Allen School Colloquium: Rediet Abebe (Harvard Society of Fellows)

Anna Karlin welcomes Rediet Abebe, who is recognized for her impactful research at the intersection of algorithms, AI, and social good. Abebe expresses excitement about the rapid organization of the event and her eagerness to share her research.

Abebe emphasizes the significance of algorithmic techniques in improving societal welfare. However, she outlines several key challenges in this field, such as measurement difficulties, limited resources, and information access barriers, all of which complicate interventions for disadvantaged communities.

The talk focuses on two intervention types: the allocation of societal resources and the improvement of information access for disadvantaged groups. Abebe highlights the critical measurement challenges in these areas, especially in quantifying poverty and its complexities beyond simple financial metrics.

Using poverty as a case study, Abebe discusses the inadequacy of conventional income measurements, which often miss significant aspects of disadvantage, such as income shocks. She presents empirical evidence showing the real-world implications of income shocks on homelessness and other adverse outcomes.

Abebe and her collaborators propose a model to account for income shocks in resource allocation, aiming to minimize the risk of ruin for families experiencing these financial disruptions. The focus is on ensuring families do not fall below a certain threshold of resources.

Through the modeling process, Abebe illustrates the core objectives and constraints of allocating income subsidies. She discusses using a greedy algorithm to determine how resources can be optimally distributed among families based on their individual profiles and risk assessments.

The discussion extends to alternative objective functions, such as min-max configurations where the goal shifts from minimizing ruin generally to helping those most at risk, showcasing the nuanced trade-offs in resource distribution strategies.

The potential impact of different intervention types is explored, emphasizing the varying effectiveness of income subsidies versus wealth subsidies. Abebe draws attention to the importance of designing interventions according to specific community needs and characteristics.

Addressing information provisioning, Abebe shares research focused on understanding the information needs related to HIV and AIDS across Africa. Findings indicate significant gaps in access to accurate health information, prompting efforts to leverage search data for better health interventions.

Abebe's collaborative work highlights the analysis of search data and its implications for public health strategies. She points out discrepancies in the quality of online content related to health information and the necessity of enhancing access to credible resources.