AI literacy: what it actually means.

The marketing version is wrong.

Look at any kid-AI product page and you’ll see "AI literacy" used to mean one of three things: knowing how to talk to a chatbot, knowing that LLMs make mistakes, or knowing that AI is "everywhere now." Those are real things kids should learn. None of them are AI literacy in the way the research community uses the term.

The phrase comes from a line of work at Georgia Tech, MIT, and Stanford that’s tried to nail down what an "AI-literate" person actually knows and can do. The most-cited paper, "What is AI Literacy? Competencies and Design Considerations," from Duri Long and Brian Magerko in 2020, broke the construct into five competencies grouped under three higher categories.¹ The MIT Media Lab’s Lifelong Kindergarten and Personal Robots groups, plus Stefania Druga’s Cognimates project, have been refining the framework since.^2,3

The five competencies.

Paraphrased from the Long & Magerko taxonomy:

1. Recognizing AI. The learner can tell when AI is present in a system and when it isn’t. They notice the difference between a thermostat (no AI) and a recommender (AI). They notice when a chatbot is generating versus retrieving.
2. Understanding what intelligence is, and what AI is. The learner has a working model of how AI systems differ from human intelligence: what they can and can’t do, what they’re trained on, how generalization works.
3. Interdisciplinarity. The learner can see AI as a tool that touches every domain, not as a "tech topic." They can imagine AI in music, in storytelling, in social work, in cooking.
4. General AI / learning models. The learner has a working model of how AI systems learn from data. Training data, examples, generalization, mistakes.
5. Programmability and creating AI. The learner has experience giving an AI system instructions, watching how those instructions shape behavior, and iterating on the result.

Note what’s on this list and what isn’t. "Prompt engineering" isn’t there. "Knowing the names of LLMs" isn’t there. The list is a competency framework. What can the learner do, what mental models do they hold. Not a vocabulary checklist.

How the studio maps to the five.

We didn’t design the studio against this framework in 2024. We designed it against Papert’s constructionism and Resnick’s four Ps. When we read the Long & Magerko paper later, we noticed the studio fell out of the list more or less for free. That’s how good frameworks work. They describe what mature designs converge on.

Recognizing AI → the visible tool trace.

Every Inkie action is visible: reading icarus.html, proposing 1 change to player.update(). The kid sees AI working. They notice when it stops working. They develop a felt sense for "AI is doing something here" versus "I’m doing something here."

Understanding intelligence → the AI-mistakes loop.

The first time Inkie proposes a change that doesn’t make sense, references a function that doesn’t exist, or names a wrong file, the kid notices. The "find a mistake" transparency quest turns that noticing into a habit. Kids develop a working model of what AI is wrong about most often, and why.

Interdisciplinarity → the four tracks.

The same AI partner ships across Game, Story, Site, and Movie. The same vocabulary works in each medium: wizard, propose, change, decide. The kid’s mental model becomes "AI is a creative partner," not "AI is a tech topic."

Learning models → the variation quest.

The "Same prompt, two answers" transparency quest asks the kid to run the same wizard twice and compare. They see the AI propose different things on identical prompts. They learn that AI is probabilistic. That’s an important step toward understanding training data, sampling, and generalization without needing the math.

Creating AI → custom instructions.

By age 11 or 12, kids on the studio are writing markdown files that hold project instructions Inkie follows every session: GAMEPLAN.md in the Game track, STORYBOOK.md in Story. They’re shaping an AI’s behavior through written specification. That’s the fifth competency, made concrete.

The research hasn’t caught up yet.

Honest caveat: AI literacy research is six years old. Long & Magerko’s competency framework is widely cited and increasingly influential, but it isn’t the definitive last word. Other groups have proposed adjacent frameworks: the AI4K12 initiative’s "Five Big Ideas in AI" at the K-12 level, UNESCO’s AI Competency Framework for Students, Stanford’s ongoing work on AI fluency.^4,5

What the field is converging on is that AI literacy is plural. Recognizing AI, understanding what it is, watching it work, iterating with it, shaping it. The studio takes that plural seriously. We don’t teach prompt engineering as a skill; we surface the five competencies as side effects of making real projects.

If your kid finishes a couple of Tell and Show projects, what they will have done, without anyone teaching them, is develop a working understanding of all five. The competency framework predicts they’ll be better positioned than peers who took an "AI for kids" course. Whether that holds up in longitudinal studies is something the research community will need a decade to answer. We’ll keep updating the curriculum as the evidence comes in.