Considerations for AI fairness for people with disabilities

Shari Trewin, Sara Basson, Michael Müller, Stacy Branham, Jutta Treviranus, D. Gruen, Daniel Hébert, Natalia Lyckowski, Erich Manser

AI Matters · 2019 · 128 citations · 60 references

Concepts

TL;DR

People with disabilities often face discrimination, and as AI increasingly influences decision‑making, it can both perpetuate existing biases and offer opportunities for fairer treatment. The paper outlines opportunities and risks of AI in employment, education, public safety, and healthcare for people with disabilities, proposes strategies to promote fairness throughout the AI lifecycle, and points to human‑centered design literature for guidance. It recommends reviewing AI systems for user impact, providing mechanisms to correct errors and raise fairness concerns, and involving people with disabilities in data collection and testing to build inclusive, robust models.

Abstract

In society today, people experiencing disability can face discrimination. As artificial intelligence solutions take on increasingly important roles in decision-making and interaction, they have the potential to impact fair treatment of people with disabilities in society both positively and negatively. We describe some of the opportunities and risks across four emerging AI application areas: employment, education, public safety, and healthcare, identified in a workshop with participants experiencing a range of disabilities. In many existing situations, non-AI solutions are already discriminatory, and introducing AI runs the risk of simply perpetuating and replicating these flaws. We next discuss strategies for supporting fairness in the context of disability throughout the AI development lifecycle. AI systems should be reviewed for potential impact on the user in their broader context of use. They should offer opportunities to redress errors, and for users and those impacted to raise fairness concerns. People with disabilities should be included when sourcing data to build models, and in testing, to create a more inclusive and robust system. Finally, we offer pointers into an established body of literature on human-centered design processes and philosophies that may assist AI and ML engineers in innovating algorithms that reduce harm and ultimately enhance the lives of people with disabilities.

References

60