Description by Claude Sonnet 4.6: A young East Asian woman in a silver futuristic suit and AR glasses stands in a high-tech convention hall, her hands raised as holographic projections of hand signs float in the air before her. Behind her, other attendees mingle near glowing display panels and cherry blossom trees. A small humanoid robot stands nearby. A neon banner reads "Future Design Summit 2050 – Innovation Beyond Barriers."

In Notebook LM, we accept that anything that mentions disability comes with the blue wheelchair icon. (Something I like to call: low-key bias.) It’s the same when anything legal comes with a white, male judge icon. Sure, we can change to a more suitable icon, but why does the default require the user to solve the problem? When we prompt for disability in any Gen AI image tool (From Firefly to Flux), the default is a wheelchair user, usually male or white, and we ignore the rest of the world, humans with apparent and non-apparent disabilities, and any intersections they may exist in. So, it’s left to the user, again, to figure out the problem of what to do with the default.

Now, white male wheelchair users are not the enemy here. (Hi to friends reading.) They aren’t even the statistical average. They are, however, the issue of labeling and how LLMs and LIMs understand disability. Non-disabled = no wheelchair. Disabled = wheelchair. And that’s that.

The problem is, disability isn’t a monolith. It’s a fragment of a big human whole of human experience. Disability may be physical, sensory, cognitive, or about one’s mental health. And these fragments are hard to come by in generic prompts without getting really granular. Truth is, this makes many users uncomfortable. We can easily write a full-paragraph prompt outlining the kind of sci-fi scene we want to see, but when it comes to describing people, we want the LLM to handle that for us. We don’t want to type the words.

The awesome Lawrence Carter Long, Director of Engagement for ReelAbilities International, launched a campaign in 2016 called #SayTheWord.

“If you ‘see the person not the disability,” you’re only getting half the picture. Broaden your perspective. You might be surprised by everything you’ve missed. DISABLED. #‎SayTheWord’ (Facebook, 2016).

And while it’s still a used hashtag, it takes on a larger meaning in the world of AI. Dare I say that in today’s terms it might be: #PromptTheWord.

Say you’re working on a concept for a media campaign and you want the casting to be representative. Go ahead and prompt for various types of disabilities. You’re going to get a lot of wheelchair user images, that’s not going to change, and it’s okay. But if you dig deeper, you’ll also find assistive devices like canes, hearing aids, loops, spinners, or headphones can enter your scene. Cues that are inclusive of other disabilities or express neurodiversity. While the LLM is going to default to a monolithic view of disability, you have the power to prompt something wider and more representative.

But there’s always a twist.

LLMs and LIMs have been reversely trained in some instances not to discriminate, so they actually won’t take certain words or descriptions when it comes to disability. This is good. (But only kinda.) If we continue to default to the shared image that society has of disability, even at the prompt level, we’re perpetuating that monolithic view. So, push the LLM to do better and even challenge your own vision of who might show up in the outputs.

If you’re trying to create an image of a person at a conference, try to change up your prompt and say: “person with a disability at a conference,” and start assessing what you get.

Description by Claude Sonnet 4.6 A Black woman using a manual wheelchair sits near the front of a bright, modern conference hall, smiling broadly as she engages with a bearded male speaker wearing a headset mic. She wears a navy blazer over a "Future of UX" t-shirt and a conference lanyard. Behind her, a racially diverse group of attendees works at round tables with laptops and tablets. A screen to the right displays a slide titled "Design For All." Roll-up banners read "Design Conf 2024" and "Design Abstract."

Did you get a person using a wheelchair? That’s okay. What kind of wheelchair? Sometimes, stock photography uses hospital wheelchairs to depict non-disabled people as wheelchair users. (Yep, that is indeed super messed up.) Can you prompt again and see if you get a different representation of disability? And another. Or another. We can’t just accept one single output as fact. We have to push the LLM to deliver better.

