-based platform for our users to do or to work within for our advanced data scientists, they mainly

• 7:17Speaker 04

  interact with our API in Python or R. And that just scales a data scientist in a way that's just not feasibly tractable otherwise, right? So a data scientist can quite comfortably build, manage and deploy, you know, potentially thousands of models by themselves in a way that, you know, manages risk and is still interpretable. So I think that's appealing to most data scientists that get it.

• 7:43Daniel Whitenack?

  So I had noticed that, that prior to you moving into the federal practice, that you had been a practitioner and probably still are a practitioner, you know, as a data scientist and kind of built up your career that way. I'm kind of curious, as you get into federal, you know, what have you discovered about that? And I, obviously, working for Lockheed Martin, I have an interest in federal. So I was just curious about your perspective as you've moved into that role.

• 8:07Speaker 04

  Yeah, I mean, I think within the federal space, there's a lot of opportunity and I'm intrigued by helping the federal government and particularly Department of Defense leverage automation in a way that, you know, kind of unlocks the potential of AI in organization. I think a bottleneck has been the ability to get human talent, right? So acquiring data science talent has been historically really competitive and that's can be a challenge for the government. So tools like Data Robot really help up level an organization. Um, so that's the one way I see us helping a lot. And two, it's just being able to solve problems that were typically very hard and intractable before.

• 8:46Speaker 04

  So, you know, Data Robot, in addition to just, you know, time series and classification regression, um, also handles visual AI problems now and computer vision issues. So being able to bring that to fair in, in, in the broader marketplace and the federal government, it's been fun to watch.

• 9:02Daniel Whitenack?

  And how does someone working on the sort of these sort of federal problems at Data Robot? How do you get routed to this sort of AI for good effort? Is that something that, you know, was a specific passion for you or kind of came up unexpectedly? Or how did that happen and what's the story behind that initiative?

• 9:22Speaker 04

  Yeah, that's a great question. So when I joined Data Robot, I was working largely on the commercial team at that time. And I had had a relationship with the Global Water Challenge, which is a nonprofit base here in DC. I had met them while I was at Berkeley and they, their mission is to help invest and manage investments in large scale water projects across the developing world. And so at Data Robot, I brought them on and talked to my CEO Jeremy about bringing them on as a pro bono customer. And he was very supportive of that. And, you know, following the work we had done together, the AI for Good program was really born.

• 10:01Speaker 04

  So I remember vividly being on a plane, coming home after a trip to Sierra Leone with Brian Banks, you know, our, our customer at the Global Water Challenge. And receiving an email from my CEO saying, you know, what if you had a, instead of doing this project by yourself, what if you had a team to help you do that? What would that look like? And from there, I was able to kind of take all the lessons learned that gathered from my discussions with Brian about the challenges that nonprofits and NGOs face, and build a program around that to address those.

• 10:33Chris Benson?

  And for those who aren't necessarily familiar with the Global Water Challenge, could you kind of talk a little bit about what that is in, you know, in general, before we fully dive into into how you were interacting with that? Just so that people who haven't heard of it before have a reference point.

• 10:49Speaker 04

  Sure. Yeah. So the Global Water Challenge is a nonprofit based here in Washington DC with the mission of helping bring water to communities in developing nations. So roughly one in four people around the world are dependent on non-traditional water sources like hand pumps or wells. And the Global Water Challenge sets out to help direct investment to countries in need.

• 11:12Daniel Whitenack?

  And so is that investment, are those projects seeking to like upgrade those water systems? Or just when you say it's not available, I guess what I'm trying to work through is it sounds like part of it is maybe clean water and good sources and part of it is just access at all? Or what sort of projects do they work on?

• 11:29Speaker 04

  Yeah, they have a wide portfolio, but the ones that we've been focusing on with DID robot have been around typically new construction and rehabilitation of water points. So for example, you have a community that may not have an infrastructure for running water. So they may look to direct investment to drill new wells or repair new pumps for people in these communities. So you'll have villages that don't have access to water and they will help direct investment with other NGOs or large corporations to either drill or build new water points.

