Beyond Screen Readers: The AI Revolution in Accessibility [May 1st Episode]

David Carlisle

Welcome to the show, everybody! I'm David Carlisle, here with Eric Marquette, Ruby Sturt, and Harper Bennett. And Eric Marquette, I want you to picture this: It's late 2023. A blind user puts on a pair of Ray-Ban Meta smart glasses, looks down at a countertop, and asks out loud, "What kind of tea is this?" And a voice INSTANTLY replies, "That's a box of Twinings Earl Grey, and it expires in December 2025."

Eric Marquette

Wait, including the expiration date? Just from a casual glance?

David Carlisle

Exactly. NOT scanning a barcode. Just the camera interpreting the visual field in real-time. We are officially past the era where accessibility tech is just a screen reader ROBOTICALLY reading left to right.

Harper Bennett

As an occupational therapist, I have to say, the leap from the old clunky barcode scanners we used to prescribe in the early 2010s to a pair of sunglasses reading expiration dates... it's wild. But the visual assistance isn't just about reading text anymore, right?

Eric Marquette

Right. And this brings us to apps like Seeing AI. Microsoft launched it back in 2017, but the recent updates use generative AI to describe scenes. It doesn't just say "chair." It says, "a wooden dining chair with a blue cushion, sitting three feet away, slightly angled to the left."

Ruby Sturt

"Slightly angled to the left." That's the part that ACTUALLY matters for navigation. Because if you're blind, knowing a chair exists is fine, but knowing exactly how to avoid tripping over its leg is the actual utility.

David Carlisle

And it's not just physical navigation. I run an after-school unity club, and we talk a lot about neuro—neurodiversity. See, ten years of teaching and I still trip over that word. Perfection is an illusion, folks. But for my neurodivergent students, the frontier is REAL-TIME social cue interpretation. There are experimental AI wearables right now that analyze conversational tone and facial expressions, and give users subtle haptic feedback if the person they're talking to is, say, getting frustrated.

Ruby Sturt

Haptic feedback for social cues? Does that actually reduce anxiety, or does it just add another layer of sensory input to manage? Because honestly, as much as I love tech... when I'm drafting accessibility policy for local councils, my anchor is still my mechanical typewriter. The clack-and-ding rhythm focuses me. If my typewriter suddenly started buzzing to tell me my tone was too aggressive, I'd throw it out the window!

Harper Bennett

I would pay money to watch you chuck an antique typewriter out a council building window, Ruby. But you hit on exactly the right tension. Does the tech reduce the cognitive load, or just shift it?

Harper Bennett

And that cognitive load question is huge when we talk about executive function. As a wheelchair user, navigating a physical space takes a lot of mental mapping... "Where's the curb cut? Is that elevator actually working?" But executive function tools are now doing that same kind of mapping for mental tasks.

David Carlisle

Yes! Have you guys looked at Goblin.tools?

Eric Marquette

The AI task breakdown app? I've seen it on TikTok, but what is the actual mechanism there?

David Carlisle

So, it uses natural language processing, essentially a tailored wrapper for an LLM. You type in something completely overwhelming, like "Clean the kitchen." And you hit the "Magic ToDo" button. It instantly generates a micro-checklist: 1. Put dirty plates in the sink. 2. Throw away visible trash. 3. Put away the clean dishes. And you can click a "spiciness" meter to break those sub-tasks down even further if you're completely paralyzed by ADHD or autistic burnout.

Ruby Sturt

The "spiciness" meter! That is brilliant. My sister Tash is severely autistic, and a command like "clean your room" means nothing—it's a wall of text. Breaking it down into "pick up the red socks" is how she actually processes the world. AI automating that breakdown could save caregivers literally HOURS of invisible labor every week.

Harper Bennett

And it's about keeping up with the user's pace. Years ago, I used to recommend Dragon NaturallySpeaking for voice control. You'd say "period," "new paragraph," and if you coughed... it typed the word "coffee." Now, we have AI like Be My Eyes integrating GPT-4. In 2023, they launched their "Virtual Volunteer." You take a picture of a map at a national park, and you can ask it, "Which trail is wheelchair accessible and under two miles?" It reads the map, processes the legend, and gives you the exact answer.

Eric Marquette

Wait—it reads the legend and synthesizes the CONSTRAINTS? That's not just OCR. That's contextual reasoning.

Harper Bennett

Exactly. It means I can spend less time cross-referencing three different PDFs, and more time actually doing the adaptive rock climbing I came there to do. I have a whole collection of mini carabiners on my keychain for every climb, and I prefer earning those to fighting with inaccessible maps.

Eric Marquette

Okay, but let's play devil's advocate for a second. At Jellypod, I see dozens of press releases a week for "revolutionary" AI accessibility tools. And a lot of it feels like tech-expo hype. They show a sleek video of a robotic exoskeleton, but the thing costs eighty thousand dollars and has a battery life of... twelve minutes.

Ruby Sturt

The "Hype vs. Help" divide. This is literally what keeps me up at night when I'm reviewing community strategies. Because in Australia, we have the NDIS—the National Disability Insurance Scheme. And their vetting framework is notoriously strict. If a piece of tech isn't proven to be "reasonable and necessary," it doesn't get funded.

David Carlisle

Which means if an AI tool costs a thousand dollars a month in subscription fees, it's effectively useless to the families who need it most. I have two disabled kids. If a tool isn't covered by a framework like the NDIS, or built into a school budget, it might as well NOT EXIST.

Ruby Sturt

Exactly, David. And that's why we cannot let AI replace the foundational standards. The WCAG—the Web Content Accessibility Guidelines—are the bedrock. Version 2.2 just rolled out in late 2023, adding specific criteria for cognitive accessibility, like ensuring authentication processes don't require you to solve complex puzzles just to log in.

Harper Bennett

Oh, you mean the CAPTCHAs where you have to identify the crosswalks in grainy photos? Those are my nemesis!

Ruby Sturt

Right! AI is great for building fancy new walls, but WCAG is the foundation. If a website isn't fundamentally coded with alt-text and proper heading structures, an AI screen reader is just guessing. Human-centric design has to come FIRST.

Eric Marquette

So the takeaway here isn't that AI is coming to magically fix an inaccessible world. It's that AI is a powerful crowbar, but we still need humans to decide which doors to pry open?

David Carlisle

I think that's exactly it. The tech is giving us new eyes and ears, but the empathy—the decision to actually fund and build for the margins—that still has to come from US. We'll leave it there for today. Thanks for listening, everyone.