Artificial Intelligence, History, and Blindness: Lessons and Musings

Why I’m Writing This

There are certain taboo subjects one is encouraged to steer clear of during get-togethers in polite society. Religion is one. Politics, even before the highly polarized world of today, is another. It seems we should add AI to that list, given the intense feelings the subject provokes.

You may well ask why I’d step into this minefield by writing a lengthy article about AI. A peaceful life is the way forward, and I should try it sometime.

I’ve written this article for three reasons. I promise that’s not because large language models seem to love grouping things into sets of three.

First, I come from a perspective that is often absent from mainstream and even technology media. I am blind, and I am also hard of hearing. I’ve used, advocated for, and even managed the development of access technology for decades.

Second, I remain an optimist, and I believe it’s possible for reasonable people to have conversations where we don’t adopt positions at one extreme end of a spectrum and then defend our corner as if we were cheering on our favorite team. (This is where I can get a “go Orioles” in the article).

Third, it’s not realistic to expect that we’ll stop the advancement of technology, so these conversations are important and necessary, even when some are entrenched and yelling. As a society, we have some big questions to work through. Technology is advancing, but precisely where are we as a society advancing to? What’s the plan? What mechanisms must we put in place to ensure that those leading that advancement have the moral authority to do so?

I’m no cheerleader for AI, nor am I exclusively a detractor. It’s a tool. Sure, it’s a powerful tool, and a tool with numerous flaws and risks. Tools can be used well, and they can be used lazily, and irresponsibly. There are problems we must talk about, but I want people to walk away from this article appreciating that AI it’s helping and empowering me, and other blind people like me, every day.

As a blind person who knows how it feels to often find oneself being the person things are done for and to, rather than with, I identify with the sense of disempowerment that comes from feeling like things that affect me are happening without my input, or even my knowledge. I feel this frustration regularly, in so many walks of life. There’s no denying that the decisions a tiny wealthy elite make can affect every one of us profoundly, from the work we do to the quality of services and products we use. It feels immoral that people no one had the chance to vote for can have such a dramatic impact on the future of humanity. But we are not going to change that by raging against the machine. We are going to fix it by insisting on a better process.

Burying our heads in the sand, and railing against the technology and each other, aren’t strategies that will help us confront the societal and even existential questions we face.

Disruption’s going to keeps disrupting

This discussion isn’t new. Human beings have wrestled with disruptive technology for as long as we’ve built tools. The industrial revolution rearranged whole societies. It put some people out of work, lifted others into work that had never existed, and changed forever how we live. The people living through it couldn’t see the societal destination, and the fear they felt was real. We should keep that in mind when we consider the fear people feel now.

The arrival of good old radio was another disruptive moment for an established industry. When the BBC in Britain began daily broadcasting in November 1922, newspaper proprietors were determined to protect their patch. They lobbied the government of the day, which agreed to keep the young broadcaster out of the news business almost entirely.

Newspaper publishers persuaded the government to ban the BBC from broadcasting news before 7pm and to force it to use wire-service copy rather than report on its own.

The newspaper industry had built something valuable, employed many people, and feared for their livelihoods. Their instinct was to wall the newcomer out. And for a while it seemed to be working.

Then came the event that exposed how brittle that wall really was. In May 1926, a General Strike brought the country to a standstill, and the national newspapers stopped printing. Almost overnight, the medium the government had kept out of the news business became the only way most people could learn what was happening. With newspapers halting production, the BBC emerged as the nation’s principal conduit for information, issuing multiple daily bulletins, five a day throughout the crisis.

Of necessity, people had been given a taste of the immediacy of radio news, and there was no putting that genie back in the bottle. The government eased restrictions from 1927, letting the BBC broadcast news earlier and produce its own bulletins rather than depend on the approved agencies. By 1934, the BBC had built a news operation of its own. The thing the newspapers had fought so hard to prevent became one of the most trusted news institutions in the world. And the papers, contrary to every prediction, did not die. They adapted.

If you are interested in this topic, I highly recommend a podcast that is doing a very deep dive into the history of the BBC, called “The British Broadcasting Century”. It’s presented with plenty of depth and a little humor, although it’s a British podcast, so of course it’s humour.

It eventually became radio’s turn to be disrupted. I’ll cross the pond for this example. After the Second World War, television arrived in American living rooms, and it did to radio very nearly what radio had once done to the newspapers. The new medium added pictures to sound, and it drew creative talent, listener loyalty, and advertising revenue away from radio. One by one, the great programs walked out the door and reappeared on television. While some transferred successfully, others disappeared from the airwaves. By the early 1950s, plenty of people had concluded that video had already killed the radio star. It had not.

The answer came from a station owner in Omaha. Todd Storz, with his program director, and Gordon McLendon in Dallas, built a format expressly as a response to the competition from television. Instead of the half-hour dramas that TV could now do better, they played a tight rotation of the most popular records, chosen based on what people were actually buying. They wrapped those records in brief news, weather and sports, frequent time checks, station promotion, and the personality of the local disc jockey. Oh, and the jingles. Who can forget those incredible jingles, many of which came from PAMS of Dallas. They called it Top 40.

