Quantify creativity, genius of an artist with an algorithm, that is the challenge embarked Ahmed Elgammaly and Babak Salehz, two researchers of the art laboratory and the artificial intelligence of Rutgers University in New Jersey.
Via their image analysis algorithm, they have assessed and accurately note the creativity of 1,710 works. The approach may be questionable and shock the art historian, but it has the interest of any mathematical objectivity. Result, the algorithm extols artists like Michelangelo, Vermeer, Goya, Mondrian, Picasso and Lichtenstein as the works of Durer, Ingres, Rodin get the lowest ratings. And if the algorithms can play the art critics, it can also take the brushes!
Google turned its image recognition algorithms … backwards
The idea of Google’s researchers is simple: use neural networks used by Google for image recognition in reverse, either to read images, but to generate new. The result is rather psychedelic, the images generated by artificial intelligence represent images of combinations combined with colorful video effects. The style was baptized Inceptionism by researchers have even posted a gallery of their most “beautiful” works. Art historians will remember this artistic movement in the history of art …
Stung by the publications of their rivals, researchers from Facebook in turn lifted the veil on their research. Also we had the idea to exploit their Deep Learning algorithms in reverse, to create images.
The researchers explain that if artificial intelligence algorithms are effective at recognizing shapes, classify the images, they are much less comfortable when it comes to creation. The approach studied by Emily Denton of the University of New York and Soumith Chintala, Arthur Szlam and Rob Fergus artificial intelligence researchers at Facebook is a little different. The models created by these researchers, baptized Laplacian Pyramid of Adversarial Networks (LAPGAN) literally learn to generate an image from images of the same theme. On “shows” a series of pictures of boats and software, after this learning, the algorithm is able to generate a credible ship image. A sufficiently realistic image to fool 40% of human to which it is presented.
Can we call these images of artistic works born in the neuronal networks of artificial intelligence? They owe nothing to chance. Maybe he should leave a different algorithm, that of the University of Rutger, judge their creativity. Where he will place Google and Facebook? Between Lichtenstein, Mondrian and Picasso, or rather in the classics?
Translation : Google Translate
“This Is What Happens When Machines Dream”, SingularityHub, June 19, 2015
“Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks”, Research paper, June 18, 2015
“Inceptionism: Going Deeper into Neural Networks”, Google Research Blog, June 17, 2015