Today, we sit down with Satiya Murugaboopathy– incoming research engineer in the AI & Global Development Lab, for a conversation about AI, the study of poverty at global scale, social-computer science synergies, long-term AI trends, and more.

Background and first project

Interviewer: Alright—let’s rapid fire. Satiya, can you tell me a little bit about your background? How did you get to the AI and Global Development Lab? What’s your story?

Satiya: Basically, I studied computer science in Germany—in Leipzig—and I just finished my master’s. Adel reached out to me asking if I was interested in a collaboration—also because I met him through a past application process at the United Nations. So he reached out and asked if we wanted to collaborate. I was very interested, and we found an interesting project that we were both hyped about. We spent the summer—also along with you—and I think we got some good results and had fun.

Interviewer: I love that. What makes a research project fun? Where does that joy come from? What draws you into a project?

Satiya: Good question. First and foremost, the topic should be interesting—I need to be attracted to what we want to research. And there should be some kind of “unlock”—something hidden that we can unlock. If it’s too small of a problem, I don’t feel the motivation to solve it because it feels too small. That doesn’t mean small problems aren’t worth solving—it’s just that I fancy the big ones more. A big challenge is always nice. And the biggest reward comes if we’re able to solve that challenge in some kind of way. Of course, mostly we won’t solve it in the way we initially thought—or we won’t come up with the optimal solution at the beginning. But that’s the journey of the research project, right?

Interviewer: I like that idea—contributing in a broader way too, to others, to the research community, and the aid community more generally. What do you feel like was the “unlock” of the first project you started working on with Adel and the rest of the group?

Satiya: We essentially mixed all of the strengths. I come from a background where I’ve been working on multimodal inputs a lot, so it was obvious to build on that experience. We tried that in the realm of poverty mapping—wealth prediction. I feel the unlock came from combining stuff that people hadn’t combined yet in that way, and trying to solve it better.

Interviewer: Very cool. And what was that thing that other people hadn’t tried before?

Satiya: Essentially, it was not using only satellite imagery as one modality, but also using textual information. The underlying assumption is that wealth or poverty information—information useful for poverty mapping and wealth prediction—isn’t only embedded in satellite imagery; it’s also in textual data. And the biggest source of textual data is the internet in our age, so we tried to tap into that. But we also tried to tap into LLMs, which have processed the internet (and other data sources) into their “neural memory,” as we called it. So we tapped into neural-memory reconstruction, mixed that with satellite imagery, and got quite good results.

Interviewer: Very cool—fascinating stuff.

Interdisciplinary collaboration

Interviewer: You said you’re coming at this more from the computer science side. What were some surprises working with the more social science crowd—Adel being trained as a sociologist while also having a quantitative background? What were some surprises in that collaboration, or ways the synergies unfolded?

Satiya: I guess it’s mainly that when people come from different disciplines and aren’t primed with the same terms or best practices, they tend to question more than people inside the field. People in a field tend to take the main highway because they’ve learned that’s the highway. But new people don’t know which is the highway and which is the small road—or which road is quickest. So we had some discussions that were really cool. That’s the main benefit. I felt that in later projects too—we discussed stuff that I probably wouldn’t have discussed if we stayed only in the computer science community. That unlocked some new thoughts.

Interviewer: Do any examples come to mind—new ideas or perspectives?

Satiya: Not really from the past project. From the recent project… for example, I had been talking with Adel about transformer encoder and decoder stuff. He was kind of confused about me talking about decoder-only architectures while I was using a vision transformer encoder to preprocess satellite imagery. Even though there wasn’t a big “unlock,” it made me rethink: right—we’re talking about decoder-only architecture, but we’re using an encoder in the pipeline too. It improved my own thinking—being aware of all the parts in the pipeline is essential for finding things that might be forgotten in the pipeline but could be optimized with new technologies.

Interviewer: That makes sense—being explicit about steps that might otherwise be automatic. Once we clarify steps, we can better optimize them.

Why global development and what technology can do

Interviewer: Within the global development realm, what are your interests? What draws you to the space? What are some ways new technologies can advance that space?

Satiya: I’ve always liked applied sciences. I like the theoretical side and think it’s important—that’s why I studied it in university—but I feel the most impact one can give is by applying those sciences at the end of the day. I’m aware that theoretical research is necessary for the broader picture, but for myself, I feel the impact more because it’s more direct; it makes me more satisfied and happy. And I get a lot of satisfaction applying that to meaningful topics—global development is a very meaningful topic. Also, it’s a field where there’s a lot of data in some parts that isn’t necessarily used in the most modern way. In industry, there’s a lot of data and a lot of money behind it, so people exploit new ways to benefit from that data. In global development, I feel there’s still more potential.

Interviewer: I like that. There’s so much money going into a relatively small number of private organizations focused on building AI systems. So is global development being left behind in this AI gold rush? I think what we’re trying to do is bring some of the insights from the AI transformation to the global development realm.

Risks, ethics, and limitations

Interviewer: We’ve talked about opportunities—like using agentic AI to learn about development in different parts of the world at a very granular level. What risks do you see in applying agentic AI in this space? What have you observed, or what are your thoughts on those risks or intrinsic limitations?