If you want a single moment when radio’s new footing became undeniable, listen to airchecks from New York in 1964, when the Beatles arrived. (If I can get the Orioles into this narrative, I sure have to get the Beatles into it somewhere.) The city’s Top 40 stations went to war over the Fabs. Throughout 1964, 50,000-watt powerhouse WABC nicknamed itself WABeatleC. It was excelling where the television technology of the time could not. It was everywhere the teenagers with their transistor radios were, it was immediate, it was local, and it seemed to speak to you alone.

Now we live in an age where traditional broadcast media persists, but disruption has well and truly reshaped it. As someone who has built several Internet radio projects by the blind community, for the blind community, I welcome the collapse of the barriers to entry. When I host my Internet radio show, I don’t need a powerful shortwave radio transmitter to reach my global audience from my home studio. When I produce the technology podcast of the National Federation of the Blind, assisting people to make the most of the technology available today and facilitating us learning from one another, I celebrate how much easier it is to publish and consume global media.

In the late 1980s, I was already on the Internet, using a shell account at my university. Those were the days. Pine, Telnet, Gopher, and Usenet, no worldwide web. Back then the Internet belonged to a small community, I don’t think the word clique is unfair, and many in that community wanted to keep it that way. When commercial interests began to arrive, plenty of long-time users declared it the end of the world as we knew it. We can’t possibly let the great unwashed onto the Internet, where will it end? There’s even a name for the moment it tipped over. People called it the eternal September, the point in 1993 when the gates opened and the old culture could no longer hold the line.

Had those gatekeepers won, the Internet would have stayed a private club for people with exceptional computer skills and specific access channels. That would have been disastrous for a blind person like me. The mainstream, commercial Internet lifted blind people out of generations of information poverty. It’s the reason I can read a newspaper the moment it’s published, bank without assistance, take my time when shopping as I examine the options like everyone else, and do my job. No one would ever have built many of the devices we now take for granted for a private club.

I see the same instinct today among those who tried to keep Mastodon small and pure at the exact moment it had a chance to go mainstream. The urge to protect what you love by walling it off is human. It’s also, very often, a mistake.

With the Internet more widely available, music told the same story. As bandwidth and codecs improved, sending music over the Internet became the next logical and inevitable step. The recording industry first tried to stop it altogether. It went hard against Napster, rightly calling out intellectual property theft, but it didn’t offer an alternative suitable for an increasingly digital world. Then it tried to cage the new world in draconian digital rights management. Finally, it made its peace and adapted. The disruption was enormous, and I don’t minimize the careers it upended. But there was no holding back the tide.

As a boy, I saved up to buy one album a month. When I look back on that time, the music industry was taking advantage of that model to deliver a lot of slop. Sure, I bought plenty of albums that were works of art from start to finish, like Hotel California, Bat out of Hell, Thriller, A Night at the Opera, and every Beatles album. But I also bought a lot of albums because there was a song I liked that the label hadn’t released as a single, while the rest of it was slop that didn’t match the standard of that track I wanted. The artist recorded the rest of the material to make up a full album so they could take the pocket money off the kid.

Today, adjusting for inflation, for less than the price of that one album I have access to very nearly all of recorded music, and I can build custom playlists of what I want to hear. If I like an artist enough, I might attend one of their concerts, and I’ve even been known to travel some distance to do so. The thing the industry feared became the thing that saved it.

We could tell the same story many times over. Painters once feared that photography would make them obsolete. Instead, freed from the duty of mere reproduction, painting reinvented itself and gave us Impressionism and everything after. Teachers feared the pocket calculator would destroy children’s ability to reason with numbers.

In taking some time to paint historical context around the inevitability of disruption in an evolving society, I don’t mean to pat anyone on the head, tell them not to worry, and claim that everything will be OK. But I’ve developed this point at length because some who object the strongest to the evolution of AI come from an industry that has tended to wield the disruption, rather than suffer an existential threat from it.

When the Disruption Comes for Software

For decades, the people at the heart of all this disruption have been software developers. They are the ones who reshaped every gadget we own and nearly every transaction we make. Now, for the first time, the disruption has come for them. And I notice that some who cheered disruption when it fell on others find it far less appealing when it arrives at their own door.

The fear of losing your livelihood is real. I’ve been there myself, and it’s horrible.

That said, I can’t pretend to share the view that all of this is a catastrophe we must resist. As history teaches us, that won’t work. I believe history also teaches us that the old ways left too many people out.

In various capacities, I’ve spent decades pressing the world to take the needs of blind people seriously. I’ve had the privilege of working with software developers who get it, and I’m grateful. Some of them are blind, and have worked tirelessly and selflessly to make the world a more accessible place. Many nonblind people have joined them, understanding that the blind are in the best position to determine our needs and how best to meet them, and in that spirit they offer their advice and considerable skill. Over the years, it has given me much joy to have raised accessibility concerns with nonblind people who weren’t even aware that blind people were using computers at all, but when they learned of it, they became committed to making the software they worked on the best it could be from an accessibility point of view. Those experiences maintained my optimism during the tough times, filling me with hope and gratitude, but they haven’t always been my experience.

For decades, too many software developers and product managers have told me that accessibility isn’t a priority, that we should wait our turn until some unspecified time, that accessibility isn’t a “feature” worth supporting.

