What we tried

The Computational Art Studio runs entirely in the browser — no accounts, no uploads, a kid's photo never leaves the device. Inside it we built a family of small experiments that all share one move: the child controls the machine's perception, not just its output.

  • Photo → line art. Bring a photo, and turn it into a line drawing by tuning how edges are found. The threshold is a slider: push it one way and only the boldest contours survive; push it the other and every wrinkle becomes ink.
  • Draw Alive. Draw a creature, and the studio recognizes the sketch, guesses what it is, and animates the strokes — the fish you drew starts swimming the way it faces. Draw several and they share the canvas.
  • Seeing as carving. Segmentation and Voronoi tools slice an image into regions — stained glass, mosaic, a map of what the machine considered "one thing."

What surprised us

We expected kids to treat these as filters — press, giggle, next. What actually happened: they treated the machine's sight as a physical material. One kid worked the edge threshold back and forth like pencil pressure, hunting the exact point where a face stayed a face but stopped being a photograph. The interesting question in the room was never "what did it make?" It was "what does it see?" — asked while pointing at the screen.

The other surprise: recognition is more magical and more discussable than generation. When Draw Alive guesses your sketch, the follow-up is immediate and scientific — "what if I draw it backwards? without the tail? tiny?" That's a child probing a model's boundaries, unprompted, at age eight.

What children can learn from it

That perception is a set of choices, not a fact. A machine "sees" whatever its rules make visible — and so, in a quieter way, do we. A child who has personally moved the line between "edge" and "not edge" has a concrete handle on the most important idea in AI literacy: the output depends on what the system was told to notice.

Open questions

  • Where's the line between a helpful default and doing the seeing for the child? Every default we add is a choice they didn't make.
  • How do we get the "what does it see?" conversation to happen at home, without a teacher in the room?
Try it

Move the line yourself

Open the Computational Art Studio, feed the line-art tool one photo, and find your own threshold — the point where the picture says what you mean. The handbook explains what's happening underneath.

Open the Art Studio