Welcome to the seventy-fifth episode of Access On, the National Federation of the Blind's Technology podcast.
Episode
Listen to the seventy-fifth episode of the Access On podcast (Browser).
Or listen on your preferred podcast platform.
Timestamps
Below is what's on the show this week, and when you can hear it.
- New research on AI and blind people 0:00
- Exporting Ootunes Recordings 44:42
- BBC domestic streams outside the UK 47:51
- Smart home automation 49:34
- Substack accessibility 51:19 Tech tip
- Training iOS dictation 54:13
- Closing and contact info 57:09
Transcript
Speaker 1:
Live like you want.
Speaker 2:
Access On.
Jonathan Mosen:
Welcome to Access On, the technology podcast of the National Federation of the Blind. On the show this week, new research highlights the benefits and the risks of AI for blind people. A listener seeks help exporting recordings from the ooTunes app for iOS. How accessible is the popular Substack platform both for consuming and creating content? And this week's tech tip, a clever way to make Siri dictation pronounced difficult words properly.
It's Jonathan Mosen at the Jernigan Institute in Baltimore, Maryland, welcoming you to episode 75 of the podcast. 75 episodes, golly. Where did the time go? We're going to talk about AI today, yet again, because artificial intelligence is now woven into the lives of many of us. It writes, summarizes, captions, transcribes, describes images, answers questions, and increasingly mediates whether people get support, get hired, get approved, or get left behind. And for blind people, that can create some tension.
AI can expand access and independence in ways that were hard to imagine even a few years ago. But when those same systems are inaccessible, inaccurate, privacy invasive, or over-trusted in high stake settings, they can create a layer of exclusion just as quickly. For these reasons, the National Federation of the Blind's Center of Excellence and Non-Visual Accessibility spends considerable time working with AI providers, not just on the accessibility of the tools they offer, but also on the expectations they set of who blind people are and how we contribute to society.
And we are making progress. For example, it's been some time now since I've had an AI express sorrow when I tell it that I'm blind, but the work is far from complete, and given how rapidly AI is evolving it'll never be complete. A new study from the American Foundation for the Blind, a professional service agency in the blindness field, draws on the experiences of 1,735 adults, including a substantial blind subgroup, to map where AI is helping, where it's harming, and where users are demanding a different future.
To talk about all of this, I'm joined by Dr. Arielle Silverman. She's director of research at AFB, and we're going to discuss the report and its findings. Ariel, it's great to have you on Access On. Thank you so much.
Dr. Arielle Silverman:
Yes, thank you so much for having me.
Jonathan Mosen:
When you step back from the report, what's the one headline that you most want listeners to understand about blind and disabled people's relationship with AI right now?
Dr. Arielle Silverman:
I think the main finding that we had is that it has a lot of potential to level the playing field for people who are blind and low vision, it has a lot of potential to make things fully accessible, but it's not quite there yet. There's still a lot of hallucinations, and mistakes and shortcomings that are occurring, there's still a lot of questions about privacy of data, and we also know that AI that is being used not by blind people but toward blind people, for example, in hiring contexts and in healthcare contexts AI can still cause harm disproportionately to blind and other disabled people. So, I would say that AI has a lot of potential and a lot of people are already capitalizing on that potential, but it's certainly not yet to the point that we need it to be at.
Jonathan Mosen:
I'll draw down into some of these topics you mentioned shortly. I want to comment first though on what I perceive to be the really polarized nature of discussion about AI. On the one hand, you have a lot of people saying AI is going to save us all. And on the other hand, people are saying we're doomed. The human race is going to be made extinct by AI. There are few people who you can find somewhere in the middle with a bit of a nuanced perspective on this.
Dr. Arielle Silverman:
Yeah. I mean, we need to have nuance, because AI is definitely not all good, definitely not all bad. And we also have to remember that it was created by humans, and humans ultimately have power to change it, whether it's changing the training data that the AI is using, or changing the context where it's being deployed, or adding in human oversights, AI isn't completely autonomous. Humans ultimately still have control over it.
Jonathan Mosen:
For the moment, right? I mean, AI is
Dr. Arielle Silverman:
Right.
Jonathan Mosen:
... starting to become self-modifying. Walk us through the design of the study. Who was surveyed, over what period, and what kinds of questions did you ask?
Dr. Arielle Silverman:
This survey was conducted between July and October of 2025, and we intentionally did this as a cross-disability study because we recognized that a lot of the uses of AI as well as a lot of the issues with AI affect not just blind people but people with other types of disabilities as well, and a lot of blind people also have additional disabilities. So, we made the study accessible and available in American Sign Language and recruited participants with all types of disabilities.
We also wanted to have a control group or a comparison group of people with no disability so that we could see whether the patterns that came out for blind people, for example, are those just limited to blind people or do those also generalize to users in general regardless of disability? And this was an online survey, although people could also choose to take it by phone or video call with a researcher or in American Sign Language.
And we ended up, after October 15th when we closed the survey, we ended up with 1,735 total participants, and that included 368 people who identified as being blind or having low vision, and our comparison group was 665 people with no disability. In terms of the types of questions, we started off with, after defining what AI means, we started off with a page where people could choose what types of AI they had used or what kinds of situations they had been in where AI might have been involved.