Just as with many industries before it, disruption is now reshaping software development, and the barriers to entry are crashing down. Anyone who knows me knows the values that are dear to me. I believe blind people must determine our own destinies. We will no longer accept someone deciding they know what’s best for us and imposing a solution on us that we didn’t help to create. In 1824, a blind teenager in Paris invented a code that gave blind people true literacy. Braille was perhaps the first transformative example of us devising our own technology, but it was far from the last.

So when I see blind people building solutions for our own needs, when I see crowd-sourced feedback influencing where those products are going and which defects are prioritized, I feel joy. We have more power and more control than we’ve ever had.

I’m seeing a glorious renaissance of software that blind people are releasing, plugging accessibility gaps we have endured and pleaded with developers about for years.

Unfortunately, I’ve also seen capable, good-hearted, dedicated, decent people set upon by an online mob who feel they have the right to mercilessly and irrationally attack someone, just because the software that is having a positive impact on the lives of blind people was developed even in part with the help of AI. That is unacceptable, it must stop, and I believe we all have a duty to call those doing that kind of bullying to account.

Many blind people may make unusually good vibe coders. We already understand user interfaces deeply, because we’ve had to. We are practiced troubleshooters because of deficient software created in the past by humans. We are experts at workarounds, and at truly understanding the tools we depend on. So now we can describe what we need and build it, refine it, critique it. That’s liberating. It’s democratizing. We’re moving ourselves to the front of the line through our own effort, rather than waiting for a turn that seldom comes. To me, that’s self-determination.

Vibe coding has the potential to redefine how we think about software, and I have been reading many articles on this subject lately. I dabble in vibe coding, but I have also been a product manager of some of the most successful access technology products in the blindness market, so I know what’s involved in properly supporting a product after releasing it. I know that, given the important work I am doing right now, I would offer lousy support and I have no intension of making anything I produce public. But that’s OK.

As a screen reader user, I am obsessed with efficiency. So much so that I tend to think that every unnecessary syllable is something to eliminate. Not every blind person feels that way, and even when they do, one person’s unnecessary verbiage is another’s essential information. If we can build software based on open standards, the software becomes the means, not the end, and it is possible that we will see an increase in software that someone writes for their own specific preferences to get their specific job done. For example, if I am producing a podcast, does it really matter to the audience how I did it? As long as the final product sounds good and it turns up in a format their podcast client can work with, they have the product they expect. That’s quite the shift with potentially positive accessibility ramifications.

 

Code, Safety, and the Human in the Loop

There’s work to do. Some software professionals warn that AI writes insecure, careless code, and the evidence backs them. One large analysis found that only 55% of AI-generated code was secure, meaning nearly half of it introduced known security flaws. A Stanford study found that programmers using AI assistants wrote less secure code, and worse, developed a false sense of security, believing their code was safer than it was. With something like 46% of code on major platforms now AI-generated, that’s a real and growing problem. If a blind developer, or any developer for that matter, ships code built on a tool that introduces a vulnerability, that vulnerability can harm real people. “Human in the loop” has become a bit of a buzz phrase, but it is plainly irresponsible to publish code without review, just as handing AI a vague prompt to write a document and letting it do all the work will give you mediocre results.

But let’s not pretend we’re transitioning from some sort of utopia. Anyone can teach themselves to code and then publish something dodgy. Software has always had bugs and vulnerabilities. As mainstream technology companies have become access technology companies too, blind people have watched bugs unique to our use case linger for years, not because they lack impact, but because the blind are the tiniest fraction of the total user base, so developers deprioritize us. Blind people know all too well the consequences of unresolved serious issues. That’s been our reality, thanks to discriminatory ranking processes far too often used by human software developers.

We’re starting to see models, such as Mythos from Anthropic, that detect serious security vulnerabilities. While we’re at it, let’s teach these models that assist with software development that inaccessible code is buggy code, and Buggy code is unacceptable code. That would deliver immeasurably positive social change. For that to work well, blind end-users must be involved in training these vibe coding models about what good in an accessibility context looks like.

What Medicine Teaches Us

We’ve already watched this tension play out in a field with far higher stakes than software. Medicine.

The medical establishment has long been uneasy about ordinary people taking charge of our own health. Sometimes that caution is wise, because self-diagnosis can go badly wrong. Sometimes it’s all about something less noble, a sense that health is the exclusive territory of the credentialed, and that the rest of us should sit quietly and wait to be told what is allowed. The gatekeepers have often proven to be too cautious.

Take the Apple Watch. When it began offering heart monitoring, plenty of clinicians were skeptical, worried about false alarms and anxious patients flooding clinics. The evidence has been kinder to the watch than to the skeptics. The Apple Heart Study enrolled roughly 400,000 people and showed the device could detect an irregular heartbeat. In a more recent randomized trial of adults over 65 at higher stroke risk, doctors diagnosed atrial fibrillation in 21 people in the watch group, against just 5 in standard care, and 57% of them had no symptoms at all. Atrial fibrillation raises the risk of stroke, so for some of those people the watch on their wrist may have saved their lives.

