Challenges in areas ranging from education to the environment, gender to governance, health to housing don’t exist in a vacuum. Each month, Abt experts from two disciplines explore ideas for tackling these challenges in our monthly podcast, The Intersect. Sign up for monthly e-mail notifications here. Catch up with previous episodes here.
In this month’s episode, Abt experts Jill Berkowitz and Sung-Woo Cho discuss how machine learning can amplify our work combatting malaria—25 million people protected to date—by applying what we’ve learned to other infectious diseases. We hope you'll join us at The Intersect:
Read the Transcript:
Eric Tischler: Hi and welcome to The Intersect. I'm Eric. Abt Associates tackles complex challenges around the world ranging from improving health and education to assessing the impact of environmental changes. For any given problem, we bring multiple perspectives to the table. We thought it would be enlightening and maybe even fun to pair colleagues from different disciplines to share their ideas and perhaps spark new thinking about how we solve these challenges.
Today I'm joined by two of those colleagues, Jill Berkowitz and Sung-Woo Cho. Jill works out of our international division. She's currently the technical lead for a global project-wide rollout of a DHIS 2 based monitoring and evaluation system for the President's Malaria Initiatives VectorLink Project. That system will be integrating different data streams for vector controlled decision making.
Sung-Woo currently oversees the machine learning applications on a Department of Labor funded project that's creating a better understanding of career pathways and the employment trajectories of individuals. He's also helped create and oversees the Abt Data Science Fellowship, a company-wide initiative that trains staff members in machine learning programming. Thank you both for joining me.
Jill Berkowitz: Thanks for having us.
Eric: Sung-Woo, you want to briefly describe the thinking behind the Abt Data Science Fellowship?
Sung-Woo Cho: Sure. The Abt Data Science Fellowship was designed to train up the people that we already have here at Abt who are already subject matter experts in whichever fields that they're working on and train them in Python specifically for machine learning applications. And so Python is a programming language that's extremely flexible across both text and numerical data. And so we have the structured training for them that lasts about close to two months and they go through 32 hours plus of instruction. They apply that instruction to their own data and their own projects. And in the future, we hopefully will ramp this up to somewhere near about a hundred people over the next two years will go through some type of training.
Eric: And so the idea is they're going to have these skills and they can apply them to their individual projects. How are we thinking that's going to expand our capabilities?
Sung-Woo: Exactly. The mindset is that instead of having just a few people with AI knowledge across the company, we'd have a large swath of people who have that knowledge, can incorporate that knowledge directly onto the data and the projects that they know intimately well. And so we think that this is one way to expand our AI capabilities by merging the subject matter expertise that we already have with AI know-how by using Python.
Eric: Oh, very cool, and that sort of ties in, I guess, with what Jill is doing because talk about expertise and a need to assess vast amounts of data. Jill, you want to tell us about what you're doing in VectorLink?
Jill: Sure. The VectorLink Project, as you said, is funded by the US President's Malaria Initiative, which is an interagency initiative led by USAID in conjunction with the CDC. And so we're currently working across 25 countries in Sub-Saharan Africa as well as Cambodia, and we're equipping countries to plan and implement safe, cost-effective and sustainable vector control programs; all with the overall goal, of course, of reducing the burden of malaria.
And so our primary intervention in a lot of these countries is indoor residual spraying. But what's really exciting under the VectorLink Program is that we have an expanded scope where we're thinking about vector control more broadly. And so we're starting to incorporate insecticide-treated net distribution campaigns, looking at new vector control products that are coming onto the market and how we can apply those in the countries where we're working. And all of the vector control strategies that we undertake are supported by extensive entomological surveillance that we are implementing across these 25 countries.
And so just to give you a sense, in 2018, we sprayed in 14 countries, we covered 5.8 million households and we protected over 21 million people. And so we have individual structure-level details about every single one of those almost six million structures. And we have all of that data at our disposal and we're trying to figure out how we can take advantage of that to really guide the next steps and make informed vector control decisions.
Eric: Yeah, that's a lot of data to manage. Sung-Woo, what do you think when you hear about the goal of how do we assess all of this and I guess be more efficient with the use and expand the use of that data?