• 12:02Daniel Whitenack?

  Yeah. And I guess this problem has probably become increasingly evident even more so over these recent times because I guess, you know, of course, disease spreads in various ways, but if people aren't able to access water and, you know, for for cleaning and and all of those things, then I'm sure it further exacerbates many things related to disease spread and and health and and a lot of different things.

• 12:27Speaker 04

  Yeah, absolutely. I mean, particularly now with the importance of handwashing, having access to clean water has been, never been more important. So yeah, the downstream effects of not having clean water in a community are huge when it even comes to impacts inequality in education as well as just overall disease. It's being able to move the needle on that has huge impacts for an economy. And I think one place that was really exciting about the Global Water Challenge and our work, what drew us to work with them, was their focus on data. So around the world, there are hundreds of thousands of these water points, but there is no centralized repository. And Brian from the Global Water Challenge set out to build the first standardized

• 13:10Speaker 04

  and normalized database for water points around the world.

• 13:21Speaker 01

  We deserve a better internet and the Brave team has the recipe for bringing it to us. Start with Google Chrome, keep the extensions, the dev tools and the rendering engine that make Chrome great. Rip out the Google bits, we don't need them. Mix in ad and tracker blocking by default. Quick access to the Tor network for true private browsing. And an opt-in reward system so you can get paid to view privacy respecting ads. Then turn around and use those rewards to support your favorite web creators like us. Download Brave today using the link in the show notes and give Tipping a try on changelog.com.

• 14:02Daniel Whitenack?

  So Chandler, I'm interested, you just started talking about the focus of data that Global Water Challenge has. I assume that that focus on data and the data that they've gathered in this repository is central to the solution that you built for them. But I was wondering if you could maybe describe that data a little bit more. What does it represent? What is the scale of it? And what sort of information is included in that data?

• 14:29Speaker 04

  Absolutely. So the Global Water Challenge and Brian Banks, the person driving this project, they set out to build a standardized database of water points around the world. And the reason they did this was because the water points kept breaking, right? So around the world, after a few years of being installed and having such a positive impact on the community, these water points would break. And they had kind of no idea what was going on and why.

• 14:58Daniel Whitenack?

  When you say water point, you're meaning like a tap or a water main or a well? Or what's included in these?

• 15:04Speaker 04

  Sure. Yeah. Typically, so when I say a water point, these can refer to a few things, but they generally fall in the categories of a well or a tap or a rain harvesting system, where they can either get groundwater or purified rainwater.

• 15:20Daniel Whitenack?

  Gotcha. And so this repository includes information about where those are at? Or what sort of information is included about those water points?

• 15:28Speaker 04

  Exactly. Yes. So as I was saying before, the challenge we were trying to solve was why do these water points, either wells or taps or rain harvesting systems break? So the data set that he set to normalize includes things around the location of the water point. So there's cell phone applications that can take a picture, capture a lat long and geolocation. So there's also image data to some degree. Then there is information on the source of the water as well as technology. So whether it's coming from a river or groundwater, or if it's a tap stand itself or a pump. And then it's got information about the country and region it's in,

• 16:10Speaker 04

  as well as the installation year, who installed it. And then some interesting factors such as are the communities paying for it? And is there a management structure in place in the community to maintain the well? Or water point itself?

• 16:25Chris Benson?

  So I'm curious what kind of solution, you know, as you got involved in this, what kind of solutions you had envisioned that might be able to help them? What was the motivation for you to get involved and for them to to work with you? I think you said it was Brian. So kind of how did that get going? And what was the vision that ended up being implemented? Where did that come from? How did it start? What was the collaboration that got all that going?

• 16:50Speaker 04

  Yeah, so the vision when we started working with Brian was was really his. So he had had a vision for this data from day one. And as this is the person who built the database, he knew they could do something with the data, but he wasn't sure just exactly what that looked like. Aspirationally, they wanted to be able to predict which water points were going to break in the future, or at least understand which ones were going to break. And in parallel, they also wanted to understand if they could identify a priority to these water points. So which communities are not being served? Where would it make sense to build or construct new water points? Because it's non trivial to set up construction of these things in a developing nation.

