Animals, vehicles, icons, gadgets – all kind of stuff. We just asked people from our mailing list to draw different things for us and they did.

Our neural net, however, had neither a game-component (no track of success) nor a social component (no social buttons, no sharing ), so it never went viral. Instead of typing “ home” or “ menu”, you would doodle what you think of and the neural net offers you fitting icons. The whole idea was to let people search icons not by words, but by their doodles. It just wasn’t presented as a game, rather, as a visual search for icons.Īnd, just like Google’s, with every doodle people made for it, the neural net became better at recognizing things.

I won’t shatter the skies by saying that Google’s “Quick, Draw!” is not exactly a game, but a neural net that learns to recognize visual images. Our company, Icons8, developed exactly the same doodling game, only we didn’t present it as a game.Īnd before you accuse me of ripping off Google’s idea, I should mention it was released on our website 2 months before theirs. However, like in every good Hollywood movie, there is a dramatic coincidence lurking around the corner. This article could easily end here with an outburst of a missed opportunity. What if I’d like to know how many other people decided to draw a boomerang with a hangover? Or how many of them were angry? Or southpaws? Another way of getting answers to such questions could be simply to analyze hundreds of them, to reveal patterns. But not now.ġ5 drawings are enough for a good laugh or a boost of self-confidence, but not enough to see the bigger picture. What if Google could tell more? In future, maybe. On top of that what you get is whether your doodle was recognised or not, that’s all. Given the fact I drew it after a colorful Friday evening, that’s a compliment.

Analysing doodles