Content area
Abstract
Algorithmically guided decisions are becoming increasingly prevalent and, if left unchecked, can amplify pre-existing societal biases. This thesis employs modern computational tools to examine the equity of decision-making in two complex systems: automated speech recognition and online advertising. Firstly, I demonstrate large racial disparities in the performance of popular commercial speech-to-text systems developed by Amazon, Apple, Google, IBM, and Microsoft, a pattern likely stemming from a lack of diversity in the data used to train the systems. These results point to hurdles faced by African Americans in using widespread tools driven by speech recognition technology. Secondly, I propose a methodological framework for online advertisers to determine a demographically equitable allocation of individuals being shown ads for SNAP (food stamp) benefits. In particular, I discuss how to formulate fair decisions considering budget-constrained trade-offs between English-speaking and Spanish-speaking SNAP applicants. Both application domains exemplify processes to ameliorate demographic-based disparate impact arising from decisions made by online platforms, with the ultimate goal of uplifting underserved communities.