E39: The AI Connector: Social and Data Sciences | Rob Krueger-Social Science and Policy Studies | Lane Harrison-Computer Science

Colleen:  What if we could make better decisions about society by combining human insight with the power of artificial intelligence? Welcome to the WPI podcast. I'm Colleen Wamback. Today we're talking about something that might sound a little academic at first, but actually affects all of our everyday lives, the social sciences and how AI is starting to change the game. At its simplest, social science is really just about people, how we live, how we interact, how communities work, and how economics, culture, and politics shape our lives. It also helps policymakers make decisions about things like public programs, resources, and overall quality of life. Now, here's where it gets interesting. What happens when you bring AI into that world? AI has the potential to look at huge amounts of data, spot patterns faster than ever, and help us better understand what people and communities actually need. It could even open the door to more voices being part of the policymaking process, making it more transparent, more participatory, and maybe even more effective. But of course, it's not that simple. There's a real balance here. How do we use powerful data and technology without losing the human context, the stories, the lived experiences, the nuance that social science depends on? That's exactly what we're going to dig into today. Rob Krueger, department head of Social Sciences and Policy Studies, and Lynn Harrison, a computer science professor, join us to talk about how AI is reshaping social science research and what that could mean for the future of policy. They'll also introduce us to a groundbreaking approach called Poverty Stoplight that uses AI to help individuals and communities address poverty on their own terms. Rob, Lane, thanks so much for being here. Rob, first question goes to you. Why and how do data science and social science matter to each other?

Rob: Well, I think the problem is that right now they don't. They should matter to each other, but they don't. Everybody is sort of playing in their own individual sandbox by the rules of their discipline or their profession and that type of thing. There's not a lot of cross-fertilization. You gotta find people on both sides who are willing to work together to learn from each other, and not just understand at a very superficial level, but really dig down deeply into how each other, each discipline or set of disciplines develops problems, how it comes up with solutions and interventions to those problems. What's really important is to understand the story around those numbers. It's understanding where they came from, how they're being used, what types of recommendations are emerging from a certain set of data, and I don't think that those critical questions are being asked at the moment. 

Colleen: Do you see AI being a connector to both data science and social science? 

Lane: Yeah, I think it's a bridge in that when we think about the social sciences, there's a lot of unstructured data. There's text, there's images, there's complex historical data, and as a computer scientist, sometimes we do things that are easy to us, that we can measure things quantitatively. One of the things that we do really often for research is we run these large crowdsourced surveys of how people engage with data, how they interact with data visualizations. We have to develop a lot of infrastructure to do this, and we have to design our survey instruments very well. Not only that, we collect a lot of complex data. We then have to analyze that data using the best available techniques. Now we have some tools where we can work with that data more effectively. Uh, so I think of it as a bridge, that there's a lot of opportunity now to cross our expertise and skill sets to make an impact.

Colleen: For people who may be skeptical, how can AI improve how we understand society? 

Lane: If we can, you know, use AI effectively, we can solve new problems, we can solve them in different ways. And if we do that well, then we can actually understand society a bit better. 

Colleen: Rob, were you skeptical? And do you understand why some people are skeptical of this in developing policy or reviewing policy, et cetera? 

Rob: Oh yeah, I completely understand why people would be skeptical. But it reminds me of, um, 1870s when economists were moving from being moral philosophers to social scientists- Mm-hmm what they called social physicists at the time. This was the same question that they were asking. If we understand the world more by doing economics in a scientific way that reflects physical methodologies from physics, what are we losing in the course of that? And as Lane was talking, I was just thinking it's really a method that one could use, and one needs to approach it that way in the sense of, what am I giving up by doing this? What am I missing? How will I know? So it really needs the same sort of critical attention that you would give trying to do a nationwide survey. Sure. 

Colleen: What are some ways that AI is already being used to shape decisions in areas like public policy, economics, or community development? 

Rob: Again, it's a tool, like Lane said. Mm-hmm. It's not like they're fabricating data using AI. What they're doing is they're taking advantage of the technology to look across vast data sets- Mm-hmm ... pull things together that human beings couldn't do in a longer period of time.

Colleen: We've heard the phrase, garbage in, garbage out. So when it comes to policy regulations, how critical is that from a computer science, from the, the data science point of view, to make sure that you're getting the right data, not garbage? 

