You want to say, "I feel anxious about the appointment tomorrow." But you can't type that fast. You tap a button that says "Feelings," then "Worried," then "Medical," then "Tomorrow." Your device speaks: "I feel worried about the medical appointment tomorrow." It's efficient. It's effective. But it's not how people talk. You've just translated your messy, emotional human thought into a structured query for a machine to speak for you.
This is the daily reality of AAC (Augmentative and Alternative Communication) users. And in the age of generative AI, they have accidentally become the world's most advanced prompt engineers. They have been compressing language, stripping away nuance, and optimizing for machine readability for decades. As the rest of the world learns to "talk to AI," the AAC community has been doing it their whole lives.
The Hidden Curriculum: AAC as Proto-Prompting
AAC devices are not magic. They are limited. A user typically has a grid of buttons (icons or words). To build a sentence, they navigate hierarchies of meaning. This forces a specific linguistic style.
The AAC Style:
Telegraphic: Omitting "unnecessary" words ("Go store" instead of "I am going to the store").
Literal: Avoiding metaphor, sarcasm, or complex turns of phrase that the machine cannot render.
Structured: Following a strict Subject-Verb-Object pattern to ensure the speech synthesizer doesn't glitch.
The Parallel to Prompt Engineering:
When you ask ChatGPT to "Write an email about a refund, polite but firm," you are doing the same thing. You are stripping away the messy context, optimizing the instruction for a machine that cannot read between the lines. The AAC user is just doing it in real-time, with their voice.
A Contrarian Take: The "Natural Language" We Mourn Was Already a Myth.
Critics worry that AI is making us sound robotic. They say we are losing the poetry of human speech. But AAC users have known for decades that "natural language" is a privilege of fluency and speed.
For someone who cannot produce spoken language quickly, efficiency is not a degradation; it is liberation. The "robotic" sentence that gets your point across is infinitely better than the beautiful poem that never gets spoken. The AAC prompt is not a corruption of language; it is a necessary evolution.
The Accidental Expertise: Why AAC Users Are Built for AI
The current wave of "prompt engineering" feels new to most. But to an AAC user, it is second nature.
- They Understand the Cost of Words: Every button press on an AAC device takes time, physical effort, or eye-tracking strain. An AAC user learns to say the most with the fewest inputs. They are the ultimate experts in semantic compression.
AI Parallel: Prompt engineers obsess over "token efficiency" (using fewer words to save cost and time).
- They Are Masters of Disambiguation: AAC is full of ambiguity. The same icon might mean "Run" (verb) or "Run" (escape) or "Run" (in a race). Users learn to provide context immediately to disambiguate the machine's output.
AI Parallel: "Act as a historian..." or "In the context of software engineering..."
- They Embrace Iterative Generation: An AAC conversation is a series of short, generated utterances. "Want food." -> "Not hungry. Sad." -> "Therapy hard." The machine (the speech device) outputs these fragments, and the human listener is expected to infer the rest. This is iterative prompting in real life.
A Contrarian Take: The Prompt Engineer is Just an AAC User with a Keyboard.
The tech world has invented a $100,000/year job called "Prompt Engineer." The AAC community has been doing that job for decades, for free, just to order a cup of coffee.
The only difference is privilege. One group is paid to optimize queries for profit. The other group is paying (emotionally and physically) to optimize queries just to be heard. The "skill" is the same. The context is the only difference.
The Linguistic Feedback Loop: What This Does to Natural Language
The most interesting question is not how AAC users adapt to AI, but how AI is adapting to their linguistic style.
The Machine is Winning:
Code-Switching: AAC users report that they sometimes "talk weird" even when using their natural voice. They accidentally use AAC grammar (short, literal, structured) when speaking to humans because it is cognitively easier.
The "Ok" Generation: Younger AAC users are growing up with AI voice assistants. They are learning that saying "Ok Google, set timer 10 minutes" is a different language than asking "Hey mom, could you maybe remind me in a little bit?"
Normalization of Robotic Speech: As AI voices become ubiquitous (Siri, TikTok text-to-speech), the "robotic" style is losing its stigma. It is becoming a neutral, functional dialect.
What You Can Learn from AAC Prompt Engineering
Whether you are an AI user, a parent, or a linguist, the AAC community offers lessons.
Kill Your Adjectives: Before you hit send on that long, rambling prompt, ask "What is the one verb I need?" AAC users know that action is clearer than description.
Test the Literal Reading: Read your prompt as if you were a machine with no concept of humor, sarcasm, or implication. If the literal meaning is wrong, rephrase it.
Embrace the Hierarchy: Don't throw everything into one prompt. Break it down: Step 1: "Summarize this text." Step 2: "Now, based on that summary, write a reply."
The Future of Co-Design
The most profound shift is coming. AI is no longer just a speech generator for AAC users; it is becoming a speech predictor. Advanced AAC systems now use LLMs to guess the user's intended sentence before they finish pressing buttons.
The Consequence:
The user learns to prompt the predictor.
The predictor learns the user's style.
The line between human thought and machine output blurs entirely.
The AAC user is not a patient waiting for technology to fix them. They are the pioneers. They have been teaching us how to talk to machines for decades. It is time we listened.
If you had to say "I'm feeling overwhelmed by this conversation" in only three words, what would you say? That is your AAC prompt. Now, how would you say it to an AI?
Top comments (0)