• 17:31Speaker 04

  You know, for me, the appeal was, wow, you know, there is this relatively clean data set on a really interesting problem that's out there with the potential for a huge impact. So that's what drew me to it. And our main focus when we set out was can we, you know, given data on which water points have broken in the past, can we predict which water points are going to break at some point in time in the future?

• 17:54Daniel Whitenack?

  Gotcha. And so, yeah, it seems like there's really a lot to tackle there. Nonprofits, I'm guessing typically have resource constraints. And so being able to understand where they should put their investment is is definitely important. But for this particular first project in terms of, you know, predicting where a water point is going to break, what is the sort of like, out of all water points in the database, how many are breaking on any given point? Like, what's the distribution like here? Is this something that happens fairly rarely? Or it's something that happens like all the time? Is it more compared to something like fraud detection, where you're trying to detect something that happens rarely? Or what's the situation like on that front?

• 18:37Speaker 04

  Sure, sure. Yeah, it's more frequent than fraud, unfortunately. So the the distribution of things that are broken is around 25%, you know, on average, compared to 75% functioning.

• 18:49Daniel Whitenack?

  Gotcha. And and that's due to just the, you know, obviously what global water challenges trying to address, just the the old systems and systems that aren't being maintained and that sort of thing?

• 19:01Speaker 04

  Yeah, there are a slew of potential reasons, some of which could be maintenance, some of which could be environmental. You know, perhaps a water point wasn't dug deep enough so you have a well that Becomes dried out, you know, six months out of the year during the dry season. So that could have an impact. So there's geographical inputs, there's community-based inputs as well as maintenance-based, you know, kind of failure modes that are out there.

• 19:26Daniel Whitenack?

  Gotcha. And so when you first saw this data and what was included in it, where did your mind go in terms of, you know, an approach that you could take to solving this?

• 19:37Speaker 04

  So when I first saw the data set, I was at one time impressed by how standardized it was, but at the same time digging into it, realized there were a lot of nuances and challenges. And I'm sure, Daniel, working with nonprofit data yourself, you have been exposed to this. But whenever you're dealing with human data collection, there's always some challenges that are out there. And one of which, on our side, a big one, was the ability to enter free-form text for the same thing. So there's obviously a few key pieces of data to solve this problem, particularly what type of technology is it? So is it a pump? What brand of pump is it, for example? Those things all matter.

• 20:19Speaker 04

  And when we started looking at the data set, there was roughly 1,600, you know, unique values for the type of technology when we knew that it really boiled down to about 12 or 14. So one of the first problems that we tackled with that was, you know, just some basic natural language processing to try to match categories together. And that was something that we had done by hand originally. And today, we're actually automating that process now through the use of Paxata, which is our automated data prep tool. So that's been kind of a big step forward as we move from this sort of, you know, version 1.0 of the solution to kind of a version 2.0 in the future.

• 21:01Speaker 04

  So, and back to your original question, my first thought was, we need to kind of organize and clean the data. And the second one is, how do we frame this to make it a useful problem, you know, down the road? And so we had to identify what variables would be really important to this. And a couple of things jump out that are available to us, namely, location, the age of the water point, as well as the technology and the source and the community interaction with it. So from there, we built our first predictive model.

• 21:30Daniel Whitenack?

  Gotcha. And you did mention the sort of problems with data and like in the nonprofit world and people sort of humans gathering this data, which I guess isn't also specific to the nonprofit world. But I know like for us, a lot of times it's hard, especially in developing countries, you know, you wonder about like, if I want this data, but it was the only access I have to that type of data was data that was gathered, you know, four years ago or something. You wonder about, you know, what's updated since then. So in this case, like, how is this data being generated? Is it people just going out into the field and like marking down where the water points and that sort of thing are? Are there actual instrumentation on some of this stuff?