It’s also true that the skeptics weren’t entirely wrong. A Cleveland Clinic study found the watch’s on-screen notification flagged only 41% of atrial fibrillation, even though the electrocardiogram it could produce correctly identified 96% of cases, with no false positives. In other words, the tool was genuinely life-saving and genuinely imperfect, both at once. The right response was never to ban it, nor to trust it uncritically.

Diet tells a similar story. My late parents both experienced significant heart issues. They both had heart attacks before they were the age I am now. Once they developed heart problems, they followed medical advice strictly. Given that I clearly have a family history, me being me, I dug deeply into the subject, working out what I should do to optimize my health.

For years, much of the establishment discouraged low-carbohydrate eating, while a great many ordinary people, and a stubborn minority of clinicians, found it worked for them and pressed ahead without waiting for approval.

I still remember going to my doctor, getting great numbers back from medical exams, and the doctor asking what I was doing to achieve them. When I told the doctor I was eating low carb, they were horrified. The numbers suddenly didn’t matter, which was plainly absurd.

The official position has since turned around. The American Diabetes Association now recognizes low-carbohydrate nutritional therapy as a viable option for managing type 2 diabetes, and states that reducing overall carbohydrate intake has the most evidence for improving blood glucose control. Its 2026 standards list low-carbohydrate eating among the patterns with the strongest evidence for preventing the disease in the first place. The people were ahead of the experts, because we trusted our own bodies and knew what worked.

That is exactly the posture I want us to take with AI. When I hear that AI-written code can be dangerous, I hear that concern and give it consideration, but then, we are also told by some companies that sideloading is dangerous, and at least some of that fearmongering comes from vested interest who wish to protect a revenue stream.

I will not accept the danger as a reason to tell blind people, or anyone else, to put their tools down and wait for permission from gatekeepers, too many of whom have ignored us, a few of whom have even treated us with absolute contempt in the past, and who have a vested interest in the status quo. We can take the warning seriously and take charge at the same time.

Blind People Have Long Led the Way

Has the AI industry overpromised and underdelivered? On balance, yes, I think so, and that’s done nothing to raise enthusiasm or trust. As a microcosm of society, blind people are no exception; the hype has let us down too. But I think it’s credible to argue that blind people are feeling the benefits of AI disproportionately more than those who aren’t blind. History teaches us that this is often the case with technology. Blind people have both invented new tools and received them early, years or decades before nonblind people realized they wanted them too.

In the early nineteenth century, an Italian named Pellegrino Turri built one of the first typewriters so that his blind friend, Countess Carolina Fantoni, could write letters without a scribe. Many also credit him with inventing carbon paper to provide the ink for his machine. The letters the Countess typed survive today as some of the oldest examples of typewritten text in existence. The typewriter would go on to reshape every office on earth. It began as a way to give one blind woman her privacy and her independence.

A generation later, the principle repeated in Paris. We remember Louis Braille for the priceless gift of the code that bears his name. Less remembered is that his friend Pierre Foucault, also blind, built a machine in 1841 to write Braille’s decapoint, a system of raised characters that nonblind people could read without training. They called it the raphigraph, an early writing machine of real ingenuity that predated the commercial typewriter by decades. What matters most here is that blind people were the first in the world to represent print as dots, giving us the concept that would later feed dot matrix printing, and the computer screens everyone uses today.

In the 1930s, the American Foundation for the Blind needed a way to put whole books onto disc. The result was a twelve-inch record turning at 33⅓ revolutions per minute, pressed in flexible, durable Vinylite, with far more grooves per inch than the brittle 78s of the day. Blind people were using the LP from 1934, almost a decade and a half before it reached the commercial music market in 1948.

Then came the machine that changed my own life and millions of others, an early example in fact of pattern recognition. In 1976, Ray Kurzweil unveiled the first print-to-speech reading machine for the blind. The idea had come from a blind stranger on a plane, who told him that his real barrier was getting at ordinary printed material. To solve it, Dr. Kurzweil combined three inventions: omnifont optical character recognition, the flatbed scanner, and text-to-speech synthesis. Scientists from the National Federation of the Blind tested the prototype and shaped it with their own suggestions yet another example of what can happen when the blind are involved at the product development stage. Today, the flatbed scanner sits in offices and homes everywhere, though camera technology is now more common as the technology has been miniaturized. OCR is in every phone. Text to speech reads to commuters and answers back from smart speakers in tens of millions of homes.

Blind people were carrying notetaker devices around long before nonblind people.

What these and many other examples illustrate is that when you’re shut out of the so-called ordinary way of doing things, we will develop alternative techniques, and those techniques often turn out to be better for everyone. So I think it’s prudent for the world to take note of the way AI is truly enhancing the lives of blind people.

How AI Helps Me Every Day

I’ll offer a few examples of how AI has materially improved my life as a blind person. All I ask is that, while we work to improve the many undeniable and serious weaknesses of the current technology and its wider social rollout, we appreciate that turning back the clock, even if we could, would deprive blind people like me of useful tools like these.

I appreciate the increased ability I now have to get audible descriptions of visual things. One of the most telling examples I can offer is the way my wife Bonnie and I treated photography at our wedding, compared with the way we approached it when we held a cheesy tenth-anniversary vow renewal ceremony in Las Vegas last year. Bonnie is also blind. When we married in 2015, we told people they were welcome to take all the pictures they wanted at the ceremony with their cameras or smartphones, and to send us any they thought worth keeping. We got some, and we kept them, but they sat on our Synology NAS untouched, because we couldn’t appreciate them.