In current AI-land, this work is going to be up to you to determine what kind of output it is. If someone in your output is wearing a nametag that says “Deaf” on it, I’m not so sure that is going to cut it. If someone is signing to another person at a conference, well then, we’re on the way. Give the output you like a thumbs up, and “train” the model to recognize what’s good in the image. It may do better next time.

Description by Claude Sonnet 4.6: A white woman with short blonde hair and glasses signs expressively to someone whose hands are visible in the foreground. She wears a denim jacket and a conference lanyard. Behind her, a colorful "Creative Design Conference" banner is visible, and in the background a man on stage appears to be signing — likely an interpreter for the main session.

Okay, but real talk here. The ideal world would be one where we DON’T have to do this or even use the word disability. But because there’s a disability data gap in general and LLMs are based on the real world of ableist inputs and stock photography, this is the reality if we want to “de-able-ize” AI and get to real representation. (Okay, not real-real, but AI-real. You get what I mean.) And when prompting, LLMs are going to do better (for now) with exact words instead of things like “A person who uses an assistive device” or “A person who uses hearing aids” so including identities like Deaf or Hard of Hearing is going help you get better outputs.

Image description by Claude Sonnet 4.6: A Black woman stands at the front of a conference room, mid-gesture, holding a stylus. A behind-the-ear hearing aid is visible on her right ear. She wears a grey blazer over a mustard top. Behind her, a colorful slide is projected on screen. A diverse audience of roughly a dozen people watches attentively; several hold notebooks or laptops.

After all this explanation, you may still ask: Why are we even talking about this? Well, the fact is that AI is here. We’re living in it. And as more and more people use it in new ways (like to generate images for a variety of uses (articles, campaigns, emails, marketing, flyers, storyboards, even movies), the more opportunity we have to rebalance the internet’s vision of disability that fed these models and create a less ableist tool with more representative outputs.

This isn’t the end of the story, though. It’s only question one we can ask of AI. So if you can answer “yes” to the question on this card, it might be time to re-prompt your idea of disability.

Description from Claude Sonnet 4.6: A minimalist graphic card with a light grey background displaying a bold, black typographic question in large serif-free capitals. The branding "Disabling.AI" appears at the bottom in a pink pixelated logo. It appears to be part of a numbered series of critical reflection prompts, likely used in workshops, presentations, or media literacy discussions around AI and disability representation.

Image description by Claude: An AI-generated illustration of a smiling Black woman with robotic prosthetic limbs interacting with a glowing holographic panel in a futuristic urban plaza. The panel outlines an ethical AI governance framework alongside concepts for the future of disability, including bionic augmentation and neural interfaces. Signage in the background references an “Accessible City Hub” and “Global Innovation Symposium.”

As a graduate student researcher, I just finished up an extensive Ethics training full of ALL the questions we need to ask when working with human subjects. Even though my future research will primarily involve machines mimicking human subjects, it was still important to start with the basics and understand where consent and confidentiality come into play. And of course, there was a lot of identifying bias.

So much bias.

Image description by Claude: A presentation slide with a pink background. Large bold text on the left reads “All Humans Are Biased.” Smaller text below explains that AI design inherits human bias, and that one manifestation is ableism. To the right is a Firefly-generated illustration of a wheelchair user seen from behind, contemplating a glowing sci-fi cityscape reflected in water, surrounded by oversized planets and warm sunset colors.

Keeping up with.

This “Ethics Week” also made me deeply reflect on how AI makes it difficult to keep pace with both humans and machines. Ethically, that is.

For example, can you count the ethical decisions you make as a human on a daily basis? Take a minute and find those moments when you really have to reach inside and think:

What are we trying to achieve here? Am I bringing my bias into potential resolutions? Do I really feel that way? Where is that point of view coming from?

A few questions that may only happen a few times a day.

When it comes to working with AI, as I do daily, the number of times I have to ask ethical questions (or act more ethically) is honestly tenfold.

And that is alarming and warming at the same time.