• 22:11Speaker 04

  Yeah, that's a fantastic and very important question. And it was subtle in the data. So that was one of the first things I asked Brian is how does this data come about? And there's a couple of different ways. One way is from large national efforts, like national assessments. And this is something that we came across in countries like Swaziland and Sierra Leone. But also you have manual input by smaller groups like local NGOs or local nonprofits that are uploading it. So what kind of was the magic behind Brian's idea was he was going to build a standardized way of capturing this information. And so no matter how they were keeping their own records, they had a common format to upload it in that maintained these key fields.

• 22:55Chris Benson?

  I'm curious, as some of that data was coming in, you know, we talked a bit here so far about the textual data. I remember earlier in the conversation, you mentioned something about images as well. Did you have a mixture of different types of data? Like, was imagery used as input? Or was it mostly text oriented?

• 23:13Speaker 04

  Yeah, so the image data has always been there. So there's, you know, S3 buckets or Dropbox files that are storing this image data. But we really haven't leveraged that much until, I guess, within the last two months or so, a data robot. So at first, our original models were contained, but text, numeric, and categorical data. And as we've expanded, we've begun to integrate image data as, like I said, we've released our visual AI platform and data robot, which allows us to incorporate images into the modeling, which is something we're currently exploring and is really interesting. And in fact, kind of a interesting use case from that is, you know, if I have an image that's uploaded, but perhaps someone forgot to fill out the field of the technology, can

• 23:58Speaker 04

  we train the classifier to say that this image is actually a hand pump or it's a rainstand? So can we use, you know, image analysis as an intermediate step into, you know, data augmentation and cleaning?

• 24:12Daniel Whitenack?

  Yeah, so on that front, you know, at least in, let's say, version one of what you did in predicting these water point failures, after you had done some of this NLP and you started getting into thinking about how to predict these failures, what ended up being a good, a good way to do that or a way that you found out how to do that? And what portions of the data ended up being good predictors of that behavior?

• 24:40Speaker 04

  So we're looking at the problem, we realized that there was some country to country variation, but some common things popped out. So the data that was, turned out to be the most predictive kind of across the board was, you know, the age of the water point. So certainly, its, its function over time was certainly dependent on, there was a strong relationship to how old it was, who installed it. So the, whether it was from a private government or, or sometimes a nonprofit was also predictive in certain areas, as well as also strong local effects. So we saw things like, you know, the region of the water point having a relationship to life. So places that were far away from, say, the, the large city in that country may have

• 25:25Speaker 04

  low access to parts and in some places, you know, They would tend to have a shorter life, all else being equal. So those were some of the things that kind of jumped out at us.

• 25:35Chris Benson?

  I'm kind of curious, as you were engaging in this process and recognizing some of these constraints that you've talked about, was this particular engagement, you know, in this kind of AI for good charitable approach, was it more or less the same as other data science projects in terms of you're still getting data and you're doing the normal prepping the data and running it through your model? Or was there anything in your mind that distinguished it as something unto itself? Something a bit different from your typical business scenario that you might otherwise be engaged in? I was kind of curious if they were essentially all the same, or if there was something that made that stick out from a process standpoint?

• 26:15Speaker 04

  Yeah, so when we began engaging with the Global Water Challenge, we were just treating them like a regular customer. You know, there was no AI for good program formally at that time. They were just being treated like a regular customer. And in our job at Data Robot as, you know, at the time, as customer-facing data scientist, was to enable them to own their own solutions, right? So that involved kind of teaching them how to fish, right? Working with Brian to, you know, help frame his problem better, understand the data with him, and then talking through all the blind spots and gaps in the modeling process that would come up along the way, along with helping him interpret his model. So in that sense, it wasn't unique from a process perspective, but it was unique in

• 26:58Speaker 04

  sort of the level of access to the data that I was, you know, I could get with clients. So with Brian's data, you know, it was side-by-side partnership. Everything was available to me and obviously there's can be restrictions when you're dealing with, you know, certain private companies when it comes to the level of access to the data you can get. So that was nuanced, but I think the takeaway for me and what really helped us when we built the AI for good program was that, you know, if we treat these nonprofits just using the same process we do with our customers, they can own and build these solutions over time themselves. And that was something that was really inspiring to me.

