The Internet Is Learning to See. It Still Doesn’t Know How to Listen.

I am blind, so I experience the internet differently from the way most technology is advertised.

I do not look at a beautiful homepage and decide whether the design feels modern. I hear it.

I hear the headings. I hear the buttons. I hear the links. I hear whatever a developer decided to call an image — assuming somebody bothered to describe it at all.

And sometimes I hear almost nothing useful.

A website can be visually stunning, cost a company millions of dollars and use the newest technology on earth, yet the experience through a screen reader can feel like walking into a building where somebody forgot to install the doors.

That contradiction has bothered me for years.

Now it bothers me even more, because we are entering an era in which technology companies are promising something extraordinary: artificial intelligence that can understand what is on our screens, describe what is around us and increasingly take actions on our behalf.

Apple previewed accessibility updates in May that use Apple Intelligence to provide more detailed image descriptions in VoiceOver and support natural-language interaction in accessibility features. For a blind person, those are not novelty features. They can translate into something much more important: independence.

At the same time, the technology industry is racing toward so-called AI agents — software that does more than answer questions. These systems are increasingly being designed to shop, book services and complete tasks. On Sept. 22, Reuters reported that a group of major banks had warned that AI shopping agents could create new risks involving scams, incorrect purchases and data privacy.

We are teaching computers how to operate the internet for us.

That is remarkable.

But there is something deeply strange about building an internet sophisticated enough for autonomous artificial intelligence while basic parts of that same internet are still unnecessarily difficult for human beings with disabilities.

I think we should talk about that.

Accessibility should not require artificial intelligence to rescue bad design

AI has enormous potential for accessibility. I use technology every day that would have sounded impossible not that long ago.

A camera can describe something I cannot see. Software can recognize text. A phone can speak almost everything on the screen. Artificial intelligence can increasingly interpret images that were never given useful descriptions.

I welcome that progress.

But there is a dangerous temptation hiding inside it: the idea that AI can simply repair inaccessible design after the fact.

A developer leaves a button unlabeled? Maybe AI will figure out what it does.

A company publishes an image without alternative text? Maybe AI can describe it.

A website creates a confusing form? Maybe an intelligent assistant can navigate around the problem.

That is clever technology.

It is also backwards.

Disabled people should not need a layer of artificial intelligence to translate a poorly built website into something usable.

The website should have been usable in the first place.

There is a difference between innovation and cleanup

Technology companies love the word innovation. I understand why. I am a creator myself. Building something new is exciting.

But sometimes the most important innovation is not flashy.

Sometimes it is a properly labeled button.

Sometimes it is a heading structure that makes sense.

Sometimes it is a form field that tells a screen reader what information belongs there.

Sometimes it is a video with captions.

Sometimes it is making sure keyboard focus does not disappear halfway through a page.

None of those things will headline a major technology conference.

They can still determine whether somebody can independently apply for a job, read an article, buy a ticket, manage an account or participate in a community.

That is not a minor technical distinction.

It is the difference between being invited into the digital world and being forced to stand outside it.

I do not want a separate internet for blind people

One of the things I dislike about the way accessibility is sometimes discussed is that it makes disabled people sound like some tiny category of users living off to the side.

I do not want a special internet.

I want the internet.

The same one everybody else gets.

I want the complicated websites, the creative websites, the social networks, the music platforms, the new apps and whatever strange thing somebody invents next year.

Accessibility should not mean stripping away personality until everything becomes a blank page with enormous buttons.

Good accessibility gives people more ways to interact with the same experience.

That distinction matters to me as a creator.

I make music. I build digital projects. I care about presentation. I care about design. I also rely on assistive technology.

Those things are not in conflict.

The best technology understands that.

AI could become one of the greatest accessibility tools ever built

There is another side to this story, and it is why I am optimistic.

For blind people in particular, artificial intelligence can shrink the distance between having information physically in front of you and actually being able to understand it.

Think about an ordinary situation.

A printed document arrives.

There is something unusual in a photograph.

A control on a device is not labeled.

A website contains an image nobody described.

A sign appears somewhere you have never been.

Historically, situations like these could require another person.

Increasingly, technology can fill in some of those gaps.

That is independence.

Not perfect independence. Not magical independence. AI makes mistakes, and anybody telling you otherwise is selling something.

But it is real progress.

And that is exactly why developers should take accessibility seriously now rather than later.

The more powerful these systems become, the more dangerous it will be to repeat the old pattern of inventing something first and asking who can actually use it afterward.

The next digital divide may not be about having internet access

For years, people talked about the “digital divide” primarily in terms of who had access to computers and broadband.

That divide still matters.

But another one is emerging.

It is the divide between people who can fully participate in increasingly automated digital systems and people who cannot.

Imagine a world in which an AI assistant can schedule a doctor’s appointment, purchase a plane ticket, compare services and complete complicated online forms.

Now imagine that the systems underneath those tasks were never designed with disabled users in mind.

Technology might technically become more powerful while independence becomes unevenly distributed.

That would be a spectacular failure of imagination.

Disabled people should be in the room before the product launches

There is a simple way to avoid a lot of these problems.

Include disabled people earlier.

Not during the final accessibility audit.

Not after somebody complains publicly.

Not when a lawyer sends a letter.

During the design.

During development.

During testing.

Ask a blind person to use the product with a screen reader.

Ask somebody who navigates by keyboard to complete the important workflows.

Ask deaf and hard-of-hearing users whether the media experience actually communicates what it is supposed to communicate.

Listen when something does not work.

The people experiencing the barrier usually do not need a committee to explain that the barrier exists.

They need somebody empowered to fix it.

Accessibility is not charity

I want to be very clear about this because language matters.

Making technology accessible is not an act of kindness toward disabled people.

We are customers.

We are creators.

We are employees.

We are students.

We are business owners.

We are developers.

We are musicians.

We are parents.

We are travelers.

We are people trying to live ordinary lives in a world where an extraordinary amount of ordinary life now happens through a screen.

Access to that world should not depend on whether somebody remembered us at the end of a meeting.

The future should require fewer workarounds

Blind people become very good at workarounds.

You memorize which button is actually the unlabeled button you need.

You switch apps.

You find a different website.

You learn strange sequences that make an inaccessible system barely usable.

You ask somebody else to do something that you should have been able to do yourself.

You keep moving.

There is resilience in that.

But resilience should not become an excuse for bad design.

The goal of technology should not be to produce increasingly impressive obstacles and then congratulate disabled people for figuring out how to climb over them.

The goal should be fewer obstacles.

That is where artificial intelligence genuinely excites me.

Not because I want AI to replace human judgment.

Not because every problem needs a chatbot.

And definitely not because I believe newer automatically means better.

I am excited because, used responsibly, this technology can give people access to information that previously depended on somebody else’s eyes, ears or assistance.

That is powerful.

But only if accessibility is treated as part of the architecture of the future rather than a repair job for the mistakes of the past.

We are building machines that can describe photographs, understand language and increasingly move through the digital world on our behalf.

Surely we are capable of building a button that tells a blind person what it does.

The internet is becoming intelligent. Now it needs to become more human.