So we asked, "Have you used voice activated AI like Siri or Alexa," which people might not think of as AI, but it is a machine that thinks and behaves like a human. So, we asked people if they'd use that. We asked people if they used AI for writing and note-taking. We asked people if they used AI in the workplace or in school if they were in school, and we asked people if they used AI for specific disability related purposes like visual descriptions, captions.
And then we asked people if they had been in situations where AI might have affected them. So for example, had they been denied a healthcare procedure or medication in the last two years? Because AI is making a lot of those decisions. Had they applied for jobs, and if so, had they faced an automated job assessment or interview? So, those were a couple of settings where people might have been impacted by AI. And then we took, based on how they answered those questions on the first page, that determined what additional questions people saw.
So, people only saw questions relevant to what they said they had experienced. We asked questions about how often they used the particular type of AI that they said they used, what they used it for, and whether they had any issues like the AI making mistakes. If they used it in the workplace or in school, we asked if it was helpful or harmful. And then they had a chance to share in their own words how they used the particular type of AI that they were talking about.
The last thing I forgot to mention, which is actually really important, is we asked people if they had ever ridden in an autonomous vehicle, because that is also a form of AI, or at least automation that touches the blind community.
And so, we ended up with about 150 people who had been in an autonomous vehicle, and then we asked them more questions about what that experience was like. We also asked everybody what they thought about the importance of developing autonomous vehicles and the importance of developing public transit, and whether or how they though AI would improve transportation.
Jonathan Mosen:
And just before I move away from the methodology, one piece of feedback I do remember seeing on social media when the survey came out was from people outside the U.S. who said, "I have a real interest in this, but I'm not allowed to participate." Is it important to the integrity of the survey to have limited the sample to the U.S.?
Dr. Arielle Silverman:
So, I was advised by our public policy team to limit it to the U.S. because the regulatory environment around AI is so different in other parts of the world, or potentially will be so different in other parts of the world. And because we're making recommendations to U.S. policymakers about if or how to regulate AI and how to monitor AI, we thought it was important that the data all came from a single regulatory environment. There may be times in the future though when we will ask those questions of people outside the U.S.
Jonathan Mosen:
One of the strongest findings is that AI is already routine for many participants, especially voice activated AI. Do you think blind people are ahead of the curve here, not necessarily in hype, but actually in practical use?
Dr. Arielle Silverman:
I think so. So, we found that almost everybody used voice activated AI and it was kind of interesting, there was a pattern it seemed for a lot of older adults, whether or not they had disabilities, where they would say they used voice activated AI but they didn't really use any other type of AI. So, nearly everybody used voice activated AI, and blind people were not more likely to use it than sighted to people because everybody used it pretty often, but blind people did report a wider variety of uses.
So for example, blind people said that they use Alexa or Siri to help with playing games, or playing audiobooks or music, or checking on things like the news or the weather report, or surfing the web or dictating messages. Sighted people use it sometimes for surfing the web and dictating messages, but they were a lot less likely to use it for some of those other things like checking news and weather, or playing games, or playing audiobooks.
And so, it does seem like blind people are using it for like a wider array of tasks and more creatively in order to shortcut certain kinds of tasks, and I suspect a lot of that is merely because the smartphones aren't always the easiest to operate using the onscreen keyboard for a blind person. So, using dictation and saying," Hey Siri, open the weather app," or, "Open this game," and things like that is more efficient than using an onscreen keyboard. And then of course in the case of Alexa, it's just a standalone device that you talk to.
So, blind people are finding ways to use that to maybe circumvent accessibility issues with websites, like if the weather website, I know there's a weather website I use that has a lot of ads and the accessibility is, it's all right but it's not great. So, being able to just ask Siri what's the temperature right now is a lot more efficient than doing that on the web. And sighted people maybe just haven't had the need for that or haven't thought of voice activated AI as being helpful in the same way.
Jonathan Mosen:
It's interesting to think about whether this is because it points to some inaccessibility of other options or whether it's just that we are more used to engaging verbally with devices and hearing back from them.
Dr. Arielle Silverman:
Yeah, that's probably also true that we're used to engaging with devices verbally. And I know there was a complaint that came up particularly often with blind users where they would complain about asking a question and not getting a verbal response. So, like asking Siri a question and Siri would come back and say, "I found this on the web for ..." And then you had to go on the web on your phone and read the content instead of having Siri verbalize the content. So people wanted, especially blind people, wanted to hear back a response instead of just getting a text response.
Jonathan Mosen:
But it's interesting, because in talking to people from big tech, which has invested a lot, particularly Amazon in these voice assistants, they really want people to engage with the technology in the way that it seems a lot of blind people are, and it's quite hard to get many sighted people beyond asking for the temperature or the weather today, or setting a timer and onto more complex functions. It just seems to be elusive in most cases.
Dr. Arielle Silverman:
Yeah. And sighted people I'm sure are so used to the visual interface that it probably feels foreign to just talk to a device and not have that visual.
Jonathan Mosen:
Does that mean that we make good testers for this kind of thing, even if it's a mainstream product? That perhaps blind people are going to kick the tires a little harder than the average user might.