• 27:45Speaker 05

  What's up? This is Daniel Whitenack, one of your Practical AI co-hosts, and I hope you're enjoying this episode and staying healthy during these crazy times. I'm working on some pretty cool AI stuff here from my home office, but I've also found that I'm having to get a bit creative and be intentional when it comes to honing my AI skills and virtually connecting with the AI community. If you're in a similar situation, or you've been inspired by the Practical AI we talk about on this show, I want to invite you to a live, online AI training event I'm hosting this May called AI Classroom. In AI Classroom, I'm going to teach you the practical skills I've learned over the years using the latest open-source AI technology.

• 28:28Speaker 05

  You'll learn AI theory along with practical, hands-on implementations in both PyTorch and TensorFlow, and after the training, you'll be able to understand the latest AI models, implement your own models in code, train computer vision and NLP models, create model inference servers, and experiment with state-of-the-art methods like reinforcement learning. AI Classroom is taking place this May. It'll be taking place live and completely online in a high-quality virtual classroom, so no travel is required. There'll also be two cohorts with convenient time zones for Eastern and Western hemispheres. Don't miss out. Tickets and more information are available at datadan.io, that's datadan.io,

• 29:13Speaker 05

  and Practical AI listeners can use the code PRACTICALAI10 for 10% off. See you online in AI Classroom.

• 29:32Daniel Whitenack?

  As we get into, I guess, you know, how this inspired, you know, more AI for Good efforts at Data Robot, and also like your learnings from how to work with a nonprofit and that sort of thing, I would love to hear about like, I guess, where this project ended up in terms of, you know, positive or negative results, and then how that inspired further work. It sounds like things have expanded past that, so I'd love to hear about that story.

• 29:59Speaker 04

  So where it ended up with the Global Water Challenge was... Was really interesting and to be fair, it's still an ongoing story. So during, I guess, two thousand nineteen, we're able to go to both Sierra Leone and Liberia. First, just with Data Robot to Sierra Leone and the second time to Liberia on behalf of the State Department, where I was asked to be a part of the Water Expert Program and I went with GWC to participate in a water data workshop in Liberia.

• 30:28Daniel Whitenack?

  That's great.

• 30:29Speaker 04

  It was very cool and it was just an awesome experience from both, you know, data science perspective where I'm working over here, I'm pulling data, you know, out of a table, to actually going on the ground and meeting the people who are collecting it and having conversations with them and trying to communicate the power of machine learning and the importance of the data they were collecting and how it could be used was just, you know, a humbling and awesome experience all wrapped up into one. So following those two trips, we had very positive relationships with the government in Sierra Leone. In two thousand nineteen, the Ministry of Water, you know, kind of reaffirmed their commitment to evidence-based decision and actually passed a national policy requiring the use of data and decisions

• 31:14Speaker 04

  about water services, which is pretty cool and also, again, humbling. If you think about it, I think the story of this project is the story of the power of data and what it can do. And if you think about it, in two thousand eighteen, you know, the use of data at the national level in Sierra Leone to kind of inform decision-making by policymakers, again, with very constrained budgets, was very low. And then in two thousand nineteen, we do know that working with the Ministry of Water, they're able to use some of the insights from our tool to inform decision-making and budgeting that year. So that was definitely, you know, exciting and a near-term win for us in the project. And that helped kind of...

• 31:59Speaker 04

  Shift things out of this R and D phase to where Brian and the Global Water Challenge are in the process of pursuing more funding. And if you want to contribute, feel free to them. But we're looking to build a much more sustainable tool that can be deployed to many countries around the world. So it kind of was the launching point for the project.

• 32:18Chris Benson?