Lane: Yeah, the right data going in can make all of the difference. But one of the things that we keep coming back to it in that field is we need to have humans in the loop, whatever process is there, because the AI can only see so much. It's people who really have the intuition, the knowledge to know if something's right or not. But certainly the data quality that goes into it is a huge component of it. It is a time saver, but how do you make sure that the human, as you said, is still in the loop? Yeah, I think it's a time saver if it actually leads to better results. If your measure is writing this policy document and sending it out, sure, it's a time saver, but is it-- does this policy actually move the needle on meaningful measures? Then who knows? And I think a, a lot of what people are trying to, to do now is to figure that out across everywhere where AI is used in work, is how do we in- incorporate this into our workflows to actually do things better?

Colleen: What about biases? 

Lane: Yeah. Biases are, uh, certainly present in AI and I'm... gotta give credit to the folks who recognized this early on and sounded the alarm bells and to, for us to look at this critically. I'm happy that the community has several scholars and venues where people discuss these issues and come up even with tools and frameworks for dealing with bias in a better way. 

Colleen: Okay. Before we get to the case that we're going to dive much deeper in, Rob, how do you build trust with communities when technology is part of the process, especially communities that maybe don't understand AI and deep learning?

Rob: You don't lead with the technology. The l- leading with making better decisions, making better informed decisions, making decisions that will empower them and help them realize their own agency is a way to address community concerns. We don't go in and say, "Oh, we're gonna use AI to solve your problems," because I think we'd probably be laughed out of the room. No, I think it's really, it's the way it should be, is that we're building trust between us, the research team, and them, and the technology is this sort of intermediary that we use.

Colleen: Coming up, we'll look at the poverty stoplight 

Commercial break: Hey, I'm John Kane, one of the hosts of the WPI Podcast. If you're enjoying this episode, make sure to check out our other conversations exploring the people, ideas, and innovations shaping WPI and the future of STEM. Find us at wpi.edu/listen 

Colleen: Welcome back. Now let's take a closer look at Poverty Stoplight, a groundbreaking approach that puts the definition of poverty back in the hands of those experiencing it. Using an app, families complete a pre-survey, a main survey, and build a life map. The survey covers fifty-five indicators across areas like income, housing, health, and education. For each one, families rank their situation as red, yellow, or green. Red meaning extreme poverty, yellow poverty, and green non-poverty. They can then prioritize challenges, set goals, and map a path forward, while also generating insights that help policymakers and organizations make more informed decisions and drive meaningful change. Rob, give us some background 

Rob: I came to know Martín Berti, the CEO of Fundación Paraguaya, and his team both there and in the United States. And he was telling me about all these data sets. He's got 500,000 records, and it was like 550,000, then 650,000, and it just kept going up. And I thought, what can we do with all of that information? What can we do in an age of AI to actually do better social science? And so Martín was kind enough to give me access to confidential information behind a firewall, and we started looking at it in different ways, and we looked at it against the DHS surveys. They're like a census survey that's done around the world that focuses on human health issues. So we started to think, what are we actually learning from this deeper dive into individuals that the survey doesn't necessarily capture? And we found some really interesting stuff. It's like for some, having six years of education, the survey suggests that you're not impoverished with education, but rather Poverty Stoplight, in contrast, says, "Yeah, six years of education, but I can't read or write." Bringing in the role of bias. There's-- They're typically Westerners working for development organizations, and they assess someone's house and they'll say, "This house is in disrepair. People need housing here." But if you talk to the people, it doesn't even show up on their agenda. They're perfectly happy with their housing. And we pulled this together to try to think about, can we glean a good policy question? Can you glean economic development opportunities from this information? By looking at the stated needs of people in the community, can you identify economic opportunities, entrepreneurial opportunities for them in local places? And we just started digging in it, and then I, I'd met Lane a couple years back when we had the director of digital development from USAID on campus, and he came to that meeting and learned about, one, his perspective on data, but also his way of representing data and- So I was looking for somebody to bring those kinds of chops to the project. 

Colleen: Lane, tell me about your journey. 

