Cripping AI: Reimagining AI Through Lived Disability Experiences
Drawing on crip theory, this paper proposes "cripping AI" as a transformative framework that moves beyond mere accessibility to dismantle ableist assumptions, center disabled epistemologies and labor, and reorient AI research and development through the lived experiences of diverse disabled communities.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Idea: Fixing the Blueprint, Not Just the Door
Imagine the world of Artificial Intelligence (AI) is like a massive, high-tech building. For years, the people building it have been told to make it "accessible." They've added ramps, wide doors, and Braille signs. This is good, but the paper argues it's not enough.
The authors say the building itself was designed with a hidden assumption: that everyone inside is "able-bodied" (perfectly healthy, fast, and standard). When a person with a disability enters, the building treats them as a "problem" that needs to be fixed so they can fit the building's rules.
This paper proposes a new approach called "Cripping AI." Instead of just adding a ramp to an old building, they want to redesign the entire blueprint based on how disabled people actually live, think, and move. They use the word "crip" (a reclaimed term from disability culture) as a verb, meaning to actively challenge the idea of what is "normal."
The paper offers three main rules for this redesign, which they call "tenets."
The Three Rules of "Cripping AI"
1. Expose the Politics (Stop Pretending There is Only One "Normal" Way)
The Analogy: Imagine a dance floor where the music only plays one specific rhythm. If you can't dance to that exact beat, the DJ tells you that you are the problem, not the music.
The Paper's Claim: AI systems often assume there is one "ideal" human way to function (like speaking perfectly fluently or seeing clearly). If you don't fit that mold, the AI labels you as "disabled" and tries to "fix" you.
What They Want: Stop trying to force disabled people to dance to the "normal" beat. Instead, recognize that "disability" is often just a mismatch between a person and their environment. AI should celebrate different ways of moving and thinking, rather than trying to erase them.
2. Honor "Cripistemologies" (Listen to the Experts Who Live It)
The Analogy: Imagine a map of a city. For years, the map was drawn by people who have never walked the streets, only looked at them from a helicopter. They drew straight lines where there are actually winding paths. Now, the people who actually walk the streets are saying, "That map is wrong; we know the shortcuts and the hidden alleys."
The Paper's Claim: Currently, AI is often trained by sighted people describing what blind people see, or hearing people describing what deaf people hear. The paper calls this "able-bodied knowledge" being treated as the "truth."
What They Want: The "truth" should come from the disabled people themselves.
- Example: A blind person doesn't just need a description of a picture; they need to know what matters to them (like where the curb is, not the color of the sky).
- Example: A person who stutters isn't just "broken speech"; their pauses and repetitions carry meaning and emotion that AI should learn to understand, not delete.
3. Respect "Crip Labor" (Stop Treating Disabled People as Passive Recipients)
The Analogy: Imagine you are invited to a party, but you have to bring your own chair, your own food, and your own music, and then you have to clean up the mess afterward. The hosts say, "We gave you a party!"
The Paper's Claim: Making things accessible is hard work. Disabled people often have to do extra "invisible labor" to make technology work for them (like constantly correcting a voice assistant or re-typing a message because the captions failed).
What They Want: Stop treating accessibility as a one-time fix. Acknowledge that disabled people are active partners who do the heavy lifting to make systems work. AI should be designed to reduce this extra work, not add to it.
Real-World Examples from the Paper
The authors tested these three rules on three specific groups to show how it works in practice:
1. Deafness and Sign Language AI
- The Old Way: Trying to turn Sign Language into written English as fast as possible, often ignoring facial expressions and body movements that are crucial parts of the language. It's like trying to translate a symphony into a single sentence.
- The "Cripping" Way: Recognizing Sign Language as a complete, complex culture. Instead of just translating it to English, build AI that understands the unique grammar of sign, supports the diverse ways deaf people communicate with each other, and respects that deaf people are the experts on their own language.
2. Blindness and Visual Assistive AI
- The Old Way: Sighted people describe a photo to a blind user, assuming the sighted person's description is the "correct" one. It's like a sighted person describing a painting to a blind person and saying, "This is what it looks like," without asking what the blind person actually needs to know.
- The "Cripping" Way: Blind people have their own unique ways of understanding the world (using sound, touch, and memory). AI should help them piece together information from many sources, rather than forcing them to rely on a single, possibly biased, description from a sighted person.
3. Stuttering and Speech AI
- The Old Way: Speech AI tries to make stuttering "disappear." It cuts out the pauses or tries to smooth them over, treating the stutter as a mistake to be corrected.
- The "Cripping" Way: Recognize that stuttering is a valid way of speaking. The pauses and repetitions are part of the speaker's identity and can build trust and connection. AI should be designed to understand and respect these patterns, not silence them.
The Conclusion: A New Kind of Team
The paper ends by saying that we can't just add more data or ask for feedback from disabled people at the very end of the process. We need to change the whole team.
- Don't just ask: "How do we fix this for you?"
- Do ask: "How do you already solve this? How can we build a system that works with your way of living?"
"Cripping AI" is about realizing that the world is full of different kinds of minds and bodies. Instead of trying to force everyone into a "standard" box, we should build AI that is flexible enough to fit the whole, messy, beautiful diversity of human experience.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.