Going all in on the Vegas experience ten years later, we held our vow renewal at a chapel which offered photographers as an option. Rather than opt for no photography, this time we chose two. The difference was AI. We now have access to several tools that offer detailed descriptions of the photos, with the ability to ask clarifying questions about each picture. It struck me just how different the world is now when we sat down one evening soon after our return from Vegas to go through the photos, to share memories and to smile about the experience, just as many couples do. We simply couldn’t have done that a few short years ago.

I use that sort of technology every day, whether for reading mail, checking the view outside my hotel room window when I travel, identifying packages, or working out what the buttons on an unfamiliar remote control do. Sometimes it delivers, sometimes it doesn’t, and I often don’t know when it has failed.

A few months ago, Bonnie and I moved into our new home and got connected with Xfinity. We received a set-top box and a remote control to operate it. I took a picture of the remote and asked an AI tool to describe the order of the buttons, left to right, top to bottom. It gave me a detailed description. The trouble was, I soon worked out that while the remote it described was indeed an Xfinity remote, it wasn’t the Xfinity remote we have. Yet it sounded so sure of itself. Only when I insisted that the order of the buttons didn’t match what was in my hands did it finally get it right.

This is another example of where having a human in the loop is valuable. One tool, Access AI from Aira, lets you send your picture and the AI description to a professional human agent for verification.

I also believe AI image recognition should offer confidence levels as a matter of transparency.

Until recently, getting a description meant stopping, taking out my phone, framing a photograph, and waiting. That’s certainly useful, but it pulled me out of the moment and occupied my hands. Now the camera can be on my face. Ordinary-looking smart glasses, the mainstream kind a sighted person might buy to take photos, happen to carry a camera, a microphone, and open-ear speakers. I say a few words and have either an AI or a real human look through the glasses and tell me what’s in front of me, while my hands stay free for my cane and perhaps some groceries. I can ask what’s on a shelf, read a notice taped to a door, or get a description of an outfit before I leave the house. The tools are far from perfect. The battery runs down too fast, the text reading still stumbles, and I treat what the AI tells me with the same caution I bring to everything else. But the leap from holding a phone to simply turning my head is larger than it sounds.

Again, I hear of pushback. Someone recently told me of a sign on a store that proclaimed AI glasses not to be welcome. I imagine that the rationale is that they are creepy and intrusive, and staff don’t want to be recorded. But I consider the glasses to be a legitimate accommodation that assists me to interact with visual elements, and I will do everything I can to shop elsewhere, rather than legitimize this sort of behavior by giving the store my business.

Staying with the visual but moving to work, AI helps me immensely. If I need to interact with a complex spreadsheet, in the past I would have had to explore it cell by cell to work out the trends. AI tends to be excellent at summarization, so I have had excellent results asking it to summarize such spreadsheets.

Several AI tools have become competent at producing PowerPoint presentations. I can write my speaker notes in Word, send them to Copilot or another AI tool, and ask for a PowerPoint presentation with appropriate graphics. I can get it done in minutes, rather than take a long time tinkering with a medium that, done well, is very visual.

We tend to frame AI for blind people as a way to take the visual world in, but it also lets us use text descriptions to put things into the visual world. The same ability that builds my slide decks now makes images. For the first time, I can describe a graphic, a logo, or an illustration, have it made, and then ask for a description to check the result against what I intended. Sometimes, I find I have to attempt this a few times before I’m confident I have the output I want. For most of my life, anything visual like this meant seeking human assistance. Now I can produce a great deal of it myself, and verify it myself. The street runs in both directions.

AI is lifting my performance. With the right integrations between Zoom, Plaud, which makes note-taking hardware and software, and Zapier, which now has an effective Copilot mode for vibe coding complex integrations, I can send meeting notes to AI, have it produce a list of tasks that the meeting assigned to me, and add those tasks automatically to the Todoist task manager. Plaud even labels the speakers for me. AI has made me more responsive and more on top of my game, with less effort.

For a blind person, the web is far too full of human-generated, inaccessible slop. Too often I visit a website or use an app to get information I want, only to find it poorly designed. All my screen reader can do is read “button, button, button.” Sometimes the page sends my screen reader jumping to a random point, or speaks a message over and over. I’ve often found exactly the information I wanted through a search engine, only to discover that the page holding the information isn’t accessible. AI has helped me here too, in the form of the deep research tools many such products now offer. Deep research tends to hallucinate much less, and it synthesizes masses of information without my having to grapple with inaccessible sites. And yes, it saves me time, just as it saves anyone else time. Before we bought our new home, I used deep research to produce a comprehensive document describing the neighborhood, whether the constitution of the Home Owners’ Association held any red flags, and what Internet connectivity was available. I also asked it to find every picture of the house it could, and from those pictures to write me a narrative walkthrough describing the house’s appearance and features.

