A bit of state-of-the-art dystopian extrapolation.
I. A contrived, nonetheless virally very succesful meme generated upon the initiative of Australia’s Transport Accident Commission, as part of a state public awareness campaign. Embodied by a wonderful, original, ominous and mesmerizing sculpture by Patricia Piccinini. This is how an evolved human is supposed to look like, one that is biologically equipped to survive heavy accidents, especially car-crashes.
II. The AI, a machine perception-based autonomous vehicle, that’s coming into our futures from several directions at once, here represented by the most cutesy model of them all, from Google. This is the Hello Kitty of self-driving cars.
III. And last but foremost, a recent (February 2016) scientific paper from OpenAI and Google researchers, describing methods of Attacks against Deep Learning Systems (such as used in the self-driving cars), that can, in a highly sophisticated, humanly undetectable way, fool an AI to see something that isn’t there, or to ‘think’ that it sees something else instead.
Now connecting the dots...
Adversarial Examples in the Physical World
Explaining and Harnessing Adversarial Examples
Practical Black-Box Attacks against Deep Learning Systems