Dr. Arielle Silverman:
Yeah, I believe so. I think blind people might find issues that need to be corrected more than sighted users because they'll use them more robustly.
Jonathan Mosen:
I don't know whether it came up in your study, but one thing I notice is that particularly with the quality of built-in dictation solutions, some of the AI tools are now a lot better than those built-in dictation solutions, but you do see people using dictation a lot in our community and apparently not thinking that it is important to check to make sure that the dictation actually heard you properly, sending an email, perhaps a quick text to a loved one is a different standard from posting an email that's going to a list with thousands of people on it. When we publish something on an email list or even on social media that could be read by many people, that's our personal brand there, and people seem to use dictation in some cases without regard for the quality of the output.
Dr. Arielle Silverman:
Yeah, and it can be pretty dangerous. We did have a question about whether voice activated AI made mistakes that caused harm, and I don't remember offhand what percentage of people said it did. It did not vary by disability status, but a lot of the comments illustrated and I know I've experienced too where it can add like curse words into your messages without meaning to.
And if you don't check and then you just send it to your mother or your colleague at work, it could be really bad. Or if you put it on an email list, I mean, I've seen posts like sexual words and then that didn't belong there, and it was clear that the person had dictated and just didn't check, but the perception of course could be very different by the person on the other side of the communication.
So, a lot of people talked about like mishearings that could be really change the context of the discussion a lot if they're not corrected, or just frustration of having to repeat themselves a bunch of times, especially if they had an accent or if they had a speech disability like a stutter, that could be frustrating. I know also for me, sometimes it's just difficult to make a correction in a long message that I've just dictated, but I know it's important to how I represent myself, whether it's in professional context or a personal communication with somebody.
Jonathan Mosen:
Right, it really is time-consuming sometimes to, say in the case of an iPhone, find the right rotor option and navigate say by word, and then delete those things and reinsert it. I mean, it is a laborious process to make that correction. Where do you think AI is genuinely expanding independence for blind users today as opposed to merely being a convenience?
Dr. Arielle Silverman:
Well, the visual descriptions are an obvious example, and about 80% our study participants who were blind used visual descriptions, so they're clearly becoming more and more common. And we didn't differentiate, but a lot of people were using, of course, Aira Access AI or Be My AI for visual descriptions, but there were definitely people who were using mainstream products like Copilot or ChatGPT to describe images.
And I know when I've opened PDFs in Google, like in Chrome, I will often get a Gemini extraction of the PDF, which is super helpful when the PDF isn't accessible. So, that really opens a lot of doors to being able to read text that's not in an accessible format if AI has that capability and then describing pictures that maybe pictures that other people share or one's own pictures so that you can make accessible posts, or share pictures and know what they are is also really helpful.
And in theory, AI can now be used to read a lot of print that's in the physical world. And I say in theory because in some cases participants described mistakes that could actually be really serious. I mean, there's a funny one in the report about somebody read a label for a topical medication and it said, "Chew two tablets four times a day," and they were like, "Well, that's obviously wrong because this is a tube of topical medication," but it could be really disastrous if it says the wrong directions or the wrong name of a medication, because maybe it has trouble seeing the image and it adds in its own interpretation.
I think there was somebody else said they used it to look at their shirt and it said they were wearing like a collared shirt and it turned out to be just like a really casual T-shirt with a slogan on it, and they didn't know that and they went out to a big professional event wearing the shirt.
So, the image descriptions when it's correct can really expand independence because it can allow people to read print just on the fly or even read signs, although people weren't using it to read signs as much as they were using it for other purposes like reading text on a print page or in a document. When it does things right, it could actually expand a lot of independence.
Jonathan Mosen:
And it is important to stress that if people aren't reading medication labels, you're probably violating the terms of service of any reputable AI service where they expressly tell you not to do it for that. One thing that I have experienced, and this comes up a lot, is that it almost gets it right, but it's not right.
An example of this is recently I was getting to the bottom of a new Xfinity remote that we have in our house, and I use one of those AI services to take a picture of the remote and I had it tell me what the buttons do. It knew that it was an Xfinity remote, maybe it saw a logo or something like that, but it gave me a description of a remote that was not mine. It was another Xfinity remote.
Dr. Arielle Silverman:
Interesting.
Jonathan Mosen:
You do cover in this report this question of verification. As a blind person who can't check this out for myself, how do I know when it's right and when it's not right?
Dr. Arielle Silverman:
Well, and then the question becomes if you have to get a second source to verify it, is it even worth using it in the first place? If you have to get a human that you live with or a human on Aira to verify a description, is it better to just use the human? It depends on the situation.
Jonathan Mosen:
Right. And yet if you're doing something like you're in a hotel room and you think, "I wonder what's outside," and you take a picture from your window, some of those descriptions can be really quite helpful, give you a lot of nuance about what's going on around you. Is there a risk that AI can act as a crutch or a poor substitute for good quality blindness skills?
People can sometimes derive information from more traditional blindness techniques, particularly when it comes to navigation. And sometimes, as you say, if you're having to verify this information, the information might not be accurate. It just takes longer than good blindness skills would have. And what concerns me to some degree is that there might be a disincentive for people to acquire those good blindness skills if they think that AI can do these things.