Satiya: That’s the biggest topic we should be talking about if we want to apply AI—or agents, or AI agents—to global development. Because there are huge biases that AI and agents have. If we’re in a production scenario evaluating sensor data, maybe there’s less prejudice embedded, so it’s not as big a problem. But if we’re talking about humans—data collected on humans, on different countries, ethnicities, cultures—there are huge risks. We have to address that and keep it in mind while building systems and frameworks. If we think of ourselves as enablers—doing groundwork for other researchers who might not have as much CS knowledge—then we have a huge responsibility. We need to make sure what we publish and what others use is clean, and also “harnessed” so it doesn’t escape what we want—so it stays within the ethical guardrails we build around it.

Interviewer: That makes sense.

Where the space is going

Interviewer: Where do you think, in 5–10 years, this space will be? What trends or developments do you see, and how do you project those forward into the future of AI and global development?

Satiya: I think it will be mainly driven by trends and progress of AI in general. Whatever CS or AI is building, we’ll look at it and ask: how can that help us? Will it help us? My vision is that we can build systems—or emphasize the need for systems—that are very bias-free, because our scenario is very sensitive and prone to bias. I hope we can build or adapt existing LLMs or AI agents so they are less racist, for example. That’s my vision: building such systems.

Technical challenges

Interviewer: Do you see any technical challenges that are specific to AI and global development? Any unique technical aspects of this space?

Satiya: I actually think that, like we talked about, the space isn’t filled with as many AI systems as other research areas—maybe because of missing money, or missing incentives for people to build systems for these projects. So currently, we have a lot of low-hanging fruit—where the hard task is doing the work, but not necessarily that we need innovation in every case.

Advice for students

Interviewer: What advice would you have for young folks—high school students, college students—who are interested in getting involved? What would be your advice for how they can get involved?

Satiya: No matter if you come from the social science side or the computer science side, you bring good experiences. In your field, build as much experience and gain as much knowledge as possible. Then come into the team, and we’ll find synergies. That’s it.

Learning in a fast-moving field

Interviewer: How do you like to learn? What have you found effective? And what are your thoughts on learning in a quickly evolving landscape—like AI—where the literature and tools move really fast?

Satiya: That’s a really good question, because it’s a challenge most of us face in AI. It’s hard to keep up. During my study years, I was always working in parallel—also because I needed to earn money—but it was helpful to build things, not only stay in the books. If you build things, you automatically search for solutions and learn from real problems. That has changed as well—nowadays you don’t just go on Stack Overflow—but the idea stands. Building and experimenting is key, from an early stage. Other than that: choose your focus, but don’t be too focused—if that makes sense. It’s hard to choose a focus early because everything is interesting. I experienced that too. In my early bachelor’s semesters I looked into medical computer science—computer-assisted surgery—stuff like that. Getting practical experience across fields helps you understand them better, then choose your focus. Computer science is still young compared to other engineering fields, and it’s too broad in a sense; it could be separated into more subfields. So we need to choose our subfields, stay up to date, try to become experts—but not “tunnel focus.” Still catch up with other fields too.

Interviewer: That makes sense.

An AI crisis?

Interviewer 2: I was curious—did you read the article that’s everywhere now… the “2028 global intelligence crisis,” the Citrini Research article? I was wondering whether you thought the hypotheticals in the article were realistic or not. If you haven’t read it, no worries.

Interviewer: Good question. Satiya—what do you think? Basically: how will agentic systems affect the economy and labor market, especially for young people?

Interviewer 2: The idea is that throughout human history, human intelligence has been a scarce input—while capital and resources are generally abundant. But agentic artificial intelligence could make human intelligence abundant too, and the article explores what might happen because of that.

Satiya: Do I think that’s true? I’m not sure. I’m an optimist, though. ß

Interviewer: There are two perspectives. One is: there have been many “intelligence explosions” in economic history. In the 1950s we created computers that could do radically more complex math than humans could do before. Tools like Wolfram Alpha, Mathematica, and other software have automated large amounts of computation since the 70s, 80s, 90s. Tools that expand what one person can do have accelerated productivity—and in some cases made knowledge workers more valuable. Excel, Wolfram Alpha, Python—these tools expanded intelligence and made workers more productive. It’s not surprising that as agentic coding comes online, we see individuals commanding extremely high compensation because the impact of one person in that space is higher. Will it require adjustment? Yes. Will it increase inequality? Probably. But will it crash the economy to zero? If history is any guide, probably not—though it could be different. What do you think, Satiya?

Satiya: I agree. I think it’s a tool. Many people trying to scare us that it’s “taking over humanity” are wrong. There’s a famous example—an excavator does the work of a thousand people with a shovel, but that doesn’t mean the excavator crashed the economy because building got efficient. One could say the challenge in the AI boom is that it’s fast. But every revolution has been faster and faster over time. The dot-com revolution was faster than many before it, and people adapted. So I feel nobody is “left behind” in that sense. The task changes. It enables us to do more and do better—and that will improve humanity. That’s the optimistic view. Pessimists might say the opposite.

Interviewer 2: Yeah, that makes a lot of sense.

Interviewer: It’s a good question, and we’ll grapple with it. Game theory suggests tools will expand—it’s going to happen. It’s like a waterfall. You’re at the top and going down. The best thing you can do is prepare yourself for the fall. There’s no point wishing we were before it started. We’re halfway through—you have to go through it to the end. But, I think we’re about at time. Thank you so much, Satiya, for joining us.

Satiya: Thank you!