Why learn it by playing with it?
AI hands you fluent, confident answers with none of the working shown — so the first real skill is telling a genuine understanding from a plausible guess, in yourself and in the machine. Everything here is built to grow that skill, not to hand you conclusions. Here's the argument, from the ground up — and you're free to disagree with any of it.
Augment, don't replace
The phrase augmented intelligence comes from Douglas Engelbart, who argued back in 1962 that tools should raise what a human can do, not stand in for the human. That's the whole frame here. If a page simply told you what to think about AI, it would be doing the opposite of what it preaches — automating your conclusions. So the aim is to amplify your judgment and leave the deciding to you.
You understand what you can change
Prose can describe a moving system, but it can't move. Bret Victor spent a career on the alternative: show the data, give feedback the instant you act, and let people work the system with their own hands. Michael Nielsen's Reinventing Explanation makes the matching case — that interactive, visual media can explain things plain prose simply can't reach. You tend to keep what you touched, not what you were told.
Make the invisible visible
The cognitive scientist Judy Fan calls these things cognitive tools — number lines, diagrams, graphs, even augmented-reality overlays — human inventions that make the invisible visible, so each generation can build on the last. Her larger point is a loop: we discover useful abstractions about the world, then engineer with them to build new things — and the building makes something new visible to discover. That loop is the shape of this whole site: discover → explore → build → and round again.
Where does "augmented intelligence" come from, and what does it claim?
Douglas Engelbart (1962): tools should raise what a human can do, not replace the human. Amplify judgment; leave the deciding to you.
The craft is real, and written down
This isn't a hunch. A whole lineage made the interactive explanation a genuine craft — Nicky Case, Amit Patel's Red Blob Games, Dan Shiffman, Jack Schaedler, Vi Hart, William Ngan — and Case even codified the patterns (Puzzle It Out, Place Your Bets, Role Play, Sandbox Mode). It's only honest to add that the field is small and its makers are human: Case has written openly about burnout, and some catalogues haven't grown in years. We're standing on what they built, not pretending we invented it.
How to tell a real explanation
David Deutsch gives the test. A genuine explanation is hard to vary — change one part and it breaks. A weak one is easy to vary: it would have "explained" the opposite just as smoothly. When AI hands you an explanation, ask whether you could remove or swap a piece and still reach the same conclusion. If you could, don't trust it yet. That test is yours to apply — here, and everywhere else.
In Deutsch's test, what makes an explanation trustworthy?
It's hard to vary — change one part and it breaks. An easy-to-vary explanation would have "explained" the opposite just as smoothly, so it earns no trust.
And it has to stick
Understanding that has evaporated by next week isn't much use. Recalling something strengthens it far more than re-reading it (the testing effect); spacing that recall to land just before you'd forget resets the forgetting curve. That's why small recall prompts are woven through these pages — and why there's a page that turns your own recall data into a way of seeing how memory works, both yours and the machine's.
Why recall instead of re-read?
Pulling an idea back out of your head strengthens the memory more than reading it again (the testing effect). Spacing the recall fights forgetting.
See memory work — yours, and the machine's →
The cards you just answered became data. This page draws that data — your forgetting curve and your ladder — and sets it beside how an LLM "remembers", so you can see where the two are alike and where they part ways.
Now the foundations
The argument above is why we teach this way. These short reads are the what — AI from nothing, in plain words. There's no required order.