In response to the economic crisis caused by Covid-19, the Togolese government implemented a new national policy to provide the country’s poorest households with subsistence cash transfers to help them survive the next few months. PI Blumenstock has supported the Togolese government by developing machine-learning algorithms that help identify the poorest households with the greatest need for these cash transfers. While the technical system holds great promise, it also raises important societal and ethical concerns. For instance, how might the use of predictive algorithms in the provision of social welfare impact physical safety, economic justice, and privacy or leave out critical populations? Even if the data are carefully protected, whose privacy might be at risk and in what ways? Toward mitigating large-scale unintended harms, this research entails developing and testing a new tool in the Togolese context—Social Impact Scenarios—that (1) engages impacted communities to identify situated societal and ethical concerns and (2) employs narrative storytelling to make those concerns, tensions, and tradeoffs visible, and at times actionable, to technologists and policymakers throughout the technical design and data-driven policymaking process. While Social Impact Scenarios will be developed in the Togolese context, our intention is for Social Impact Scenarios to be used more broadly as a tool to support data-driven policies that account for situated societal and ethical concerns.
Joshua E. Blumenstock
Associate Professor, University of California, Berkeley
PhD Student, University of California, Berkeley