Lane: Yeah. I think growing up in North Carolina, where I'm from originally, rural South, we were no strangers- Mm-hmm ... to poverty, and that's always something that weighed on me. And so when the opportunity presented itself to work with some really meaningful data, uh, in a way that really aligned with what we do here at WPI, what we do in my research lab, I thought that was a great opportunity. So when I saw the Poverty Stoplight Data, uh, I think it was one of Rob's students just happened to show up at my office hours and showed me a spreadsheet, and was like, "Can you do anything with this?" And I was like, "Ooh, I think we can." Uh, and a few weeks later, we had a, a prototype that analyzed the data in a new way. And one of the things that stood out, and I learned later from Rob, is a lot of the measures that the government uses for poverty are very top-down. They're very coarse in that, you know, you have this much income, or you have this much education. Coming from the bottom up, letting people define it, I think is a really powerful aspect of that. So this idea of multidimensional poverty versus very few dimensional is really intriguing. And in data visualization, we have a whole host of techniques for multidimensional data. So just trying out a few of those on some of the Poverty Stoplight Data really rapidly led to some really interesting insights, where you could look at the multidimensionality of it on your screen rather than just peeling off one dimension at a time. And I think doing that and then tying it to geospatial data, and now thinking about even more data sources we might use, really helps tell a fuller picture of what people are experiencing and where some interventions could potentially come into play.

Colleen: Do you remember one of your first aha moments? 

Lane: Oh, yeah, absolutely. As soon as we ran, like, a high-dimensional clustering algorithm on this, I was excited. It just... It, it was-- It worked. Like, the clusters were meaningful. It, it's so many aha moments. I'm thinking of one where we just started typing in water to see what people's priorities were, and then seeing that pop up, and it turns out several of them were close to each other, which you could look geospatially to see if perhaps there are some water issues in that area. And if you were to fix that, that would be a change. Yeah. 

Colleen: Rob, you're nodding vigorously, agreeing with the water or just- The whole thing ... your aha moment? 

Rob: It's, it's the whole thing. Seeing the, the graphical representation of the data in high-dimensional space was... It was cool, and I can get into the nerdy stuff in a minute. But I think that we-- It was water. There was one around education when we were looking at the Rwanda data. There was one around electricity. And it occurred to us that if you had some knowledge of the place, you would be able to understand in much more detail and much more depth why this place or this cluster of people feels like they need more energy, energy to-- for refrigerators or for lighting and things like that. Or education, you could identify, oh yeah, that's a bad school. I know that they have a hard time getting the teacher to come. These aren't necessarily first-world problems, but they are really interesting in that it's not we go and we try to understand why kids aren't going to school. We understand that the parents know that there's a challenge here, and it doesn't have to do with flip-flops and guinea worm. It has to do with the fact that the teacher doesn't show up- Or the kid is malnourished in some ways, or there's some other, say, gender-based violence or something like that that intervene in a child's education. 

Colleen: We are gonna get to the nerdy stuff, I promise. But Lane, you said it hit home because in North Carolina there was poverty. There is poverty. That poverty may look different than Rwanda or Paraguay. So is that what you mean by geospatially, that either lack of resources or just how people perceive things? Either one of you can jump in on that. I just wanted to make sure I understood that and that our listeners understand it. 

Lane: Yeah, I'll do a quick one. I think it, it very much is how those dimensions light up is gonna depend on the space that you're in to some degree, because it's a different place and different environment and different values, different norms. So I think, you know, how people experience and think about poverty in a place like North Carolina is gonna be quite different than Rwanda, but there are also some commonalities too. Sure. So I think being able to have that global perspective is one of the things that's really valuable and unique. It remains to be seen how to best harness that, but, uh, I think that's something that you can definitely follow up on. And geospatially on the slightly nerdier side, there are a lot of GIS systems that we see used by, you know, the state, different governments to the extent that they have contextual information. Being able to combine that with something like the Poverty Stoplight might be one of those things that Rob was talking about earlier, where you combine disparate data sources and to help you, uh, construct an effective policy. Okay. Colleen: Do you wanna get nerdy, Rob, or do you wanna add to that? 