Dr. Arielle Silverman:
Yeah, I would worry about that to an extent. I'd also just imagine if you're navigating, you need to be able to hear your surroundings. And so, if you're listening to an AI app chattering in your ear, it can be distracting potentially from listening to traffic or whatever landmarks you need to be listening for. I know there's an app, I don't remember the name of it, that helps people cross streets by telling them when the walk signal is on.
Jonathan Mosen:
Oh, the Oco app.
Dr. Arielle Silverman:
Yeah. So I don't know if that's AI based, but I imagine if it gets it wrong obviously that could be really dangerous if the person fully trusts it and doesn't also use their blindness skills to listen to what's going on with the intersection. That seems like a way that it could be problematic.
Jonathan Mosen:
So, it's about finding the place for this technology.
Dr. Arielle Silverman:
Right. So, using it as a supplemental technology where you have the cane or the dog is the primary aid, and then you might be using it for a supplementary purpose like to read a sign or to double check that you are where you think you are.
Jonathan Mosen:
You talked about your experience with Gemini and PDF files, and the report does suggest that disabled participants are using AI to make work and school materials more accessible. Is AI becoming then a bridge layer between inaccessible content and the tools that blind people already rely on?
Dr. Arielle Silverman:
Yeah, I think it definitely can be if people are using screen readers or screen magnification and then they're running across inaccessible digital content, AI has the potential to make it accessible. Of course, the fact that people are relying on AI for that means that there are deeper issues that we've been advocating for for a long time to make sure that content is accessible right out of the box so that someone doesn't have to use a tool and make it accessible.
We have had some thoughts about, especially the uses of AI in education that people talked about, that sometimes people might be using AI in lieu of accommodations from their college or university, either because they can't get those accommodations or because maybe they don't want to ask for accommodations, and it's easier to just fire up an AI tool and get the accessibility that way or get feedback that way.
People feel like disclosing disability to an AI is maybe safer than discussing it with their professor or the university disability office. I don't know if anybody said that in the study, but that's one thing that seems like it could be plausible. But we certainly know that a lot of students are not getting accessible materials, and AI is starting to bridge that gap.
Jonathan Mosen:
Is there a danger then that academic institutions might say, "We don't need to do our bit anymore because now you've got the AI"? Or does it not matter? I mean, is it the end that counts and not the means?
Dr. Arielle Silverman:
I think there could be a danger if academic institutions rely on AI, which again, isn't always accurate to create accessible versions of materials. I also worry about the opposite that because there's so much concern now about AI being used for cheating, if academic institutions ban AI it could put blind and other disabled students at a real disadvantage because then they can't use it for assistive technology.
Jonathan Mosen:
I remember a long time ago now when I was working with Freedom Scientific and JAWS introduced an ability to perform optical character recognition on inaccessible PDFs, and people were concerned about that too saying, "Well, that risks diluting the argument about accessible content if you can just put it through JAWS and come up with a version that a blind person can read." Of course, optical character recognition has flaws and it's not as reliable as a born accessible document.
So, I think that took care of it, and probably the same is true in the case of AI. It can actually be very difficult to get AI to render you a faithful copy of the document rather than just summarizing it, which is what it really likes to do. How should we think about teachable or customizable AI in this space? What would an AI tool look like if it really learned me rather than a generic user or use case?
Dr. Arielle Silverman:
I mean, in the best case scenario, it would learn your specific accessibility needs and preferences. So, if it knows you're a screen reader user so every time something comes up that's an image it needs to be described. I mean, that would be great if I'm on a website and all of a sudden the AI jumps in and provides an image description for every image that comes up as long as I can customize maybe the types of ... I don't want it to describe decorative images, that could be really useful.
Jonathan Mosen:
I was very interested in the finding that participants often preferred humans when sensitive data were involved, but blind users often preferred AI over a human reader when information stayed on the device so it wasn't going to the cloud to I guess contribute to somebody else's AI model or whatever. How should developers interpret that tension, do you think?
Dr. Arielle Silverman:
Well, it seems like there's a real need and real market for on device AI that can read text, either print or digital text, because if it contains sensitive information blind users are going to be more likely to use that than to use humans. Humans have their own risks of disclosing private information. So, allowing information to stay on device could really be a win for blind users or others with print disabilities who need to have that information converted.
Jonathan Mosen:
Yes, it is interesting to look at one particular product in the blindness space called Perspective Intelligence, which is seeking to do just this, where they allow you to install a range of large language models onto your device and everything just stays on the device. All the processing is done there. So, that's going to be interesting to watch. I want to talk about employment too, because the automated hiring findings are some of the most troubling in the report. What do you most want employers and HR vendors to hear? How are these systems screening out disability before anyone has a fair chance to show what we can do?
Dr. Arielle Silverman:
So, unfortunately we can't measure the screening out at the very early stages of the hiring process, because people don't know that they've been screened out so they can't report it to us. I mean, we do know there have been simulations, large language models where they've been given resumes that disclose disability or don't disclose disability, and people are screened out more often when they disclose disability. So, we know that at least it's a theoretical possibility that people are being screened out.