Sung-Woo: Yeah. I mean whenever I hear that there's a situation where there's a lot of data available for use, my immediate thought is towards machine learning applications in machine learning. And so when you talk about indoor residual spraying in locations for spraying and getting a more pinpoint version of where we ought to be spraying, my mind immediately jumps to some of the things that we could do in terms of geospatial mapping and incorporating machine learning applications onto a mapping grid so that we can try to better predict where some of these outbreaks might occur in the future. And so these are things that people have been doing outside of our industry, I think, for a while in terms of incorporating a geospatial mapping with machine learning and AI. And so I think that that's something that we can try to pursue in the future.
Jill: Yeah, that would be incredible. I think, as we all know, we're working in a space where there's limited resources and so we're trying to have the greatest impact. And so when we make these decisions about where should we put this new net technology, where should we spray? How do we figure out where we're going to have that greatest impact? They're really tough decisions and we know that there's a lot of data streams that are required to inform those decisions. And if there were a way we could wrangle that altogether and help us to make those choices, I think that would be incredible.
Eric: What are some things we could do to actually enact that? What do we need to bring together to sort of put this into play?
Sung-Woo: For starters, with the data that we already have for a project as vast as VectorLink, and I'm sure that people have started to think about this and this is not a novel idea at this moment; but, I think that having that type of information, laying it out in a geospatial way and using all the variables that we have at our disposal. And perhaps even linking that dataset that we have with other variables outside of the things that we already have through VectorLink, and merging all that information together to try to get this "super" database in order to make our predictions of where malaria outbreaks could occur, I think that that's something that's feasible technically and maybe something that we could try to think about in the future.
Jill: It sounds like the application of AI and using some of these machine learning technologies absolutely have application on our VectorLink Program, but as we know, Abt has a broad portfolio of other international projects, whether it's ag, it's diverse health portfolio. And I know that the application of AI in these sorts of ways, it seems like it could be really powerful across our health systems work. I wonder if given Abt's footprint about how we work across 40 plus countries and we're in all different types of health program areas, Ag, climate resilience, I think that our company moving forward and increasing our AI capabilities really has endless possibilities.
Sung-Woo: I think the interesting thing about AI is that it's directly forcing us out of our silos across different topic areas. Because we're learning more and more through the way that we train these algorithms and clean the data prior to it when we train these algorithms is that these data, whether it's an international or domestic, they pretty much are similar enough that we don't really have to think of ourselves as working within specific subject matter silos. I think that that's also the interesting part of it. And I think that the international work is really leading that, especially with regard to geospatial data and to more 3D-esque data rather than the large administrative datasets that we've largely been accustomed to working on in our domestic lines of work.
Eric: I feel like you guys are implying something, so let me put some words in your mouth and you can then correct me. Jill, when you're talking about working across our divisions, across different areas of health, are we talking about applying machine learning and AI, say in malaria, and then applying those findings to Zika or the flu? What's reasonable? What's a reasonable expectation in terms of amplifying the effects and use of machine learning?
Jill: Yeah, absolutely. I think that the experience that we can garner under one specific technical area or one specific health area only opens doors about how we can apply those same fundamental principles to all of these other global health challenges, so you think Zika, you think flu.
Eric: Oh, that's very comprehensive.
Jill: But even, I know that now we're concerned about famine and early warning systems for famine. All of what we can accomplish under one specific project will only improve our ability to support a broader range of program objectives in a more integrated way.
Eric: Yes, I was nodding vigorously. I think what we're saying here is we're developing the capability to solve all health issues, right? Can I get an amen?
Jill: The potential is there.
Eric: The potential is there.
Sung-Woo: Yes, yes. The potential is there, I agree.
Eric: Excellent, excellent. Well, that seems like a good note to end on. Jill, Sung-Woo, thank you both so much for joining me.
Jill: Thank you, Eric.
Sung-Woo: Thank you.
Eric: And thank you for joining us at The Intersect. Jill, Sung-Woo and I were recorded live at Abt Studio One in Rockville, Maryland.