Rob: I, I'll just, I'll add a little bit to that. But it's-- Lane mentioned he's from North Carolina and w- but before getting to that, we were talking about electricity in Rwanda. Yeah. It's something that we picked up on, and we can zoom into the neighborhood. I work with a Rwandan in physics here, and so I talked to him about it, and he explained to me why this area was that way. So we're looking at the North Carolina data, and one of the things we saw people felt like they were-- they could use more of was community engagement. And so Lane's there and he's looking at it. He goes, "Oh yeah, that's rural North Carolina." So they, there might not be somebody living within 10 miles of them, and so they're not politically engaged. They're not connected to their community and things like that. So Poverty Stoplight really does a good job of helping folks together decide on what is something that's surmountable with a policy intervention and really just what isn't, you know, and, and where to prioritize and things like that, and help people establish their own priorities. 

Colleen: Why is this a shift from traditional models?

Rob: It's a shift because it turns the process on its head. So from, say, US census data, I could look at the same county in North Carolina that we're looking at, and I could say there's people there who are financially impoverished. The property values there are very low. The education level is very low, and therefore- That's the information I have to, to move on it. But that's me interpreting the data in a certain way. It's not me understanding what is it from the perspective of the people who are experiencing poverty. What do they wanna fix? Using a certain instrument to try to help people that they don't really understand, and turning it over and s- like letting the people who are experiencing it speak for themselves, and recognizing that they have agency, they have knowledge of their own situation in ways that those prior instruments, which were very top-down, didn't. Just bring me through the whole process from the subject's point of view. That's really a, a great question, and it tees up a bunch of different things. One is a relationship that Lane and I are working on or cultivating with the Department of Public Health in the city of Worcester- Mm-hmm ... to work with developing a, an instrument to serve homeless people in the region. But it also has to do with a book that I mentioned Martine Burt earlier, publishing a how-to manual for people going out and doing this in the United States. There is no how-to manual for doing this. So through the WPI Press, and hopefully with some support from the NextLetter Foundation, we're gonna be able to disseminate this, what I'm about to say, broadly across the US. And it starts with getting folks who are in the policy arena, getting them accustomed to, and to buy into the bottom-up approach to measuring poverty. And that's not as hard as you might think, because, because a lot of the folks have public health backgrounds, and so they really do understand that community perspective and wanna see that captured in any type of survey. And then you go out in the community and you start asking people questions. You have focus groups or s- or you meet with thought leaders in communities and things like that to try to understand what that community's array of indicators are, and then do it until you start hearing repetition and you say, "Okay, I think I got it now." Then you go back and you iterate with the community and with the organization that you're working with to come up with the indices for the stoplight, and then you go out and you pilot it with people. And if it is really meaningful to them that they relate to it, and you get feedback, and then you go out again, hopefully with a final instrument, and people can then use it through an app on their cell phones. They use it through the computer. You can fill it out with a pencil and paper however you want. And then the more these folks go out, say the homeless group goes out on every Wednesday morning, and they'll spend three or four hours out there talking to homeless folks about what their needs are, what their challenges are, things like that. This-- Through this guided survey or interview, we can start to really understand the nuance behind it, because you're not homeless because you don't wanna live in a house. You're not homeless just because you're an addict. Mm-hmm. You're not homeless just because you don't have money to pay your rent. There's all these other kinds of things that, as Lane mentioned before, multidimensional poverty that comes together to put people in these situations. 

Colleen: Lane, how large a data set do you need to start seeing these clusters or trends? 

Lane: That's a good question. I haven't answered that formally, but informally I would say just a couple hundred- Okay is enough to, because you have so many dimensions with one family or one person that you've got a lot of room to work with. Yeah, I don't think it's too much. I think if anything, there's always room for more, and we're here talking partly about AI. It's thinking about how could we use some of the emerging technologies to save time and effort- Mm-hmm to allow people to tell more of a richer story, like a multi, multi-dimensional poverty. Because even if you have a great set of indicators, there's still more going on. 

Colleen: Okay. So it can be tailored while seeing the trends, what the data says, visualizing it, and, and hopefully enacting change. I don't wanna put it too simply because it's a process. It is a layered approach, but am I on the right track? 

Lane: Yeah. Obviously it is a process, and that's one of the benefits of it, but it helps inform policymakers in new ways about populations that they're trying to serve. W- whether it could be actually all the way down from the caseworker, the person who goes out in the field and does- Yeah the interviews, to their manager, to the director of that group, to the, in this case, the Commissioner of Public Health. So whether you're trying to get more resources for something or you've identified something new that hasn't been considered by, say, the state legislature in Massachusetts, then you've got data that they can recognize and respect to help make a data-driven approach. 

