Research Approach

Our original research guides the process of innovation and identifies insights and methods that accelerate progress across the social impact ecosystem.

Machine Learning and AI for Social Impact
Machine Learning and AI for Social Impact
Susan Athey, founding director of GC Social Impact Lab, highlights the approach of the lab using digital technology and social science research.

Innovation Through Collaboration

Over the past decade, digitization has transformed both academia and industry. A new set of research tools and practices has emerged to enable data-driven, fast-paced, iterative innovations of digital products and services.

New practices in analytics, experimentation, and artificial intelligence — the foundation of today’s “tech firm toolkit” — require careful attention to the context in which products and services are delivered. Many of the interventions relate to social science concepts such as incentives, behavioral economics, and the psychology of well-being.

Our approach to a research collaboration starts by carefully studying the problem and its context. The lab incorporates expertise from the relevant problem domains (such as education or health care) as well as general social science insights. The “tech firm toolkit” is then applied to create a data-driven approach to innovation. Often, we start by analyzing historical data, and then proceed to develop approaches to measurement and experimentation that promote rapid testing and data-driven innovation.

Through this approach, new social science insights are developed about the design and efficacy of interventions, as well as about the underlying mechanisms. We also develop new statistical methods and approaches to experimentation that are tailored to specific problems and challenges. We document the impact of approaches such as personalization, highlighting fruitful paths for future innovation both within and across domains. The lab also develops new approaches to measurement for particular use cases, such as measures of student learning while using educational mobile applications. Measures of learning progress that work well for evaluating long-term use of an application are typically different from the measures that are available for guiding rapid experimentation, but often the two types of measures can be linked in useful ways.

Our research is disseminated through a wide range of academic outlets, and is also synthesized in educational materials designed for practitioners and students.

From the perspective of collaborating organizations, working with the lab offers access to scientific expertise and advice in the areas of artificial intelligence, machine learning, data and analytics, behavioral science, and incentive design, along with access to Stanford student talent. In most cases, collaborating organizations are working on problems at the frontier of both research and application, so the value of academic expertise is particularly high.

Drive Measurable Impact

We have several areas of focus in our research on methods. In one area, we develop new methods that are needed to effectively apply the “tech firm toolkit.” For example, we create methods that use machine learning to measure the impact of interventions and understand how this impact varies across different individuals. The value of personalization may be high when individuals have different backgrounds or learning styles. We further work on approaches to allow organizations to experiment much faster and more efficiently, for example, algorithms that make experiments adaptive. Adaptive experiments use data about what is working to adjust treatment assignments during the course of the experiment, thus focusing learning on treatments that are most likely to be effective. We design such experiments in a way that leads to better outcomes for the subjects in the course of the experiment, and further still allows for the data to be used to test hypotheses about what works and why at the end of the experiment.

Another area of focus is to develop generalizable approaches to inform, educate, nudge, and engage individuals through technology, using social science research as a guide.

Finally, we study approaches for improving the alignment incentives for innovation for firms with what is most beneficial for social welfare. Such “market shaping” approaches include governments or philanthropists paying for outcomes, advance market commitments, and income sharing agreements where workforce training programs are paid based on the improvement in worker income. Our work on measurement is integrally connected to the success of market shaping, since reliable measures of relevant outcomes are essential to ensure that firm payments are aligned with societal goals.

When social impact organizations adopt best practices in measurement and innovation, the entire social sector benefits from increased transparency. Philanthropists and governments are empowered to direct resources to solutions that have the greatest potential for impact, driving more resources into the sector as a whole, and providing further incentives to improve measurement and strive for effectiveness.