stumbled upon @ https://aiexperiments.withgoogle.com/autodraw
We draw familiar shapes, doodling it’s called. And not everyone is a Leonardo, but human perception matches childish simplifications with a “well known” real life object.
I guess machine learning is not so different than human learning finally, as without familiarity and a “difference machine” (human brain) with a stock of similar images we could not recognize any particular drawing either.
So in a way, machine learning uses the same techniques as children and through exposure can eventually build up the “experience” of such recognition.
Don’t forget to smile to a robot with API access to Google Cloud Vision: