#HTE

They’ve Taught an AI to Compose an Original Rembrandt Painting

Earlier this year there was an unprecedented collaboration between ING, Microsoft and TU Delft, with the willing participation of Dutch museums Mauritshuis and Rembrandthuis, to teach an artificial intelligence to paint a Rembrandt.

Not to duplicate an existing Rembrandt, but to paint an original one, with a unique composition and using a person who never existed as the subject.

The idea was that the AI could analyze and learn Rembrandt’s style and then produce, on request, an original portrait based on certain parameters (“Thirtysomething male, black hat, white collar, looking to the right.”) Here’s how they did it, and here’s the frighteningly convincing image that the computer reproduced—on a 3D printer, no less, to simulate the height of brushstrokes:

It was called the Next Rembrandt project, and I have very mixed feelings about it. While it is unquestionably an impressive technological achievement, ought we be moving AI into those non-numerical areas of human creativity? I suppose one could argue that the Next Rembrandt painting is not truly creative, just a mere average of existing values; but the video made it sound as if the algorithm was a bit more sophisticated than that.

It’s not difficult to imagine, as a next step, feeding an AI the works of all great masters in a particular genre, and asking the machine to create works that can adhere to the genre while “expressing” the work in a style unlike these masters’ work. Then, what’s the step after that? You tell us.

http://www.core77.com/posts/58038/Theyve-Taught-an-AI-to-Compose-an-Original-Rembrandt-Painting