Designing an Online Infrastructure for Collecting AI Data From People With Disabilities

By Joon Sung Park, Danielle Bragg, Ece Kamar, Meredith Ringel Morris. 2021

56 citations recorded by the source corpus.

AI technology offers opportunities to expand virtual and physical access for people with disabilities. However, an important part of bringing these opportunities to fruition is ensuring that upcoming AI technology works well for people with a wide range of abilities. In this paper, we identify the lack of data from disabled populations as one of the challenges to training and benchmarking fair and inclusive AI systems. As a potential solution, we envision an online infrastructure that can enable large-scale, remote data contributions from disability communities. We investigate the motivations, concerns, and challenges that people with disabilities might experience when asked to collect and upload various forms of AI-relevant data through a semi-structured interview and an online survey that simulated a data contribution process by collecting example data files through an online portal. Based on our findings, we outline design guidelines for developers creating online infrastructures for gathering data from people with disabilities.

Subfields: Artificial Intelligence, Computer Science Applications, Safety Research, AI Ethics, Fairness & Society, Human-Computer Interaction

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Science Applications

In Safety Research

In AI Ethics, Fairness & Society

In Human-Computer Interaction

Citations in the corpus

Designing Disaggregated Evaluations of AI Systems: Choices, Considerations, and Tradeoffs (2021, background): “[63] found that many people with disabilities were unable to capture and upload video or speech samples that”
Disability-first Dataset Creation: Lessons from Constructing a Dataset for Teachable Object Recognition with Blind and Low Vision Data Collectors (2021, background): “Often data from disabled people differ in key ways from data from non-disabled people which leads to applications that do not perform as well for people with disabilities [20, 42].”