  Yeah, that's a super cool story. And I just wanted to note to listeners that Data Robot on its AI for Good page has a video of you and Brian doing this work together. And I wanted to call that out and we'll include it in the show notes. It's just a few minutes long and I would urge anyone to take a look at it just to kind of, you know, get the imagery of what you went through. I think you surprised me a second ago when you were talking about, you know, being able to shift national policy. That intrigues me because we have the privilege of talking to people that engage in AI for good on a fairly regular basis and hear about cool projects, but most of them don't change country, a country's policy based towards evidence-based action, you know, from the work they've done. And I kind of wanted to

• 33:02Chris Benson?

  get a sense of how did that feel when you realized that not only had you had the specific impact that you had on by providing the service that you walked into the engagement with, but when you realized as well that you were actually changing the way a country was thinking about using data to affect good for its population. What was that like? I'll let you answer that and then I have another follow up.

• 33:29Speaker 04

  Yeah, I remember getting the email saying that they were leveraging the insights from, you know, our models and our, even the simple data visualizations was a huge leap ahead for them. Right? And I remember hearing that and being told that it was being used to inform the budgeting process for the following year. And I was just floored and that made me, you know, a little, I don't know if scared is the right word, but I was just like, wow, that I did not realize this was going to be happening so quickly. But at the same time, it was just another proof point that people were starving for information. Right? You have people that are trying to make decisions that impact a lot of people in their citizen base.

• 34:12Speaker 04

  And, you know, the ability to just synthesize a little bit of information can go such a long way. So I think we're just on the front end of the wave when it comes to the, the way to leverage data, you know, across these developing nations for water policy.

• 34:27Daniel Whitenack?

  Yeah, just to interject there, I totally echo what you're saying. And I know that like some of the proof of concepts that we've done around like dialogue systems and chat interfaces in emerging markets. To be honest, like the interface has not been that great. Like it probably wouldn't fly in the U S, but people are so hungry for information, accurate, good information in certain contexts that to some degree that that doesn't matter. Although we of course strive for good interfaces. But yeah, I totally resonate with what you're saying that, yeah, it just can be so powerful to be in these situations where you're creating something that allows a new view onto information that that people are so hungry for. So yeah, it's really cool. Chris, did you have a follow up?

• 35:12Chris Benson?

  Yeah, I was just, I wanted to actually reference another thing, you know, that you had done. And that was, I know you have a blog post called Why Most Tech for Good Campaigns Fail and How We Can Fix Them. And I was wondering if you could kind of address those steps that you take us through and why that works, but also, you know, if you have any thoughts about how, you know, people or organizations like yours that are trying to do these AI for good projects might be able to influence or impact policy going forward. I know it kind of surprised you in this case, but if you have any thoughts toward even how to extend this so that you think you might get systemic change for the long term and how a government thinks about this. I'd love to know that, what your advice is in general.

• 35:57Speaker 04

  Sure. Yeah. So two parts, how do we, you know? Look at delivering this in the most effective way for nonprofits and then, you know, for companies' interest in this space, how do they think about potentially impacting policy? And I'll tackle the first one. So when it comes to why we were observing a lot of these Tech for Good initiatives, and I hate to say failing, but just not delivering the results they intended, a lot of that was just based off experience from Brian and what we heard and his just life at the Global Water Challenge and his exposure to people in the sector. So the big idea is that you need to partner with these nonprofits and NGOs and help them build solutions that they can maintain. So, and hackathons are all

• 36:42Speaker 04

  well-intentioned, but it's probably not realistic to expect small nonprofit without a big data science team to maintain a code base over time, right? So, like, even if they build an app, you know, models to get stale, models need to be refreshed. So, our idea was, and we happen to have the benefit of, you know, a very powerful enterprise AI platform behind us, but just help them understand how to think through their problems, right? So how do we appropriately identify use cases that matter to them? How do we frame them? How do we think about sort of the ethical considerations for what we're doing? You know, how do we think about acquiring data for it appropriately? And then how do we go through the iterative