Colleen: Can AI help reveal patterns or needs that might go otherwise unnoticed? 

Lane: I think if used in the right way and developed to re- recognize or to reveal something, it potentially could. I think the, the question I would put in the place of that is what tools do we need to harness the capabilities of AI to reveal patterns or needs that might otherwise go unnoticed? Because I think all of these things are possible for folks to do, but can you do it on a state or city budget within- Yeah ... a time period so that you can actually deliver and ask for more resources? One of the approaches that we're taking with Rob is to think about what does a, an entire toolkit look like end to end when we're talking about the surveys, when we're talking about people going in the field, when we're talking about aggregating that data into reports or visualizations or dashboards that would then go up and perhaps influence policy later? There are a lot of opportunities there to have some focused efforts to reduce the time and cost and expand the capability set so that we can... One term that I like to think about is can we listen at scale, and can we do that better? 

Colleen: I love that. In a previous life, I, I did polling and first of all, you have to get people to commit to answer the questions, but the data is only as good as the questions you ask. And a lot of time it's just about the quantity over the quality. And you said it's about being a good listener. So y- you wouldn't think that AI adds in that extra maybe, uh, not human, but extra layer. But in fact, here you did because you're able to take all of that data and maybe find the nuances or find something that will rise to the top. And when it rises to the top, you see it visually.

Lane: Yeah, I think there are principles that will emerge as we figure these things out. To your point, I think that- I- if we talk about using AI to summarize a lot of what people are saying or a lot of this data, you certainly want someone to look at the individual original data at some point. That's gonna be true whether you're doing policy analysis or intelligence analysis. You need to look at that source document that's trusted. But if you can use AI to, to help surface that and navigate it well, and to help build your intuition of what's in the data, of course there are risks that come with that, but I think it's a powerful and engaging way, and you can cover a lot of ground doing that and maybe even make it more enjoyable. To your point about taking the survey, it should be enjoyable also to analyze your survey, and that's what we have in mind. 

Colleen: Rob, I'm gonna throw this one your way. What have you seen happen when individuals and families take ownership of identifying their challenges and possible solutions? 

Rob: That's a good question, and we're trying to measure that scientifically, but haven't gotten there yet. In Brazil, we're piloting one of our tools, and I sat and watched a family do it for the first time, and they're speaking Portuguese, but in real time I could tell that they were having a very constructive conversation about their financial and about their multidimensional poverty. And then later when I heard the transcript, it was... It's, yeah, it's liberating. You can talk about it with others and know you're not alone or- Exactly ... work together. Yeah. That's one of the things they said is, "Oh my gosh, somebody else feels, you know, what I feel when we all go through our busy daily lives and don't really connect with folks. We might think we're in this alone, we are the only ones experiencing this type of division." Yeah. 

Colleen: When I was writing up these questions, I kept writing poverty spotlight, not poverty stoplight. But maybe I was onto something because you're looking at all of these different factors, putting the spotlight on what poverty is, what problems arise from it or cause it, et cetera. We're gonna switch now over to the actual name, the poverty stoplight. How do you take all of this information and start to inform and start to affect change? 

Rob: First of all, Colleen, you're not alone. Good. Lots of people call it the spotlight. Okay, good. There's a number of ways to try to do it. We've talked to the government of Rwanda and they put us in touch with the Ministry of Local and Municipal Government, and that's the way you do it there. In Worcester, you start somewhere in high enough up in the organization that they can make a decision and allocate resources, but not too far away from the people who are actually going out and doing the work because the success of this is- The range of things. It's from people giving honest and sincere answers to people being open and to listen and interpret these things in the way that they were meant, and really following through with that. And then working it all the way up and so that people who oversee the folks who go out in the field can start seeing connections across their different groups that the groups themselves might not see. So it works its way All the way up. But there's another part of this too that's important is, is what counts as data. And when I talked about the 1870s earlier, since at least then, data was considered to be something that could be quantified in a single number. And while this does the same thing, ultimately you're red, yellow, or green, and you're, or a three, two, or a one. But there's all this context that sits behind it that you can understand better what people are feeling and what their experience is. 

