An artificially intelligent (AI) machine that creates surreal nude portraits has been built by a teenager in Virginia.
AI whiz Robbie Barrat fed a neural network – an AI that functions like the human brain – thousands of naked portraits and then trained it to create its own racy artworks.
In a Twitter post, he said the software often paints people as fleshy blobs spouting random tendrils and limbs, adding: ‘I wonder if that’s how machines see us’.
An artificially intelligent (AI) machine that creates surreal nude portraits has been built by a teenager in Virginia
While most of the women in the images appear lumpy and misshapen, some of the subjects closely resemble slender, standing figures.
Mr Barrat, who recently graduated high school in West Virginia, said he is currently doing ‘deep learning interning’ for AI computing giant NVIDIA.
He said the AI created images by interpreting a set of ‘rules’ about the shapes, sizes and colours of figures that it learned by looking at thousands of nude portraits.
To train the algorithm – a Generative Adversarial Network (GAN) – he mostly used portraits from ‘a few hundred years ago at least’.
GANs are a class of artificial intelligence algorithms that uses two different neural networks, one called a ‘generator’ and one a ‘discriminator.’
Speaking to CNET, Mr Barrat said: ‘The generator tries to come up with paintings that fool the discriminator, and the discriminator tries to learn how to tell the difference between real paintings from the dataset and fake paintings the generator feeds it.
‘They both get better and better at their jobs over time, so the longer the GAN is trained, the more realistic the outputs will be.’
AI whiz Robbie Barrat fed a neural network – an AI that functions like the human brain – thousands of naked portraits and then trained it to create its own racy artworks
In a post to Twitter , he said the software often paints people as fleshy blobs spouting random tendrils and limbs, adding: ‘I wonder if that’s how machines see us’
While most of the women in the images appear lumpy and misshapen, some of the subjects closely resemble slender, standing figures
Mr Barrat, who recently graduated high school in West Virginia, said he is currently doing ‘deep learning interning’ for AI computing giant NVIDIA
Sometimes the GAN will slip into a ‘local minima’, in which the generator and discriminator are stuck in a loop in which they keep trying to fool one another.
‘In this case, the generator keeps generating fleshy blobs that fool the discriminator pretty well, so overall they stop getting better at painting,’ he said.
Mr Barrat compared the computer-generated pieces to the works of 20th Century American artist Sol LeWitt.
To train the algorithm, he used a Generative Adversarial Network (GAN) fed mostly with nude portraits from ‘a few hundred years ago at least’
GANs are a class of artificial intelligence algorithms that uses two different neural networks, one called a ‘generator’ and one a ‘discriminator’
Speaking to CNET , Mr Barrat said: ‘The generator tries to come up with paintings that fool the discriminator, and the discriminator tries to learn how to tell the difference between real paintings from the dataset and fake paintings the generator feeds it’
The two AIs both get better and better at their respective jobs over time, so the longer the GAN is trained, the more realistic the outputs will be, according to Mr Barrat
LeWitt would send a list of rules for a drawing to a museum, where visitors would interpret the rules and create the artwork there.
After he shared some of his AI images on Twitter, Mr Barrat received a number of responses from fascinated users.
@SapphirePelican tweeted: ‘Imagine in a hundred years when AI can create infinities of staggering artworks, symphonies, cuisines, and novels in the blink of an eye.
Often the GAN will slip into a ‘local minima’, in which the generator and discriminator are stuck in a loop in which they keep trying to fool one another, creating ‘blobby’ results (pictured). In this image a single model produced by an AI is shown in two different positions
Mr Barrat compared the computer-generated pieces (pictured) to the works of 20th Century American artist Sol LeWitt
After he shared some of his AI images on Twitter, Mr Barrat received a number of responses from fascinated users. @SapphirePelican tweeted: ‘Imagine in a hundred years when AI can create infinities of staggering artworks, symphonies, cuisines, and novels in the blink of an eye
AI systems rely on artificial neural networks (ANNs), which try to simulate the way the brain works in order to learn
‘It will force humans to reconsider what they call art and push its boundaries ever more ferociously.’
@krdluzni wrote: ‘I find it disturbing that my brain still recognizes human structure in these even when they are this warped.’
@PraxJarvin said: ‘I’m reminded of the aliens on Star Trek who reffered [sic] to humans as ugly bags of mostly water.’
ANNs can be trained to recognise patterns in information – including speech, text data, or visual images – and are the basis for a large number of the developments in AI over recent years
Mr Barrat told CNET: The way that it paints faces makes me uncomfortable. It always paints them as like, purple and yellow globs – that isn’t in the training set so I’m actually still not sure why it does that’
In response to Mr Barrat’s post on Twitter, user @krdluzni wrote: ‘I find it disturbing that my brain still recognizes human structure in these even when they are this warped’