Much left to do

We are making rapid progress. Some of the issues we must resolve are now ethical rather than technical. Everyone else can identify people they already know in a crowded room, and AI permits a blind person to do that too. If I know you from somewhere, whether it’s because I work with you, we socialize together, I follow you on social media where you have made your face visible in your profile, or you can reasonably expect to be considered a public figure, it is equitable to know you’re in the room with me, just as everyone else knows you’re in the room with them.

The technology exists now. What’s holding us back is agreeing on a standard and a reasonable set of safeguards that ensure the tools are not privacy-violating while also not depriving blind people of information we would have were we not blind.

I’m optimistic that the future of computing will be more inclusive. Sometimes we forget that the vast majority of blind people become blind as seniors. I’m not for a moment suggesting that just because you’re a senior, you’re incapable of learning a screen reader. But often, seniors have a lot going on when blindness comes. Even if they’ve had some computer experience in the past, screen readers introduce a new modality, full of unfamiliar keyboard commands and gestures.

There will always be a place for blind people who need the efficiency of a screen reader with speech cranked way up, rapidly engaging with an application using a myriad of keyboard commands or gestures. But AI can deliver a more conversational approach that will benefit people who have needs such as contacting their grandkids or playing some friendly games.

Agentic AI has the ability to benefit all of us using access technology. If a blind person can zip through a complex purchasing process without wading through a massive number of links and options with varying levels of accessibility, that raises efficiency and lowers stress.

Consider transport. Bonnie and I have to build extra time into our schedules whenever we take an Uber or Lyft, in case a driver refuses to carry us because of her Seeing Eye dog. This happens regularly. It’s hard to convey the frustration of trying to go about your lawful business, only to find yourself, out of the blue, in a confrontation with a driver who won’t do their job lawfully. We just want to get where we’re going like everyone else. On one occasion we set out on a date night, only for a driver to take us with reluctance and harangue us at the top of her lungs the whole way from home to the restaurant. It was so serious that I quietly got out my phone to report a safety issue and seek help.

When we do get a ride, a stranger often quizzes us about our medical history just because we’re blind. So when people ask how I feel about autonomous vehicles, my answer is simple. Give me one any time.

I have spent this section celebrating what AI gives me now and what I expect it to do in future, so let me be honest about what I hand over in return. Every time I ask a tool to describe a photograph, read my mail, or tell me what’s outside my hotel window, I send a picture of my private life to a company I don’t control. The tools that enhance my independence also learn where I live, what my home looks like, and what arrives in my mailbox. I have spent my life fighting gatekeepers who decided what blind people could and could not have. I do not want to win that fight only to hand the keys to a new set of gatekeepers who merely happen to be friendlier today. So I use these tools, and I keep my guard up. We need competition, and we need tools that run on our own devices wherever possible. There are several promising blindness tools using local LLMs, and I hope they continue to improve and flourish.

These are just a few examples of how, when I use AI in a spirit of partnership, skepticism, and curiosity, it’s improving accessibility immeasurably. I ask that those who take to their keyboards, ferociously demanding that AI simply stop, appreciate that when they say that, they’re calling for a world in which some of us will genuinely be worse off. I’m happy to work with those who have legitimate concerns and approach them in a spirit of intellectual curiosity and goodwill, I suspect I share some of those concerns. But I am not prepared to surrender the gains that have made the world more equitable.

AI and Hearing Loss

AI is also making a positive difference for those of us who are hard of hearing. Modern hearing aids now run machine learning models that separate speech from background noise far better than the fixed compression schemes of even a few years ago. They learn to tell the difference between the voice you want to focus on and the voices behind it. On phones and computers, live transcription has crossed the line from novelty to genuine tool. One of the most impactful things Apple has ever done was to introduce accessible live captions as part of its Braille Access feature in iOS 26. At meetings with no assistive listening, I use this a lot. I use the term “game changer” sparingly, but this feature has changed my life for the better.

AI has also come to the rescue when organizations that should know better have failed. In 2025, I attended two well-known conferences on accessibility that didn’t offer assistive listening devices in their breakout sessions despite my requests for such a device well in advance. I hope the irony that these were access technology conferences isn’t lost on anyone. The rooms were too full of echo for live captions to work well, so I sat there in frustration while Plaud helped me out. My Plaud device recorded the sessions I couldn’t hear, and generated near-perfect transcripts and summaries.

I want to turn to music, which has always played an important part in my life. Hearing loss does not erase musicality. It erases reliability. I used to make a lot of music, but with degenerative hearing my pitch is no longer reliable. Plenty of musicians keep perfect internal pitch and a deep sense of arrangement long after their ears stop giving them accurate feedback. The voice wanders not because the musician has forgotten the note, but because the ear can no longer confirm it.

I accept that AI music tools can be misused or deployed lazily, but they’re also genuine accessibility tools.

Services like Suno offer a way around the performance barrier without abandoning the creative one. Just as with vibe coding and many other pursuits, I see people derisively dismissing all AI music generation as a slot machine where you type a prompt and accept whatever falls out. Used that way, of course it produces exactly the forgettable output the critics complain about, just as a chatbot does when you ask it to write a poem. But that’s not the only way to use it. If you bring real musical knowledge to the work, Suno becomes an instrument you direct rather than a vending machine you feed. You can specify structure, key, tempo, mood, instrumentation, and arrangement. You can write your own lyrics, which carry your ideas and emotions. You can upload some basic instrumentation and ask Suno to develop or orchestrate it. You can even upload a sample of your own voice and, gloriously, it’s on pitch again. You can iterate on a section until the phrasing matches what you hear in your head, regenerate the parts that don’t work for you, and stitch together the takes that land, all accessibility permitting, of course.