Colleen: So that's where, and Lane, this is for you, you are getting these stories. But when you are trying to get to policymakers to impact change, they need to see the data, the clusters. So are you seeing that change as more people are becoming part of this?

Lane: Yeah, that's one of the most challenging things is w- if poverty is multidimensional, how do you measure across all of those dimensions, and how do you do that in geographic areas to, to show that progress has been made? And I think some of the things that I've seen folks do now are perfectly reasonable. You're zoning in on one thing and showing how many go from red to yellow or from red to green. I think there's an opportunity though to again look at this from a multidimensional perspective and see, are people moving in the same way? Are they citing interventions that are perhaps different? How can we look at this from a, a full perspective? Develop the metrics, of course- Yes ... that the policymakers need. But kind of show statistically, mathematically that those are a faithful reflection of the data that they collected, show that they correlate with other measures that may be more traditionally valued by folks in policy. There are some, like, social determinants of health that we heard several times. So if you can develop those foundations, I think you're in a good spot. But I, I think it's even more than that. Sometimes they need stories. So- You need both. Yeah. Yeah. So you want some stories of what's behind all of these multidimensional indicators. How has a person moved from a red to a green on some meaningful dimension in their life? And reflect on that. 

Colleen: As AI tools get more and more advanced, do you see that impacting the level of information you can get to or surface? 

Lane: Yeah. What a great question. So Rob earlier was talking about prototyping a tool that we've been testing out. With the Poverty Stoplight, we started exploring this idea of allowing people to answer the questions, but also to speak to us as they're answering the questions and have a conversation And much like we're speaking into microphones now, computer technology is really strong for that, and we've developed some NSF-funded survey software, uh, where we can actually chop this up really nicely for analysis along with what they answered and their transcript and their audio. And we're currently exploring how we can have not just the red, yellow, green, but the story behind it at the same time. So I think that's in the very near future. 

Colleen: So the AI is listening for what? Certain phrases or..?

Lane: A really simple example would be say that they answer, "I'm red on neighborhood safety." But if you have 50 reds on neighborhood safety, they're all gonna be for different reasons. But if you can actually capture some of those stories behind that, uh, and use some natural language processing and AI LLM tools, it really reduces the time and cost of doing that analysis, which I think is the key thing here. That was always possible to read all the transcripts, but now you can say, "Oh, look at these first," or maybe we're clustering within them. So if you can pull that off and actually surface some of the underlying issues, then I think you're in business.

Rob: What Lane said is spot on. And just to try to draw out an example is I think of the poverty stoplight as tabula chart for a guitar. So you can see the notes, but what Lane's talking about is we can now see what lies between the notes and the resonance and the, the picking pattern or the strumming pattern or something like that. So it gives you so much more information instead of just these dots that look like they're independent notes. When we looked at the community survey for the city of Worcester, it's asking the same questions that we're asking. But instead of assuming somebody is loads positively on that or they don't load on that, we're asking people to say, "How do you feel about it? What are you feeling about it? Do you feel like you are deprived in some way? Are you doing okay? Are you thriving in this area?" And getting that to be a de rigueur for policymakers would be outstanding. That would be like true success. 

Colleen: All right, I have two last questions and one for each of you. Lane, if listeners remember one thing about AI in the social sciences, what should it be? 

Lane: You should definitely be using it and you should be using it responsibly, and you should be developing tools so that you can use it better. 

Colleen: All right. Rob, if there's one thing listeners should take away from the poverty stoplight, what would you want that to be? 

Rob: That AI stands for actual intelligence. That you're really drawing upon information that was produced for a certain purpose and is being evaluated, analyzed in a way that represents real people's feelings. 

Colleen: Excellent. I really just think we scratched the surface on what can be done and what needs to be done, and I would love to have both of you back here in the future to see how things are going. So thank you so much. Thank you, Colleen. Thank you. This has been another episode of the WPI podcast. I'd like to thank Varun Bhatt and Astrid Detwiler in the Global Lab for their help. You can get all episodes of the WPI podcast on Spotify, YouTube, wherever you download them, and of course, on WPI's website. Thanks again for joining us. I'm Colleen Wamback.