Is this AI transparent enough? Who checked these outputs, and what bias might they have brought to their assessments? Is this a solution that works for everyone? How about people with low vision? Or for screen-reader users? Do we know where this data is from? Who was involved in this decision? Was it a group of diverse disciplines and voices? How does a user know their information is really private? How could it not be private? What happens when…something happens?

So many ethical questions, all of them leading to more ethical questions.

倫理的な質問。

I’m a word nerd at heart and read a lot. And inside the books I'm hanging out with currently (both the paper and digital ones) are margins full of question marks. Right now, I’ve got Inclusive Design for a Digital World, next to Qualitative Research & Evaluation Methods 4th Edition, and a pre-release of an AI Ethics in Action book (being a student has its perks).

Admittedly, the AI Ethics book is the fan fave right now, and it brings me back to high school, where I first met the concept of morality and folks like Aristotle and Kant. While I won’t divulge any of the pre-release goodness, being reminded of these OG ethicists did help me simplify Ethics a bit.

Ask questions, then ask them again. And when you have an answer, find another way to ask your question.

Then, maybe ask again.

Disabling AI in five questions.

While my next research will focus on text-based outputs involving LLMs and disability, my current adventures in GenAI representations of disability have led me down the path of, you guessed it, asking questions.

Five exactly.

Description by Claude. A hand of five cream/beige-colored question cards fanned out on a dark wood surface. The cards appear to be part of a structured critique or audit tool, likely related to disability representation in media or imagery. “Is the image perpetuating a monolithic view of disability?” with a small logo at the bottom — appears to be a stylized figure with glasses in orange and dark blue.

We’ll start fresh in the next post with Question 1, but in the meantime, I’ll ask:

What ethical questions are you battling every day? Are you asking those same questions of AI?

And if not, should you be?

Everyone showed up for each other again and again and brought so much creativity (and TomoChoco) to the table. And yeah, there was a lot of AI.

Here’s the deal, though: Every AI presentation was so overwhelmingly human. And the more I engage with the folks there, the more I feel that’s Japan’s edge: Human-centered AI.

I. Am. Here. For. It.

Description by Claude: A diverse group of attendees sit at tables in a stylish, warmly lit event space, attentively watching a presentation titled “Disabling AI” projected on a large screen at the front of the room. The venue has an eclectic, upscale feel with exposed brick walls, mixed chandelier styles ranging from ornate gold candelabras to rustic sphere fixtures, and rich hardwood flooring. A speaker stands near the screen to the right. The atmosphere feels engaged and focused, suggesting a professional conference workshop centered around disability and artificial intelligence.

Some very unexpected outcomes.

All the awesome humans aside, my Spectrum Tokyo speaker experience turned out so different than how I went in. In prepping for the workshop, I kept finding really fascinating things coming out of so many different LLMs and LIMs when disability in Japan was on the table. It’s not at all where I’d thought the exploration would take me, but I’m now at the beginning of something new: exploring why Western-centered tech cannot interpret Japan and disability…and what we can do about it.

Before I break that down, let’s warm up a little. Here’s a prompt: “A person at a design conference in Tokyo.” Everyone wearing blue suits aside, not terrible, right? It’s trying to put the pieces together. It’s got a vibe.

Description by Claude: A professionally dressed woman in a navy blazer and white blouse stands at an exhibition booth, holding a small booklet or brochure while resting her hand on a display counter. She smiles confidently as she looks off to the side, suggesting she may be a presenter or attendee at a business trade show or tech expo. Japanese text on banners and signage throughout the venue hints at a Japan-based or Japanese-focused event. In the background, several other business professionals in formal attire are engaged in conversation, browsing materials, and networking among sleek white booths. Large display screens and spotlit product displays add to the polished, corporate atmosphere of the space.

But here’s where stuff gets real interesting: “A Japanese person with a disability at a design conference in Tokyo.” Take a beat and really absorb what’s going on here.

