This object demonstrates how machines are seeing archetypes in order to recognize what it sees. The archetypal object is made by combining several objects into one and can be recognized as a desk lamp by an actual recognition system (used by Tesla and Facebook among others).

When a machine sees an object, the recognition system will first reduce the noise in order to see the essence of the object. When the essence matches with an archetype in its trained database, it can recognize it. But when the essence is distorted or trained with the wrong data, the object can be misrecognized.


OR 1


Object with lights on


Eventhough the two blurred images are almost identical, the recognition system highly believes that one of them is a desk lamp and the other one a ping-pong ball.