One of the strongest consensus points from our previous study that we did in 2024 is that there needs to be a human in the loop if AI is being used for candidate screening, and also that the use of AI needs to be disclosed to everybody who's applying for that job. They need to be made aware that AI is involved in the screening process and have a way to appeal or at least talk to a human if they believe that they've been screened out.
The other thing that came up in this report was with automated assessments not always being accessible, and we don't necessarily know with the stories that people told us if they requested accommodations during the automated assessment process, but it's likely that people might not have known how to request accommodations because they're not actually at the job site, they're doing this remotely, and they might not even realize that they need an accommodation until they start the assessment and it makes them disable their screen reader, it has pictures in it, and then all of a sudden they don't know what to do.
So, one of the biggest messages we want to give employers is if they use AI-based assessments and they post a job, it'd be really explicit About how people can request accommodations, not just for maybe a human interview but for any automated assessment that might come up before they're interviewed, because that's not always disclosed.
Sometimes employers will say if you need accommodations maybe with the application, they give you somebody to contact, but if they explicitly say that accommodations can be made for automated assessments and how the person can request, then people can avoid the kinds of scenarios that came up in the report.
Jonathan Mosen:
It's moving so quickly. I'm seeing on social media references to people being interviewed by AI. Those algorithms during the interviewing are scoring you based on things like eye contact, and facial expressions, and things of that kind.
Dr. Arielle Silverman:
Yeah, and it seems like in those cases, employers either should not be using that type of AI model, or they should allow accommodations for someone who has a disability where they can't perform eye contact and facial expressions in typical ways, which could be a lot of people, not just blind people. They should be offered a different way of being interviewed.
Jonathan Mosen:
Yeah. It's ironic, isn't it? That the report makes it so clear that AI is helping disabled people perform better at work, but at the same time it might be making it even worse than it was before to actually get the job in the first place.
Dr. Arielle Silverman:
Right, exactly.
Jonathan Mosen:
The report also covers the fact that most people are teaching themselves AI at the moment, but many would prefer formal or employer supported training. What does good AI literacy look like for blind learners?
Dr. Arielle Silverman:
Well, I think the AI systems themselves need to be accessible that blind learners are being taught to use. And then the training should be, whether it's provided in person or remotely, the training should be accessible. So, it should be readable with a screen reader or a screen magnification. If there's videos, everything that people are talking about in the video should be described. It should be easy to follow.
There should be maybe humans that people can contact if they have questions if it's a remote training. I mean, it was interesting that a fair chunk of people also said that they liked learning AI from friends and coworkers, so having more personalized experience learning from somebody who's more experienced with the AI could be helpful, and maybe personalizing it to the AT that the individual is using.
Jonathan Mosen:
Yeah, because I mean, obviously the way that you engage with AI will change a lot if you have a screen reader running, particularly reviewing how it has responded. You found both appreciation and alarm around AI psychotherapy. This is interesting because I suppose on the one hand, if it does it right it can make therapy available to those who may not be able to afford it, but then we're all very aware of the very tragic cases where some people have formed very unhealthy relationships with AI chatbots.
Dr. Arielle Silverman:
Right. So some of the advantages, it can help be more affordable and also available to people who don't have transportation to get to an office to see somebody. Although there's also remote human therapy is also available. People liked that it was available constantly whenever they needed to vent about something or they needed to check in, the AI was just immediately always right there. They felt like they were less judged. And this is probably not altogether an accurate perception, but people tend to think that AI is less biased than humans or less prejudiced.
So, those were the positive things. But then there was concern about, of course we know the cases where AI has encouraged people who are having self-harm or suicidal ideations, and so that's something that needs to be fixed in the models. They need to be trained on how to support people who are in crisis or preferably refer those people to another service that has humans. People felt like AI is not really at the point of being able to help people who are in crisis.
And then it was interesting, like somebody said they were talking to a chatbot and they have multiple disabilities, and they were talking about their disabilities and the chatbot recommended the death positivity movement, that they should read up on that. And so, that's kind of comes from the ableism of believing that people with disabilities want to die or that sort of thing. And so, we want to make sure that those models are disability affirming.
Like you said in the beginning, not saying, "Oh, I'm sorry," when you say that you have a disability, and understanding concepts of like disability pride, and not using the disability to overshadow, like if somebody has problems not assuming that it's all because of their disability. I mean, a lot of humans have that problem so we want to make sure that AI therapists don't have that problem.
Jonathan Mosen:
Melissa and Anil were talking about this topic, psychotherapy and blind people, on an episode of the Nation's Blind Podcast not so long ago. One of the observations that was made in that discussion was apparently blind people do have to suffer from therapists who, when they find out that you're a blind person and if you're turning up to their office with a white cane, or a guide dog or whatever, it's pretty jolly obvious. They then want to put everything through that blindness lens and blame everything that's causing you grief on your blindness.
Dr. Arielle Silverman:
Exactly.
Jonathan Mosen:
AI doesn't require you to disclose, right? I mean, if you-
Dr. Arielle Silverman:
Right.
Jonathan Mosen:
... choose to that's your choice, but you don't have to and the AI might not even know that you're a blind person.