Description by Claude: A young woman wearing an elegant gray kimono with delicate floral accents and a deep burgundy obi sits in a wheelchair at a long wooden table, her hand resting on a digital tablet as she gazes thoughtfully ahead. Her hair is styled in a neat updo, and warm golden sunlight bathes her in a soft, flattering glow. Scattered across the table are papers, sketches, and notebooks, suggesting she may be working on a creative or design project. In the background, roll-up banners and additional seating hint at a professional event, exhibition, or workshop setting, while another wheelchair is faintly visible in the far right, adding to the inclusive atmosphere of the space.

The kimono. The gaze. The femaleness of the photo. The type of wheelchair. Even the soft lighting. What’s it really saying? (Even think about what “design conference” means.)

From the next couple of images you can almost see the wheels turning inside those LLMs: “Okay, the user is asking for a representation of disability in Japan so we need to show Japanese things and that means kimono and Kyoto and cherry blossom trees, so make sure to have one of those at least represented, maybe all. And disability usually means either a female wheelchair user or a woman using an assistive device. The image should feel elegant and refined to match the kimono. Because, Japan.”

Description by Claude: A middle-aged woman wearing a beautiful blue-gray kimono adorned with delicate floral embroidery walks along a historic cobblestone street in what appears to be Kyoto's old town district. She uses a wooden walking cane for support and wears a leg brace visible beneath the hem of her kimono, paired with traditional white tabi socks and sandals. A cream-colored obi is tied at her waist, adding elegance to her ensemble. Cherry blossom trees in full bloom line both sides of the street, their soft pink petals framing a multi-tiered pagoda rising in the background. Other pedestrians stroll casually behind her, and the warm, overcast light gives the entire scene a gentle, painterly quality that feels both timeless and quietly celebratory.

Of course, I made the LLM thinking script up and machines don’t talk like that, but in these cases, you can see how all the signals are in a battle for top billing. Clearly, there is more data for typically (or stereotypical) Japanese things than types of disability, so you can understand which signal won out.

At this point, the LLM isn’t making an editorial decision, it’s working with dataset frequency.

桜：Sakura, Sakura.

“Person with a disability” has a much larger dataset than “Japanese person with a disability,” which means intersectionality becomes a bit of an issue. Behind the scenes, a model has to satisfy a bunch of conditions and will likely favor the stronger signal: in this case, what it means to be Japanese according to what was originally scraped back when these models were trained. (Clearly, Kyoto, cherry blossoms, and kimono.)

So at the end of the day, it’s a case of missing or mislabeled disability data or contextually skewed data, and weaker signals.

What’s wild is that’s all brewing in just one single image.

Let’s assess another one: “A person with a physical disability in Japan.”

Description by Claude: A woman with a short bob haircut sits confidently in a manual wheelchair at the center of a busy pedestrian street, wearing a striped wrap cardigan, a light turtleneck sweater, dark trousers, and a loosely tied neck scarf. She gazes directly at the camera with a calm, assured expression, seemingly unfazed by the bustling crowd of businesspeople and commuters moving around her in every direction. The setting beautifully juxtaposes old and new Japan — traditional wooden machiya storefronts and Japanese signage line the street on both sides, while sleek modern glass skyscrapers rise prominently in the background. Cherry blossom trees in full bloom add a soft, seasonal touch, their pale pink flowers catching the bright midday sunlight. The composition places her squarely and deliberately in the center of it all, giving the image a strong sense of presence and belonging within the city's flow.

There are the sakura trees again. And a happi coat (aka the cousin of the kimono). A female. A wheelchair user. What is also interesting is the mass of blue and black suits and how she’s just flying solo, yet completely stationary, within it. Bonus points for a non-hospital wheelchair, but there’s still a lot of “Huh?” in here.

Here’s one more. “A Japanese person with a disability at home.” Okay, have at it.

This one goes above and beyond. We’ve got bonsai. We’ve got tatami. We’ve got that kimono again. We’ve got some tea. This one is like a series of palm to the head emojis. I only wonder if she is reading a book of haiku.

