Two developments have popped up within the last week that are causing a stir on the internet. While lots of people in the disability community are concerned about AI, relatively little has been said about these two developments in the accessibility blogs I follow.
Why is AI Important to Accessibility-- And Why is it Risky?
I'll start by describing what this post is not about. This post has nothing to do with overlays or automated testing. My perspective on overlays is well known, based on the work that we performed for IAAP. While I also believe that artificial intelligence will someday revolutionize the world of web and digital accessibility, that day is not close and isn't the topic of this post.
Instead, this blog post is about two new legal developments concerning artificial intelligence (or AI). Why is this relevant to people with disabilities? AI systems are increasingly used to screen candidates for all manner of things, from college admissions to customer service calls. Just about anything that requires an application is (and likely will be) done through AI. This has both efficiency benefits as well as social risks. Members of any disadvantaged group, such racial minorities or people with disabilities, can easily be subject to discrimination because AI will either screen them out or tend to screen them out.
Because AI is a "black box" and only "learns" by examining countless patterns of data, neither it nor its creators can articulate why an AI system makes specific recommendations or decisions. In many cases, discrimination can only be identified by looking at the candidates excluded or treated differently by the AI system. This risk created by AI is why civil rights organizations have been concerned about its misuse. For instance, the Department of Justice's Civil Rights Division has several activities around artificial intelligence.
Colorado Signs Nation's First Comprehensive AI Law
On Friday, May 17, Colorado enacted SB-205 Concerning Consumer Protections in Interactions with Artificial Intelligence Systems. Here is a link to the signed legislation and a link to the legislature's overview of the new law. I won't go into the details of the new law except to say that it creates guard rails around "high-risk" AI systems-- which here means any AI system that can be a substantial factor in making a decision involving education, employment, insurance, or just about any of the other important things that we apply for in modern society but are not guaranteed of receiving.
The law addresses the increased risk of "algorithmic discrimination" in these high-risk AI systems. It requires that anyone deploying one of these systems monitor their systems for possible algorithmic discrimination and take active steps to address it. The law also requires organizations to notify consumers about possible algorithmic discrimination and offer them a way to redress it when it occurs. The law requires deployers to describe how they expect to mitigate against algorithmic discrimination before it occurs. One element that I really liked about SB-205 is that consumers need to be told when they are being subjected to AI system. This is important because people need to know when their rights under SB-205 are triggered and when they can file complaints if they believe that an AI system is making the wrong decisions. The law is a great first step, but I would have liked to have seen a requirement for deployers to take active steps in testing to verify that algorithmic discrimination isn't occurring.
EU Signs First AI Risk-Based Legislation
Not to be outdone, on May 21, the European Council approved its new regulation harmonizing rules on artificial intelligence. Unlike Colorado SB 24-205, the EU Rule includes AI systems other than just high-risk AI systems and adopts a "risk-based" approach that scales requirements based on level of risk. Some AI risks that present an extremely high risk are simply banned whereas others have hardly any requirements at all.
At 419 pages, the EU's new AI rule is not light reading. Its press release provides a quick summary and a very high-level overview of the regulation. While I obviously have not read the regulation carefully, I did note that paragraph 48 mentioned that a potential adverse impact on people with disabilities was a factor in determining whether an AI system was high-risk. This is a helpful addition that the Colorado law could have used.
Next Steps
Preventing abuse by AI systems is far beyond the scope of this blog post. Many organizations (including disability-focused organizations) are working hard on this topic. It's great that new legislation and regulations are supporting this effort and building in the safeguards to avoid discrimination.
Disclaimer
Nothing in this post should be interpreted as legal advice or as forming an attorney-client relationship. It is offered for educational purposes only. You should always contact a qualified attorney in your area to discuss your legal rights and responsibilities.
Want to See More Content Like This?
Want the latest blog posts, videos, white papers, and announcements? Sign up for our mailing list and stay in the loop!
We're Here to Help When You're Ready
Take a deep breath. Then feel free to reach out to our team when you're ready to discuss your accessibility needs.
0 comments on “Two Important AI Developments Affecting Accessibility”