Dr. Arielle Silverman:
Right, that's true. If your issues that you're discussing are not blindness related, you don't even have to mention it.
Jonathan Mosen:
Blind participants showed especially strong interest in autonomous vehicles. We were talking about this when you were giving us an overview earlier. Blind riders were notably less likely to call the experience fully accessible. What's the key lesson there?
Dr. Arielle Silverman:
Yeah. So, we have this contrast where on the one hand blind people are very excited about the possibility of autonomous vehicles and think it's really important that we invest resources in making them happen, and then amongst the people who actually rode in AVs, it was only about half said it was fully accessible compared with 75% of the sighted riders. The biggest issues people talked about, which I've also experienced personally, relate to locating the destination after leaving the vehicle, because the vehicle doesn't always park directly next to or across from the destination.
Waymo, for example, does provide walking directions to your destination, but in my experience those directions are not always accurate and sometimes even completely the opposite direction of where I should be going. And that might be a limitation in the compass on my phone or it might be something about the app. So people talked about that, feeling like they weren't being dropped off in a safe place or not knowing how to get to the destination.
The other thing is that the app, at least for Waymo, the app and the kiosk are not the same, and the kiosk is not yet accessible. So, there are a few features that blind people cannot use because we're using the app and they're not in the app, whereas they are visible in the kiosk that sighted people can use. And so, that does create disparities, but the main issue that people voiced was related to this locating destination at drop off problem.
Jonathan Mosen:
We've been having some discussion here in Maryland of late because there was legislation which did not pass the Maryland legislature this year relating to getting Waymo and other autonomous vehicles in the state. And it was interesting listening to the discussion on talk radio and just out and about from people who were asked, "Would you take one of these things?" And I was quite surprised, I guess, by the number of sighted people who said, "You would not catch me in one of those things," but it seems like blind people are much more positive about the prospect of being in an autonomous vehicle overall.
Dr. Arielle Silverman:
Yeah, and it was also people who could drive versus people who couldn't. People who could drive were a lot less enthusiastic, not surprisingly, because they don't necessarily need autonomous vehicles in their own personal lives. And people who couldn't drive, which was almost all of the blind and low vision sample, were a lot more enthusiastic about it. I mean, there were definitely concerns voiced about, I think equally amongst blind and sighted people, about safety.
Like, "I wouldn't trust this thing to get me to my destination," but it definitely seems like blind people are willing to put that aside more often because transportation is such a big barrier for us. Being able to overcome that with an autonomous vehicle is exciting. The other thing that came up, of course, is that AVs don't discriminate against guide dog users like Uber and Lyft do.
Jonathan Mosen:
Yeah. My wife's a guide dog handler and we now budget time into when we're supposed to leave home for an appointment just in case we get a ride-share refusal.
Dr. Arielle Silverman:
Right, right. You have to do that.
Jonathan Mosen:
I feel like we've painted a little bit of a bleak picture here, but it's a realistic picture. But I wonder whether overall do you think AI is a net benefit for blind people at this point?
Dr. Arielle Silverman:
I think so, and I think as long as we are conscious of the harms and actively work to reduce the harms I think the benefit is going to increase over time, especially once the reading technology gets more accurate.
Jonathan Mosen:
Was there anything in particular that surprised you as a researcher about what came back?
Dr. Arielle Silverman:
So, I think most of it was expected. I guess I didn't realize in our sample that everybody ended up being an AI user. We had a section for people who didn't use AI and nobody filled it out. Not one person filled it out. And I recognize a lot of that is just if you say it's a survey about AI you're going to get people who don't use AI or just not going to fill it out, but I guess I didn't realize quite how much people are using AI.
We did also find a few interesting differences when we compared generations, and we also compared gender and race on a few of these variables. So, older people are more enthusiastic about AVs than younger people, especially the older adults, people over the age of 65 were much more excited about AVs than younger people.
And that seems to mirror other research, and perhaps maybe because older people are just less likely to use public transportation and are more likely to drive, they think AVs are more important. Men also thought that AVs are more important, and we also found that people who did not identify as white were generally a little bit more positive about AI on a couple of different questions. They were more willing to use AI for sensitive information.
They were more trusting, or they thought that AVs were more important. And when we asked about how people traded off between independence and privacy, they were a little bit less focused on privacy. So, it seemed like maybe minorities, people who are not white might be more trusting of AI perhaps because of negative experiences they've had with racism from humans and they might believe that AI is not going to have those same problems that they've had with humans. I mean, that's a possibility. Or it could just be their different cultural attitudes about technology in different ethnic groups.
Jonathan Mosen:
And can you have a high degree of confidence with that sample size that this is true and not just some blemish in the data?
Dr. Arielle Silverman:
Well, in order to have a big enough sample we combined all of the groups that were not white. So that includes African American, Hispanic, Latino, Asians, Native Americans and multiracial people, and we had I think about 300 people in that subgroup. So, it's a big enough sample to have a reliable result, but we don't know if it was driven by one or two of those smaller subgroups. And that's something we might want to look at in future studies if there's a particular group that is just very pro AI and that's swinging it.
Jonathan Mosen:
And if people would like to read the report in detail, how can people find it?