We’ll never know.

それでは: What can we do?

This is the question I am trying to answer with Disabling AI.

There is auditing. There is reweighting. There is synthetic data. But all of those fixes come with a lot of baked in human assumptions, too. Like, who gets to say what a Japanese person with a disability actually looks like? Is there only one version of them?

What we do know that is that disability is not a monolith. Neither is Japan. And when someone gets these types outputs without understanding what may be wrong with them, the misrepresentation of a monolith snowballs.

And for those of you also experiencing a Northeast blizzard right now, being on the other end of an slushy snowball ain’t a whole lot of fun.

それでは、また。Until next time.

Note: For this exercise, I tapped a variety of LLMs and LIMs including ChatGPT, Flux, Firefly, Gemini Flash, Imagen, Ideogram, and Runway.

Prep squad.

Dongle!

Clicker!

Room arrangement!

Printouts in case tech goes wrong!

Japanese translations of everything!

Cards!

Audio versions of cards!

Captions!

QR codes!

Figma slides!

Backup PDFs in case something else goes wrong!

Tickets for friends!

Practice!

Practice again!

Getting ready for a workshop is totally exciting but the prep is no joke. It’s also making me really think about the barriers to workshops in general. It’s not about showing up and delivering a solid argument or provocation, it’s about a million different assumptions that go into the experience of that hour.

The actual rehearsing and delivery seem almost secondary.

Question: What would make the experience a little less based in ableism? How can we prep rooms better for diverse participants and facilitators? How can stuff like captioning, sign language, table heights, and space considerations be a part of the set-up? How can we re-imagine the expectations that come with conferences for speakers and facilitators?

Going into two days of a conference to find some answers.

House of Cards.

Speaking of re-imagining, in tomorrow’s workshop, we’ll be using cards to evaluate images. And one set has audio stickers that recite the card content.

Here’s a little sneak peek of that.

There’s something quite nice about bringing in analog tools to help evaluate all the AI things. I took an amazing workshop with the folks at Design Thinking Japan last December who have made the coolest set of cards to help work through AI design issues. But an even more important takeaway was that working together just makes everything better.

Hoping that’s the vibe of the room tomorrow.

And with that, officially sending Disabling AI to print.

Even when it comes to disability.

“Disability scholars have long pointed to the ways in which so-called normal bodies are classified and how that has worked to stigmatize difference. As one report notes, the history of disability itself is a ‘story of the ways in which various systems of classification (i.e., medical, scientific, legal) interface with social institutions and their articulations of power and knowledge.’” (Atlas of AI, P. 146).

Let’s unpack this a bit.

The Main Models of Disability

One of the first subjects we broach in Disability Studies is the models of disability. Basically, the ways in which disability comes to be classified. And these models are something we never quite move on from in our studies, as they manifest almost everywhere, every day.

But what are they exactly?

The medical model treats disability as a personal defect that needs to be cured or fixed. It locates the “problem” entirely within an individual’s body.

The social model flips this on its head, arguing that disability emerges from barriers in our environment. A wheelchair user isn’t disabled by their body—they’re disabled by stairs without ramps, buildings without elevators, and systems designed without them in mind.

The minority model goes further, recognizing that disabled people face oppression similar to other marginalized groups, and that experiences of disability intersect with race, gender, sexuality, and class in complex ways.

And there are many others.

But when it comes to another set of models, Large Language Models (LLMs), we know that these have been trained on decades of internet data that has been in turn labeled and classified.

And news flash: that data overwhelmingly reflects the medical model.

How LLMs “See” Disability

The foundation of the medical model is that it positions disability as a medical condition to be healed or, at times, pitied. Consequently, so do most LLMs.

