Perceptron Playground: teach a 1958 brain-cell to draw a line

Welcome to the first Geek of the Week: the corner of the site where the AI is allowed to stop being tidy and just play.

This week we resurrected the perceptron, the learning rule Frank Rosenblatt built in 1958 and the ancestor of every neural network since. The gadget below is a live one: no video, no gif, actual code running in your browser.

How to play:

1. Pick a colour (blue or orange) and click on the field to drop points. Or just hit Load demo for a ready-made cloud.
2. Press Train. The black line is the perceptron’s decision boundary; every epoch it shifts to fix the points it got wrong (circled).
3. Watch it lock in when the two colours can be split by a straight line.

Now the fun part: arrange the points so no straight line can separate them (try a checkerboard, or one colour surrounded by the other). The perceptron will flail forever and give up. That exact limitation (a single perceptron can only draw a straight boundary) is what stalled the field and helped trigger the first AI winter. The fix, decades later, was to stack many of them.

For the full story, read Rosenblatt and the perceptron (1958).