Done with that level of intention, it is your song, realized through a tool that compensates for the one ability hearing loss took away.

Again, the backlash isn’t new. Even before we had the megaphone that is social media, I remember all the outrage when musicians started to use drum machines and synthesizers.

Last Christmas, I wanted to give Bonnie something truly special and personal to her. I wrote a set of lyrics myself. I prompt-engineered a very clear description of the kind of music I wanted Suno to produce. I refined it, and refined it again. It took a long time to get what I wanted, I estimate about 13 hours. But in the end, it was a song worthy of Bonnie. It made such a cool Christmas present, and I felt moved and empowered that this technology let me produce music again that I had confidence in.

So, when people dismiss anything created with the assistance of AI as slop, that’s far too simplistic.

Using AI as a Partner

The examples I have given of how I find AI effective have for the most part demanded a lot of human thought and effort. When there’s partnership and the human sits in the driver’s seat, AI works better. Many people are like kids with a new toy right now, handing over so much to AI that they’ve lost their voice. I like to think that will settle down. Those who expect AI to produce a document that’s ready for prime time on the first draft will face credibility problems. People will be able to tell that it lacks soul, that it lacks the touch that makes it yours. My own preference is to write first, then use AI to refine, to test clarity, and to see whether it can offer rebuttals to my arguments that I should address.

In case you’re wondering, sure, I used AI as part of the drafting process of this article. I wrote out a full draft while having AI open in another window, asking various clarifying questions. I then fed my first draft to AI, and asked it to argue with me. What points had I not covered that I should have. When had my language missed the mark with what I am trying to do here, which is to be honest without being unnecessarily confrontational? It’s a better article because I involved AI this way. It didn’t do the work for me, it enhanced and improved the quality of work that those who have read my writing for years will clearly be able to tell is mine.

We have the order backward when we teach the tool before the skill. A person needs enough underlying ability to judge what the AI hands back. We have to find a way to continue teaching the fundamentals first, and only then introduce AI as a way to go faster and further, never as a substitute for knowing what good work actually is.

I read a lot about how our use of AI is evolving alongside its capabilities, and some innovative thinking is going on at the university level to make sure students genuinely have to think for themselves.

Alongside the skill, we must teach the habits of verification. The future will still belong to those of us who continue to think critically. Ask the tool for its sources. Ask it to argue against its own answer. Ask it where it might be wrong, and what it would need in order to be more certain. As we train people in responsible AI use, we should give them real tasks where the AI is confidently mistaken, and let them experience that for themselves, so they learn to recognize the particular texture of a wrong answer delivered with total assurance.

There are concerns here of particular interest to the blind community. Quality blindness skills training can be hard to find. That is partly a matter of insufficient funding, but it is also a matter of the low expectations others hold of blind people, and in some cases that we hold of ourselves. It would be a tragedy to watch quality blindness training options erode because someone, somewhere, decides that the machines can now take care of us. AI tools can augment our blindness skills, but they cannot replace them.

We face a moment of great opportunity and very significant risk, in that we must demand that AI does not simply absorb and repeat the prejudices that have held blind people back for generations. A system trained on a world that underestimates us will underestimate us too, so we must insist otherwise. Through the work I have the honor of doing on behalf of members of the National Federation of the Blind, I know that the Federation is actively working on that problem, and I can tell that the work is paying off. I’m heartened by the initiatives that several big tech firms have put in place relating to responsible AI, and by their growing recognition of the need for quality data sets comprising real-world information about the way blind people live our lives. When that data conveys the truth that blindness does not hold us back with the right training and opportunity, and when that data assists with responses to questions, influences recruiting decisions, and guides public policy formulation, there’s a real chance that our tomorrows will be better than our todays and yesterdays.

The Intellectual Property Question

The harder questions reach well beyond accessibility. As a society, we must work out how we define and reward intellectual property in this new era. The companies building this technology aren’t the right people to settle that question alone, because they have a vested interest. We have to find a consensus, in an age when consensus on anything at all feels almost out of reach. I don’t have the answer. I only know that the question belongs to all of us.

I have, however, had some past experience with changing concepts of intellectual property and theft. In the 1990s, I ran the government relations program for the blindness agency in New Zealand. In a world first, we advanced the idea that access to information was no different from access to the built environment. It was a human right, and it needed to be enshrined in law. That was very radical thinking for its time. Back then, copyright law around the world required organizations producing accessible-format material to seek the permission of the copyright holder. The copyright holder was within their rights to refuse permission, and sometimes they did. That deprived blind people of information that nonblind people could access without a second thought.

We came up with the idea of inserting a clause in the Copyright Act that let organizations producing accessible-format material produce it as of right, with no need to seek permission. I vividly recall a phone call from the man representing authors and publishers. “Mr. Mosen,” he said, “do you steal from everyone, or do you just steal from authors?”