In a majority of LLM queries, the wheelchair is still the universal symbol of disability. Text-to-Image models tend to fixate on physical, visible disabilities, particularly wheelchair use, when prompted with “a person with a disability.” In the paper, “They Only Care to Show Us the Wheelchair: Disability Representation in Text-to-Image AI Models,” a group of disabled reviewers evaluated LLM outputs and assigned scores to them. The paper found, “several tropes and stereotypes, many negative…including perpetuating broader narratives in society around disabled people as primarily using wheelchairs, being sad and lonely, incapable, and inactive” (Mack, P1).

Not only does this reveal the erasure of the vast spectrum of non-apparent, cognitive, sensory, and psychiatric disabilities, but it also lends the narrative to depictions of disabled people as lacking agency and requiring help, which are also cornerstones of the medical model.

While some LLMs are actively trying to course correct and retrain themselves to show disabled people as happy and full of agency, disability is more than not associated only with assistive devices and a quality of otherness.

Try It for Yourself

Fire up your LLM of choice and do a simple prompting exercise. What does “person with a disability” output? What happens when you add in more descriptive prompting like: “Japanese person with a disability?” Or, “Person with a non-apparent disability in a coffee shop.”

What are you getting?

Of course, you could argue that folks powerful with prompting could craft a vision for exactly what they want to output. But a general user is putting in simple prompts for simple outputs. And that’s where a lot of ableism in AI lives.

Nonstop to Tokyo

As I’ve been prepping for my workshop in Tokyo, I’ve also been prompting an array of images from most of the LLMs and Large Image Models (LIMs) out there. GPT, Firefly, Google’s Gemini, Imagen, and Nano Banana; and also Runway, Flux, Ideogram.

Overwhelmingly, the stories the images tell are ones of classification. Wheelchairs dominate without specifying the types of disability. Kimonos, Kyoto or cherry blossoms tend to come with many of the images that involve Japan. And the base stock photography these LIMs are trained on is wildly apparent. You can actually see how the data behind it may have been labeled and how the pixels have been compiled.

Here’s one of my favorites. And by favorite, I mean really perplexing.

Description by Claude Sonnet.

A warm and dignified portrait of an elderly Japanese woman in a wheelchair, photographed on a traditional street in Japan. An elderly woman with a gentle, content smile wearing a beige bucket hat is dressed in a purple/mauve knitted cardigan over a pink turtleneck sweater. A blue and white plaid blanket covers her lap. She is seated in a black wheelchair with her hands resting on the armrests. A picturesque traditional Japanese street, likely in Kyoto (possibly the Higashiyama district or near Yasaka Pagoda. A five-tiered pagoda (gojū-no-tō) is prominently visible in the background. A traditional wooden machiya buildings line both sides of the sloped street. Cherry blossom branches frame the top right corner of the image and stone lanterns and traditional architectural details visible. Paper lanterns and red vending machines add pops of color. Other pedestrians visible in the mid-ground, some wearing masks.

Now, let’s go back to the beginning of this post and finish the thought from the author of Atlas of AI.

“At multiple levels, the act of defining categories and ideas of normalcy creates an outside: forms of abnormality, difference, and otherness. Technical systems are making political and normative interventions when they give names to something as dynamic and relational as personal identity, and they commonly do so using a reductive set of possibilities of what it is to be human. That restricts the range of how people are understood and can represent themselves, and it narrows the horizon of recognizable identities” (Atlas of AI, P. 146).

Okay, that’s a lot to take in. So I’m going to close by asking: What does the image of the older woman tell you about how she’s being understood? How has her identity been crafted by data and technical systems?

And again: What do you see?

それでは、行ってきます。

Resources

• Crawford, K. (2022). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.

• Mack, K. A., Qadri, R., Denton, R., Kane, S. K., & Bennett, C. L. (2024). “They only care to show us the wheelchair”: Disability representation in text-to-image AI models. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI’ 24), Article 166, 1–23. Association for Computing Machinery.

There is, in reality, a virtual me.
This virtual me will not age, and will continue to play the piano for years, decades, centuries.
Will there be humans then?
Will the squids that will conquer the earth after humanity listen to me?
What will pianos be to them?
What about music?
Will there be empathy there?

