Toward a better theory of learning

Before he began his PhD studies in the Stanford University department of economics, Anirudh Sankar was part of a team of researchers examining the impact of incentives on immunization in India.

The work, with Nobel Prize-winning economist Esther Duflo, was fascinating and Sankar’s first introduction to the field of development economics, but it also made Sankar think about other ways to change people’s beliefs and practices.

“The immunization project was based on behavioral nudges, and those have their place,” he explains. “But it’s not based on a conceptual understanding of immunology: ‘What exactly is this infection? What am I getting when I get an immunization?’”

Now, Sankar is working on a project that, by studying the adoption of a novel irrigation system for small-holder farms in eastern Uganda, aims to shed light on how people learn. Last summer, as a recipient of Graduate Student Research Funding from the Stanford King Center on Global Development, he traveled to Uganda to explore the idea further through qualitative interviews with farmers and the people who help them install and operate the irrigation systems. The project is in its early stages, but ultimately Sankar hopes it will contribute to the literature around social learning.

“Residents of low-income countries are often bombarded with technologies created from the outside, and when people choose not to use the imported technology, there are all these questions about ‘why aren’t they adopting the technology?’” Sankar explains. “I wanted to think about technology adoption and what it’s like to confront technology as a black box,” meaning someone can see the results of an intervention but does not have any understanding of how the intervention actually works.

Sankar’s interest in development economics evolved over time. He was a math major as an undergraduate student at the University of Chicago and went on to study math and cryptography at the University of Waterloo, where he earned a master’s degree. After graduating, he worked for two years as a data scientist working on advertising optimization.

“I was scientifically fulfilled, but I wondered, is there a way I can apply my skills to something that could help people?” he remembers.

Sankar began searching for roles in development and found the Abdul Latif Jameel Poverty Action Lab (J-PAL), which Duflo co-founded. He spent two years there, including on an effort to apply machine learning techniques to the immunization research, before deciding he wanted to pursue a path in economics. A co-researcher at the University of Chicago, Vesall Nourani, helped Sankar clarify his own interest in understanding how people learn from each other and apply that new knowledge to their own lives.

In math, Sankar says, “we learn about things deeply and conceptually.” In development economics, “why is it that we underemphasize developing people’s ability to generate knowledge themselves and be curious about things?”

Through Nourani, Sankar learned about a social enterprise that uses motorcycle taxi, or boda boda, drivers to deliver, install, and run irrigation systems for small-holder farmers in Uganda. The company, Agriworks, hires drivers to travel between farms setting up the irrigation pump, which is powered by their engines. With the support of the King Center, Sankar interviewed about 30 farmers and boda drivers who service the farms. He found that a vast exchange of knowledge takes place during the irrigation visits: The driver tells the farmers about what he’s seen on other farms; the farmers tell the driver about how their own farms are doing. The drivers are paid to power the irrigation pumps, not to provide information. Nevertheless, they do provide information. Lots of it.

“What is their incentive?” Sankar asks. “It’s curiosity.”

Over the hours-long visits, “an informal chat ensues,” he adds. “And through this informal chat, a process of social learning also happens. It’s like having a private tutor without calling it that.”

Sankar and his team of co-researchers at the University of Chicago are still designing their experiment. But they hope to be able to test not just for the effects of the boda drivers’ influence on farmers—whether their crops perform better, whether they continue to irrigate, and whether they change other practices as a result of irrigating—but also how farmers’ conceptualization of irrigation and agricultural production changes.

In particular, the team is interested in diverse social learning, or the kind of learning that takes place when we learn from people who operate in different contexts than we do. Their hope is that their experiments will inform the conditions under which diverse learning can be effective, which can then be applied to interventions that help people adopt technologies that will improve their lives.

“Social learning is well documented, but the effects can be muted,” Sankar says. “One reason is that people are worried about context differences: ‘It’s great that it works for you, but you’re doing many things differently than me.’”

Learning diversely is rare, Sankar continues. “People try to learn from people who are as similar to them as possible. But people doing something similar may not have anything interesting to tell you. So the challenge is to learn diversely but with enough explanation so that one can extrapolate to one’s context”. He noticed in his interviews that boda riders and farmers frequently engaged in such explanations.

Sankar and Ben Davies, a fellow Stanford PhD student in economics, are developing a theoretical model of mechanistic understanding, or the ability to generalize across contexts. In other words, they are interested not just in what someone learns, but in how they learn it. What they discover could have far-reaching implications.

“It’s really important to understand how to develop agents’ generation of knowledge and to cultivate and support that,” Sankar says.

Without the ability to adapt new ideas and technologies to their own lives, every potential beneficiary of a new idea or technology would need to receive a custom solution for their own circumstances.

“That’s going to be extremely cost prohibitive and doesn’t scale well,” Sankar says. “If you end up with those solutions, you don’t have a solution.”

The alternative, which Sankar hopes to help advance, is to promote people’s understanding of an idea or technology’s mechanistic principles—or inner workings—so they will take the steps necessary to tailor it to their own circumstances.

Sankar says the King Center’s support of his project has been crucial.

“I can’t imagine being able to work on this without having done the exploratory interviews,” he says.

He adds that he’s also benefitted from feedback he’s received from the King Center’s community of scholars, including Faculty Director Pascaline Dupas and faculty affiliate Marcel Fafchamps.

Stanford Economics Professor Matthew O. Jackson, who serves as one of Sankar’s advisors, says Sankar’s work combines “theory and empirics in impressive ways” and provides “new understandings of exactly what contributes to difficulties in learning from others’ experiences.”

“By providing a deeper understanding of how social learning works and why it can sometimes fail, it will have widespread applications for economic development and more broadly,” Jackson says.