Three Canadian provinces released AI literacy frameworks this spring. I read all three. Each one uses the phrase “critical thinking” as if it names a single room everyone enters through the same door. Each one assumes the student arrives at the machine with no prior fluency in pattern recognition, no existing relationship to systems logic, no trained ability to spot situations that break the normal rules. Each one teaches from zero.
I sat in a classroom in Rotterdam in November 2022 as an observer, watching a fourteen-year-old autistic student debug a machine learning model. She found the bias in the training data in six minutes. Not because someone taught her to look for it. Because she had spent her entire life noticing when a system’s stated rules did not match its actual behaviour. Every school she’d attended had trained her in this — accidentally, by being a system that said one thing and did another. The teacher beside me whispered: she’s a natural. As if what she did was talent and not survival.
Here is the comparison. In Helsinki, a vocational school piloted an AI module in early 2024 built by Reaktor Education, a Finnish education technology company that develops AI curricula for schools. They structured it around what they called “algorithmic intuition” — the ability to sense where a system will fail without formal proof. Their facilitators told me, in a February 2024 call, that their strongest students were consistently the ones with ADHD or autism diagnoses. Not sometimes. Consistently. These students saw the floor plan of the logic before other students had finished reading the instructions.
In Ontario, the same semester, a province-wide curriculum rolled out AI literacy as a subset of media studies. The framework — I have the PDF — defines critical thinking as “the ability to evaluate claims using evidence-based reasoning.” One sentence. No mention of pattern recognition. No mention of systems thinking. No mention of the cognitive diversity already present in any classroom. The floor plan has one entrance and it faces one direction.
I am not making an argument about inclusion. I am making an argument about architecture.
Otto Neurath built the Isotype system in the 1930s — a visual language using pictures instead of words to make information accessible to people who could not read. It was meant to make knowledge universal. It failed, but not because the pictures were wrong. It failed because universality assumes a single viewer. The moment you design for “everyone,” you design for a specific cognitive norm and call it neutral. Neurath knew this by the end. He wrote to a colleague in 1944 that the system worked best when the viewer brought their own logic to the image. The image did not teach. The viewer taught themselves, using the image as material.
The AI literacy frameworks do the opposite. They assume the student has no existing logic. They teach from scratch. They build the room and then invite the student in.
But some students built the room already. They built it because they had to. A Deaf student reads spatial relationships before linguistic ones. An autistic student maps system behaviour before social behaviour. A dyslexic student sees the shape of an argument before its sequence. These are not deficits overcome. They are architectures of thought already constructed, already load-bearing, already doing the work the curriculum claims to introduce.
Ari Ne’eman, an autistic disability advocate and founder of the Autistic Self Advocacy Network, wrote in 2009 that autistic people are consistently described as needing to learn social cognition but never credited with the systems cognition they already possess. The framing is always deficit first. What you lack determines what we teach you. What you already know does not register because we did not put it there.
I find this funny in the way that is not actually funny. The same institutions that spent decades telling neurodivergent students their thinking was wrong are now designing curricula to teach neurotypical students to think the way neurodivergent students already do. Pattern recognition. Ability to spot situations that break the normal rules. Distrust of surface-level outputs. The ability to ask: what did this system leave out?
You could call this ironic. I call it a floor plan problem. The people who know the building best are not consulted on the blueprint because they entered through the side door. The side door is not on the official drawing. So their knowledge does not exist.
I label things on Tuesday mornings because the font was wrong. This is not rational but it is real: the wrong label makes the object harder to find in my mind. The wrong framework makes a student’s existing knowledge invisible to the curriculum. Both are information architecture failures. Both privilege the designer’s logic over the user’s.
The fourteen-year-old in Rotterdam finished debugging the model. The teacher logged it as “exceeded expectations.” Not as evidence that the curriculum’s starting point was wrong. The girl already had the room built. They just couldn’t read her floor plan.
This article was prompted by How should schools teach AI? 3 models to consider from The Conversation.