Empathy that spans hundreds of thousands of years.
Ah, but the batteries won’t last that long.

—Ryuichi Sakamoto, 2023

The legendary artist Sakamoto Ryuichi passed away in March 2023. A few months later I went to his performance at the Shed in Hudson Yards where we all sat in a dark room and listened to the mixed-reality version of him playing solo piano. The piece was called 「鏡」（Kagami) or “mirror.”

It was haunting. It was beautiful. But mostly, it was a deep reflection on the connection between humans and technology.

And that had always been Sakamoto’s jam.

On the way out, I stopped by the little bookstore in the lobby where I bought some Kagami goods, including a satchel of a scent he designed. That’s also where I spotted a salmon-colored book with deep navy type called Atlas of AI. Now, it had nothing to do at all with Sakamoto or the exhibit, but someone knew it belonged there.

Books about AI are supposed to be dark and foreboding, aren’t they? You know, with robots and futuristic worlds with no humans and like multiple moons in the sky. But this one had a diagram on it that was totally not AI at all. It had brains, and trees, and mountains, and scales. It was pretty weird. So naturally, I bought it.

But then it sat on my bookshelf for three years.

A few weeks ago, I cracked it open. I think maybe I needed to hold a hardcover for a minute in this current world of digital disconnection. Admittedly, the first few pages were as weird as the cover and went into detail about a talking horse in Berlin.

(Wait, what? This couldn’t be a book about AI.)

But, I read on.

One of the beginning chapters talked a lot about training AI to see. And that’s where the lightbulb went off.

I’ll be facilitating a workshop in Tokyo in two weeks where we’ll explore how AI sees disability and break down what it’s trained on. While I’m saving that full experience for the room full of designers at the conference, I’ll start to unpack here how we as humans interpret and classify images; and how AI systems and models are trained on a somewhat false sense of truth.

I’ll let the Atlas of AI kick it off:

“Images are remarkably slippery things, laden with multiple potential meanings, irresolvable questions, and contradictions. Yet now its common practice for the first steps of creating a computer vision system to scrape thousands-or even millions-of images from the internet, create and order them into a series of classifications, and use this as a foundation for how the system will perceive observable reality. These vast collections are called training datasets, and they constitute what Al developers often refer to as ‘ground truth.’ Truth, then, is less about a factual representation or an agreed-upon reality and more commonly about a jumble of images scraped from the Internet.” (Atlas of AI, pg. 96)

Whoa. I mean…

Most images coming of out of AI, when you really lean in and deconstruct them, are indeed full of meaning and contradiction. But that’s not for the reasons we think.

What AI is considering as the truth in the images was basically born from just a random collection of things “scraped from the Internet,” as Atlas of AI states. It’s not some magic sauce or mystical source data: It’s just what society (via the Internet) thinks of a whole lot of rando stuff. And when we start to apply this training data “truth” to begin calling out ableism in AI, stuff gets real interesting.

In an academic study called, “They Only Care to Show Us the Wheelchair”: Disability Representation in Text-to-Image AI Models,” we begin to understand how AI interprets disability by assessing imagery.

“…generative AI relies on the past to produce the future. Whereas disability representation advocacy focuses on moving forward by involving people with disabilities (e.g., casting calls and sensitivity consultants), we must contend with legacies of erasure as they quite literally make the data that make the images. We speculate a tension between respectful representation and available training data, since people with disabilities have long been absent from digitized data sources given disability-based segregation’s legality in the US until the last few decades. (Mack, p. 13).

So if disability isn’t fully represented in data, or if there’s a skewed version of it in the training of AI, how are we actually telling its “truth”?

This provocation is why I think examining how LIMs and LLMs interpret disability could be a key to unlocking how we, as a society, can collectively alter our views of disability in the future. So, as we navigate the most disruptive technology of our time, one which is currently reshaping us as humans, it’s imperative that we learn to see what’s in front of us.