• 37:25Speaker 04

  model-building process? And then, you know, how do we deploy things in a way that are useful to people? So, the truth with data is that, and I'm of the opinion that value really isn't realized until it's consumed by somebody, right? So, we can build models to our blue in the face, but until someone's doing something with it, it may not be super useful outside of perhaps, you know, insights, which I would still argue is still consumption. So, you know, our process is structured to, you know, within the AI for Good program today, help nonprofits go from, this is my big vision, to how do we deconstruct this to, you know, a machine learning problem? And then how do we go from, this is my idea,

• 38:10Speaker 04

  to this is my deployed model in a structured way, and then teach them how to learn the process, you know, through each of those tool gates.

• 38:19Daniel Whitenack?

  Yeah, it sounds like good advice for really any data science program. In my opinion, a lot of those things ring true.

• 38:27Speaker 04

  Yeah, on the second part on the policy side, I hope that more companies continue to provide resources to nonprofits and NGOs. I think what's true in this space is that they may be a little bit behind than other private companies when it comes to collecting data and storing it and organizing it, but they're coming along and times are changing quickly. So the situation where you have a lot of nonprofits with potentially a lot of very rich, interesting data that aren't sure what to do with it, that grows a lot. So I would love for other companies to continue to get involved and offer their services. And I would just say that, you know, again, we just have to account for the fact that machine learning

• 39:07Speaker 04

  models are sort of living things that go stale over time and we need to help our end users build solutions that account for that.

• 39:15Daniel Whitenack?

  And I'm curious, as we get to a point of sort of closing things out, I know you mentioned earlier, you know, getting that that email, I think it was asking about, you know, what if you had more people to work with on this initiative? Where are things at right now with the AI for Good initiative at Data Robot? And what are things you're looking at in the future?

• 39:36Speaker 04

  Yeah, so that was very exciting and I can't thank Jeremy and team enough for supporting us to build out this vision. But June of last year, we sort of launched our program and opened up our application process and we actually got applications from 10 countries and five continents, which was pretty exciting for our first year of the program in its inception. Since then, we've been working with six nonprofits. So Kiva, which is obviously a large lending platform in the nonprofit space right now. So helping them predict the likelihood of which loans will go unfunded or not. Donors Choose, helping provide supplies to teachers in classrooms. And Acosta Riverkeeper here in Washington,

• 40:17Speaker 04

  DC, helping them forecast ECOI levels in the river, given some sensor data that's being read. To some very interesting healthcare use cases with the University of California, San Francisco, Spinal Cord, Zuckerberg Spinal Cord Institute, where we're looking at ways to use OR data to help improve outcomes for spinal cord surgery. And then finally, working with the University Hospital of Mannheim over in Germany, where we've been forecasting Top World Health Organization causes of death as well as predicting patient mortality given people that come into the hospital. So those are some of the use cases we're working on now and we're excited to just continue to broaden the, you know, the impact

• 41:01Speaker 04

  that people are working with and keep the program growing.

• 41:04Chris Benson?

  Fantastic. Well Chandler, thank you so much for coming on to the show and telling us about Data Robot and the things that you're doing or have done with the Global Water Challenge. It was really fascinating and inspiring and thank you for doing that work. I really appreciate it.

• 41:20Speaker 04

  Absolutely. Thanks for having me on the show. Really been fun talking to you and let me know if you guys ever want to catch up again.

• 41:28Chris Benson?

  Will do. Thank you.

• 41:33Speaker 01

  Thank you for listening to Practical AI. We appreciate your time and your attention. Next up, let your voice be heard. Please leave us a comment on the episode page. There's a link in your show notes for easy click-ins. We'd love to hear from you. Word of mouth is the number one way people find new podcasts. If Practical AI has helped you on your AI journey, please do tell a friend. Hey, they'll thank you later. Special thanks to Breakmaster Cylinder for the beats and to our awesome partners for their support. Shout out to Fastly, Linode, and Rollbar. That's all for now. We'll talk to you again next week.