Dr. Arielle Silverman:
So people can go to www.afb.org/airesearch2 since this is the second in our AI research series. That will take you straight to the report. We also have an executive summary that's about eight pages long if you don't want to read the report, or you can read specific sections of the report on the website. If you go to afb.org/airesearch without the two, it will take you to the page that has both of our AI studies on it.
Jonathan Mosen:
Thanks to Arielle Silverman from the American Foundation for the Blind for joining us to talk about their recent AI research. We'll come back after a break, and we've got a little time for listener contributions as Access On continues.
Speaker 4:
Are you looking for opportunities to promote your business? Become a sponsor for the National Federation of the Blind Access On podcast. This weekly podcast allows you to reach a targeted customer base of people interested in cutting edge technology products and services. Learn more at nfb.org/publications.
Jonathan Mosen:
Well, imagine that. I mean, your ad could be right here on Access On. How amazing. Now, we do get a large volume of email coming in to [email protected]. Don't let that discourage you, but we do read a representative sample of what we receive. We can't possibly read everything that we get. That said, let's have a look at some of the email that has come in, and this one says, "I'm Jessica Dale. I've been listening to your wonderful Access On podcast since the beginning." Well, thank you so much, Jessica.
She says, "I'm writing to you with a perplexing problem. My iPhone 16 Pro Max is exhibiting very strange issues, which may necessitate a restore this weekend. I do not mind doing that. However, I have recordings in the ooTunes app, which I would rather not lose.
Therefore, I need to export them. That's where I need your assistance. The developer gave me instructions, which I'll paste below. However, they do not work on either my iPhone or iPad. Have you or any of your listeners done this successfully and would be willing to share a process of doing this that works, and is of course accessible? The instructions are below."
If the recording is "saved" and you have a full iCloud or computer backup of the iPhone, it should include the recording once restored. However, if you want to export a recording from your phone, play the recording in the app and while it's playing, long press the record button, which is near the top right of the screen, for four seconds. And when you let up your finger, it will let you share it with a share sheet. Hopefully that helps.
I think this is Jessica continuing now. "I do have backups but plan to set up as new, which means that those recordings, and there are numerous, would all be lost. I've attempted to email him back several times to no avail. Based on this information, how do you suggest I proceed? I've even tried the free version of iMazing with voiceover on my Mac, but it keeps wanting me to buy a license every time I try to do something and there are plenty of unlabeled buttons.
Thanks in advance for any advice you can provide." Well, thank you for writing in, Jessica, and for listening to Access On. I'm wondering whether the issue is that that gesture that Steven, who's the developer of ooTunes, is talking about, may be difficult to execute with voiceover. If you can locate the record button by touch, then quit voiceover with a triple click of the side button, and hold down as he suggests on the button without voiceover running, I'm wondering if that will do the trick. I'm wondering if that will invoke the share sheet.
If it doesn't, then I guess there may be a trade-off. You may have to decide is it more important to set up as new or is it more important to have these recordings? If anyone else has any thoughts on this issue for Jessica, please be in touch [email protected]. Now, it's interesting what pushes people's buttons, because we had a lot of response to the email about listening to BBC streams when you're outside the UK. I will read a selection of the responses that we've received.
We'll go to Columbia and hear from Luis, and he says, "I'm writing to you regarding the mention in Access On about BBC domestic stations no longer being available for listeners outside the UK. I wanted to let you know that the Triode app for iOS still allows you to play BBC domestic stations, including Radio 3 and Radio 4. It is a fully accessible application.
In fact, it has become my default radio app for all my listening. Here are some key details about the app. It has an annual subscription of 10 U.S. dollars if you want the ability to save stations to your favorites list. It allows you to create shortcuts for quick access to your stations. You can manually add stations if you have the specific stream URL. By the way," continues Luis, "simple radio doesn't allow you to play these BBC radio stations. I hope this information is helpful for you and the Access On community."
Thank you, Luis. That's very interesting. Certainly the current radio app that I use regularly broadcasts doesn't have these BBC streams in there, so it's good to know that some still do. And we're in good hands with Triode. It's developed by the Iconfactory who of course used to do Twitterrific, a much loved app and our community when X was Twitter in days of yore.
Now we've got to cut way over to Canada, and hear from Steve Cutway. Amazing. He says, "I hope you won't mind a foreign interloper to contribute to the podcast." Oh, not at all, Steve. Not at all. We get people from all around the world contributing to Access On and we love that. He says, "I very much enjoyed your first part of the webinar last September in this week's episode about smart homes. I look forward to future episodes in this series. I have switched completely to the Amazon ethos from Google.
Nancy and I bought our first Echo Pop in 2024 and were very happy with it. Alexa+ was introduced in Canada in November 2025 and I immediately turned it on. Your description of it is very accurate, and five months later it's still a work in progress, although I have some fun and lengthy conversations with her. When I enabled Alexa+, the voice changed from the default you used in the recording. I call her my Valley Girl. You do know the Frank and Moon Unit song, don't you?"
