Key points are not available for this paper at this time.
Poverty in the US is not invisible. A large number of Americans are low-income and experience homelessness. This population relies on scarce-resourced public services for survival and thriving. High-stake public service resource allocations are increasingly fueled by AI/ML to provide efficient and scalable services. While AI/ML tools are deployed with positive expectations, there is growing evidence of AI/ML causing invisible harm, often dismissed as inevitable. AI/ML tools are based on deficits and vulnerability ranking; they ignore the strengths of vulnerable individuals. This thesis aims to counteract existing exclusions and reduce AI/ML access barriers for low-income individuals, through their inclusion in AI/ML innovations and education, using mixed-methods experimental studies. Specifically, (1) designing and evaluating value-sensitive service-assessment tools that go beyond individual risk-factors and focus on strengths, (2) making AI/ML knowledge digestible for low-income individuals. Thereby, bringing about social change, AI/ML access, and algorithmic reform.
Dilruba Showkat (Sat,) studied this question.