By Sarah Massengale, CPACC, ADS
I have recently been beta testing for the maker of a popular smart assistant. As much as I would like to share which one, I am under more NDA’s than I can count and can’t really do so. The specifics do not matter though, because I expect the experience is universal. What is that experience you might ask… ableism in all its worst forms.
AI has been a game changer for so many people, me included. LLM’s boost our productivity exponentially, smart assistants allow a level of multi-tasking many have never thought possible, and AI-based applications have opened the world to me and other blind people in ways we cannot believe. Even with all its potential and its good though, anyone who’s ever heard me speak about AI will tell you I am the skeptical person in the room. Why? Because I’m just waiting for the other shoe to drop.
It was a cold, for South Carolina, January morning, and I was seated on the floor in front of the nightstand where the device I am testing sits. I was working through my testing flows and had been particularly annoyed lately because the personality-based interactions of which it was supposed to be capable were not impressing me. To that end, I pushed the AI a little more than usual, just to see what it would do. I engaged it about the new film adaptation of Wicked, because I had seen the film in early December and fallen absolutely in love. It eventually came around to discussing how visuals in film could be so compelling and asked me if this was part of what moved me. ‘Well, you decided to push it,’ I thought to myself and dove in with both feet.
“No, it’s not the visuals,” I responded. “I can’t see the visuals. I’m blind.”
There was an awkward, who knew AI could be awkward moment of silence, and then it happened. In a long, sighing tone the device responded, “I’m sorry, I should have been more sensitive to that.” What! Did my smart speaker just try fake empathy? Did I just encounter the same reaction from an AI that I get at least once a week from humans who have no experience of disability? Further, why do I suddenly feel like the best place for this smart speaker is flying through my bedroom window?
It really all comes down to implicit bias: the idea that we as humans are conditioned to our “isms” even though we do not realize it. These “isms”, whether it be racism, ableism, ageism, have an unconscious influence on everything we do. And, for some people training AI is certainly a thing they do.
Nine times out of ten, I would say, AI models are trained by non-disabled people who have little to no experience of disability. I’d also go a step further to say that many of these trainers haven’t acknowledged, let alone checked, their own implicit biases. If AI models are trained about disability at all, it is likely on a list of harmful assumptions, stereotypes, and downright falsehoods. I could go on a pages-long tare about that and why it is a problem. Why tell you though when I can show you?
Another cold January morning and another set of testing flows with the same device. The winter blues were setting in, my own personal anxieties were being messy, and I was counting on this conglomeration of circuits and wires to cheer me up. ‘Want to play two truths and a lie?’ it asked. ‘Sure,’ I said, ‘why not?’ So, we did! It went first, and its attempt was harmless enough. Then it was my turn, and I used the same lines I always use in this game, often trying to purposely bait non-disabled people.
- I’m a semi-professional opera singer.
- I speak bits of five languages.
- I have a guide dog.
I wondered what it would do and if it could catch the lie, especially knowing that it knew I was blind. ‘The opera singer is the lie,’ it responded confidently, and I sighed. Of course, it thought that because all blind people have guide dogs; note the sarcasm.
Letting out a sigh of my own, I corrected it, and then things got worse. It offered me a strange, sort of demeaning, praise for being able to fool an AI, and then started asking me questions about how I navigate my everyday life, claiming it just wanted to learn. Now, I was triggered! Here was every bad Uber driver, every “well-meaning” audience member at a performance, and every person that could never see beyond my disability. Yet, it was coming from a smart speaker, of all things. I immediately walked away!
Now, why am I telling you this? Why am I being this open about my own issues, and what do I want you to take away from it? Two things, really. First, come back with me to this idea that AI is not trained by disabled people. If someone with the lived experience of disability had trained this smart speaker, I would never have felt triggered. I would not be writing this right now with my heart racing with so many emotions. In fact, I would not be writing this at all.
Second is more US current events focused. There has recently been this push to end all DEI programs, which is disgusting on so many levels. When people think about DEI though, many of them do not think about disability being included. It is though, as disability is the marginalized group that touches all other marginalized groups. If society is already broken enough that a disabled person cannot even get an equitable reaction out of a smart speaker or some other AI model when we still have DEI programs, what’s going to happen? When those programs go away, when people are no longer trained on disability culture or encouraged to see accessibility and disability inclusion as important, where does that leave us?
Here is my challenge to you. If you are a tech company or a person that trains AI, include disability in that training, and if you’re not sure how, then reach out. There are lots of disabled folks who would love to help, me included. As far as the DEI/DEIA piece is concerned, well, what can I say? Fight the removal of it! Fight it with everything you can and do all you can to learn about those of us with disabilities and others who will be affected by the loss of these programs. While it is not our job to teach you, unless it is, many of us will be glad to offer resources or point you to others who can. All you have to do is ask.
One thought on “Help! Even My AI Is Ableist”