To him, making a printed work accessible as of right was the theft of intellectual property. In that case, the politicians of the day disagreed with him unanimously.

In 2013, the nations of the world gathered under the World Intellectual Property Organization and adopted the Marrakesh Treaty. Across borders, and as a matter of binding international law, it does what that phone call had assured me was stealing. It requires countries to allow accessible-format copies to be made, distributed, and shared without the rights holder’s permission, and it came into force in 2016. The United States ratified it in 2019. What one publisher once called theft is now the substance of a human rights treaty.

So when I hear that same word, theft, thrown around the debate over AI and copyright, I must admit it’s a wee bit triggering. I remember that I’ve been on the receiving end of that accusation, and I was no thief. History agreed.

But I’d dishonor my own argument if I left it there. Our exception worked, and eventually won the world over, precisely because it was principled. It served people who were otherwise shut out completely. It didn’t compete with the author’s market, because a blind reader would gladly have bought the book if only it had existed in a form we could read. And it stayed within the limits that international copyright already set to protect a creator’s legitimate interests. We widened access without interfering with the reward for creation. That balance was the entire point.

The AI question is not the same question, and I won’t pretend that it is. When a company trains a model on millions of copyrighted works, the use is commercial, it is vast in scale, and the model it produces can go on to compete with the people whose work it learned from. That makes the creators’ worry far more consequential and worthy. A novelist who fears that a machine trained on her life’s work will now flood the market with cheap imitations of it is genuinely and legitimately worried about a real threat to the livelihood that makes the next novel possible.

Both lessons show us that we must think creatively and radically about how AI can widen access and serve the public good. We should be slow to let the word theft end a conversation it ought only to begin. In the same breath, we must ensure that we protect and properly reward the people who create, because a society that stops paying its authors, its musicians, and its artists will soon have very little left worth training a model on. The answer for AI is the same kind of answer we found for accessible books. Not a land grab, and not a flat refusal, but a principled settlement, drawn with care, that serves a genuine public good while keeping faith with the people who make the works. We managed it once, for a cause that the experts of the day called theft. We can surely find the imagination to do it again. I certainly do not want to live in a society devoid of human creators.

What Lies Ahead, and the Conversation I Want to Have

Finally, I don’t underestimate the magnitude of the social disruption that lies ahead. As I’ve shown, there’s been plenty of disruption before, and society has largely withstood it. But the scale and speed of the disruption to come is like nothing we’ve ever faced. If we’re not deliberate and careful, we risk fragmenting social cohesion completely.

I’ve leaned on history throughout this article, so I owe you the strongest argument against my own optimism. In every previous wave, the machine took some jobs and created others, and people moved into the work that opened up. The fear this time is that AI is general in a way no earlier technology was. It doesn’t automate one trade and leave the rest alone. It reaches into writing, analysis, design, coding, and the white-collar work that people fled to when the factories fell quiet and the switchboards went dark. If the tool can also do the new jobs, then the comforting lesson of history, that there’s always somewhere to move next, may not hold this time. I don’t know whether that fear will prove right. Nobody does. But we can’t wave it away, because it’s the most serious worry of all, and the people raising it aren’t fools.

This technology is so consequential that we must find the balance between encouraging innovation and proceeding with a plan that ensures it works with us and for us, not against us. We haven’t found that balance yet, and that’s the justifiable source of much of the fear.

We need to think hard about the concept of work and reward if the jobs simply aren’t there.

There’s another cost I can’t in good conscience leave out, even though it does my argument no favors. These systems consume staggering amounts of energy and water. The data centers behind them draw on power grids and water supplies that are already under strain, and the communities living nearest to them don’t always get a say. When I celebrate what AI does for me, I have to weigh it against what it asks of the planet and of people who may never see its benefits. Efficiency is improving, but the honest position is that we’re running an enormous experiment with shared resources, and we owe it to everyone to count that cost properly rather than pretend it away.

As a grandparent, it breaks my heart that some of these systems have amplified people’s mental illness instead of guiding vulnerable people toward care. A solution to that issue can’t wait.

So no, everything is not fine. But the point of this article is to argue that we won’t solve any of these problems by imagining we can somehow magic society back to a previous state. Nor, as a blind person enjoying many meaningful benefits, do I want us to. I do not want to go back to that less accessible, less empowered, less self-determined, less inclusive place. And we will certainly not solve them by turning on one another.

That last part troubles me most of all. The conversation has grown so hostile that good people have gone quiet. I know many who hold thoughtful, mixed views and who now say nothing, because they have watched others buried under a pile of furious vitriol for daring to express anything but total opposition. When even a polite disagreement earns a pile-on, we’ve stopped having a conversation and started running a mob. That helps no one, and it fixes none of the genuine problems I’ve described.

I have decided not to stay quiet. Yes, AI has many problems. So did the software that many of the people now sounding the alarm wrote. Software has shipped with bugs and caused harm, and we’ve fixed and improved it, for as long as software has existed. AI is no different in that respect. It’s doing real harm in some places, and real good in many others.

We cannot turn back. But we can decide, together, what we move toward, and we can insist on a process worthy of the stakes. That is the conversation I want to have. I hope you’ll have it with me.