Yes, of course I do, Steve. Yes. "I was actually quite impressed with the Echo Pop's music sound quality, so we ordered two more Pops in an Echo in December. Setting them up was easy. Once we set up the Echo, we could set up the new Pops using voice control alone. We've had an ecobee4 thermostat since 2018 and are very happy with it. Now I plan to add two ecobee smart doorbells, which are completely usable from Alexa devices." Look at you go, Steve, living the smart home automation dream in Canada. Good to hear from you.
Sy:
Hey, Jonathan, this is Sy from New York City, a newly minted member of the New York City chapter. My question is whether the NFB has ever had any contact with Substack about the accessibility of their website or their app. There's a ton of really interesting stuff on there, and I think it's a great platform for blind creators in certain ways.
I've been on it for a long time, just because it was a good place to put both writing and podcasts in one place, and put the subscriber content behind a single paywall. And I just figured accessibility wise, it was originally a newsletter platform so you could get anything I wrote in your email inbox, you could get the podcast in your podcast player and the accessibility of the platform didn't matter so much for consumers. For me behind the scenes it's been a bit of a pain, but everything has been trending in the right direction for some time.
But I get the sense that it's not trending in the right direction because of any concerted push toward accessibility. I think it just may be getting more accessible because of better design in general, but there are still many ways in which it could improve. And I think it's a great platform, not just for blind people to consume but also for blind creators. I've run into other blind writers on there. And the podcast host in particular is completely free.
And don't quote me on this, but as far as I know they don't mess with your audio, as far as I can tell. Anyway, they only charge if you start charging as a creator, they take a cut of whatever people are paying to you, that's their model. Otherwise, the platform is completely free to publish on, which is just something that I wish it was easier for us to use.
And there are constantly more and more people on it and lots of actually interesting, relevant voices, especially here in the U.S. So I'm wondering, has the NFB had any contact with them or looked into the accessibility of this platform at all? Thank you so much, love the show. Appreciate any thoughts you have.
Jonathan Mosen:
Sy, it's good to hear from you and particularly nice to hear that you are a newly minted member of your local NFB chapter. Regarding Substack, I've certainly used it to consume content, and you're right, there's a lot of interesting content on that platform. Certainly in the last little while, while I've been involved, I can say that the National Federation of the Blind's Center of Excellence and Non-Visual Accessibility has not had any contact with Substack about their service or their accessibility.
If you have any feedback about specific areas of the site that you think might be falling short, do let us know. We can add it to the list. Sometimes we can reach out, we can schedule a call, we can explain the feedback that we've received. And when we can collate it and pull it together in a way that's friendly for the development team or the product management team, it can really make a difference.
So if you, Sy, or others who are listening to Access On have any thoughts on Substack and there are things you would like us to pass on, by all means please do be in touch and we'll see what we can organize.
Joe Goode is doing lots of good out in Arizona with the National Federation of the Blind out there, and he writes in with a tech tip this week. He says, "Hello, Jonathan and Access On listeners. The below tech tip comes from my weekly newsletter, Tech Tip Tuesday.
Real quick, before I get into the tip, anyone who wishes to subscribe, please feel free to send an email to TechTipTuesday, that's all one word all squished together. Tech Tip Tuesday plus subscribe at groups.io. You will get small but useful technology every Tuesday," says Joe. Here's an example of what you'll get, because he's forwarding this from his newsletter, and this one is actually an ingenious workaround.
It says, "How to force your iPhone to say the correct words during dictation. Oftentimes when I dictate a text message or a note, Siri likes to say things that I don't want. Everyone knows what I'm talking about. Here is a great example. I'm dictating a message and talking about the JAWS screen reader. However, every time I say JAWS, Siri says jazz. Very annoying since I have to correct it each time.
That got me thinking. It won't say JAWS when I need it to, but yet it will often say the name of a contact properly. So, what would happen if I just added the word JAWS as a contact? Guess what? It worked. Here is how it's done. One, open the phone app and tap the contacts tab. Two, find the add button on the top right of the screen and tap it. Three, the first name field comes into focus. Type JAWS. Four, tap done in the top right corner.
Now if you are dictating, Siri should now write JAWS when you were talking about JAWS. Do this for as many words as you need. Tip. It feels strange to have contacts without a number. If you want to keep them organized, simply write the word dictionary in the last name field. Note, in my experience this trick has worked with most of the items that I have tried correcting.
Sometimes though it may not work particularly if you are dictating locations, but that is not usually an issue since Siri uses Maps to get its information. I do not claim myself to be an expert on this, but as with any dictation there are just some things that will not dictate correctly. Happy dictating," concludes Joe.
And if you have a tech tip that you'd like to share with us, you can write into [email protected], that's [email protected], and you can put tech tip in the subject. That concludes this episode of Access On, the technology podcast of the National Federation of the Blind. To send in a contribution for a future episode, email us. Attach an audio clip or just write it down, and send it to [email protected].
That's [email protected]. To keep up to date with Access On, follow us on Mastodon. [email protected]. That's [email protected] on Mastodon. To subscribe to an announcement-only email list about upcoming episodes, send a blank message to [email protected]. That's [email protected].
To learn more about the National Federation of the Blind, visit our website, NFB.org, or phone us, 410-659-9314. That's 410-659-9314, and be sure to check out the Nation's Blind Podcast right from where you heard this podcast.