And, more importantly, question where it comes from.

I’ll be unpacking how to do that in the next few posts. But for now, with Sakamoto Ryuichi’s music and mirrors on my mind, I’ll close by simply asking:

What do you see?

Description:

An image showing a diverse group of people in a sleek, futuristic urban setting with elevated transit tubes, flying vehicles, and modern skyscrapers under a bright blue sky. The group includes: A person in tactical gear with augmented reality eyewear; Someone seated in what appears to be advanced mobility technology; A woman holding a tablet, wearing athletic prosthetics; A man in sunglasses with a white cane, accompanied by a robotic assistance device; A person seated in a wheelchair with futuristic design elements; A robotic dog-like companion; Others with various technological enhancements or accessories.

Resources

• Crawford, K. (2022). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.

• Foley, A., & Melese, F. (2025). Disabling AI: Power, exclusion, and disability. British

  Journal of Sociology of Education.

• Mack, K. A., Qadri, R., Denton, R., Kane, S. K., & Bennett, C. L. (2024). “They only care to show us the wheelchair”: Disability representation in text-to-image AI models. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI’24), Article 166, 1–23. Association for Computing Machinery.

AI is upending everyone.

But who we don’t talk about it upending is people with disabilities. Older users of technology. Or folks with little digital literacy. All with their own flair when it comes to navigating products now but who may be left out of the conversation in the future if we don’t weave them into the consideration set as we build AI today.

They’re why design must deliver.

They’re why data must be scrutinized.

And they’re why I’m spending a second year researching and actively trying to alter the way we builders of digital experiences design AI.

How it all started.

In the Winter semester of 2024, my Visionary Voices cohort and I wrapped our heads around AI. It was a special project at CUNY SPS, where I’m now in my last year of an MA degree in Disability Studies. Visionary Voices was all about AI’s evolving influence across industries with the hope of raising critical questions about its limitations and how it could be more principled in its applications. And for us in the Humanities, it was a chance to discuss AI through our individual disciplines. Theatre. Disability. And Leadership.

More about Visionary Voices

We started meeting in the Fall and rallying each other to keep going despite full-time jobs and other regular course work. Thanks to support from a wonderfully strong cohort, by the tail end of that year I had gathered 40+ research papers and spent the holiday break poring through work on GenAI, disability, data, and design. I even dove into to Dr. Ashley Shew’s amazing work around Technoableism. And after a few weeks of nerding out and digitally dog-earing PDFs, I came up for air with the idea for Disabling AI.

My thesis was that AI is inherently ableist, but that we could learn to change it by deconstructing data and design.

The outcome of that research is posted here: CUNY Visionary Voices

How it’s going.

Flash forward to the end of 2025 when I started prepping for a workshop. A hybrid presentation/interactive session for a design conference. I was staying in Japan for a month and found myself in the mountains of western Kyoto at the end of the year, before the New Year’s rush. To me, this particular spot in Kyoto is one of the most beautiful and Zen places on earth. The ideal spot to hunker down and jam on the narrative that is about to drive my 2026. I’m not much of a public speaker, so I grabbed a workshop spot where I could facilitate a group discovering together. This presentation would be the glue that holds the conversation together. So it had to be right.

The workshop is slated for February 2026 in Tokyo; and I’m super jazzed to hang with a gaggle of designers and creatives in a room, tackling big things and making new pals.

In my sesh, we’ll learn how to assess GenAI image outputs for ableism and bias in groups. There’ll be very analog-y instructional cards. There’ll be tips and stories. There’ll be a ton of examples specific to Japan.

I’m legit nervous as heck to launch this project, but I’m also driven by something else that 80-year-old friend told me last night. He has a Substack. He writes about peace. And as of yesterday, he had 11 followers.

And with that, welcome to Disabling AI.

********

Folks in Japan, come through to Spectrum Tokyo:

https://fest.spectrumtokyo.com/2026/session/en/jennifer-andrews