Google discovered that its neural net AI engineered to understand and caption photos could also work in reverse to...

Originally shared by Kevin Kelly

Google discovered that its neural net AI engineered to understand and caption photos could also work in reverse to generate photos. They take a source image and tell it to make X in the images and it will generate new original dream like images. Google just released the source code, and within a day or so people have begun playing around with it. Here is one example of someone who used it to find/make eyes in pictures.

http://www.dailymail.co.uk/news/article-3149466/A-nightmarish-vision-world-seen-eyes-Google-s-artificial-intelligence-programme-TAUGHT-recognise-sort-images.html

Several things strike me about the images.

1) They are truly novel, and incredibly sur-realistic. This is just day one. Imagine what will be possible in a few years. Especially when the code is optimized to create.
2) They are remarkably trippy, and echo the patterns one see in hallucinogens. I suspect these techniques will help biological neuroscience.
3) The issue of authorship has not been raised yet, but like the issue Google had when these same algorithms captioned black people as gorillas, the issue of what to do when it draws something taboo will be an inevitable challenge.
http://www.dailymail.co.uk/news/article-3149466/A-nightmarish-vision-world-seen-eyes-Google-s-artificial-intelligence-programme-TAUGHT-recognise-sort-images.html