Google has modified its image recognition neural net with a feedback loop.. and it now “dreams up” images of its own, even from nothing…
But if the neural network is tasked with finding a more complex feature – such as animals – in an image, it ends up generating a much more disturbing hallucination:
Ultimately, the software can even run on an image which is nothing more than random noise, generating features that are entirely of its own imagination.
“One way to visualise what goes on is to turn the network upside down and ask it to enhance an input image in such a way as to elicit a particular interpretation,” they add. “Say you want to know what sort of image would result in ‘banana’. Start with an image full of random noise, then gradually tweak the image towards what the neural net considers a banana.”
The image recognition software has already made it into consumer products. Google’s new photo service, Google Photos, features the option to search images with text: entering “dog”, for instance, will pull out every image Google can find which has a dog in it (and occasionally images with other quadrupedal mammals, as well).
So there you have it: Androids don’t just dream of electric sheep; they also dream of mesmerising, multicoloured landscapes.