advertisement
Facebook
X
LinkedIn
WhatsApp
Reddit

Google’s AI creates trippy images from white noise

We don’t think there is anything to worry about any more in terms of a robotic uprising – if Google’s artificial intelligence (AI) is anything to go by.

Take the image below. The swirly picture, which definitely has hints of a drug-infused Oscar-Claude Monet, was made by Google’s AI servers. The company wanted to see what sort of art the AI could come up with when given a blank canvas – and in this case the blank canvas was pure white noise.

Iterative_Places205-GoogLeNet_4
[Image – Google]

A bit of a back story is needed to explain what’s going on here: the researchers at Google asked the company’s artificial neural network to recreate images of objects after it showed it a pictures of them.

It “teaches” the network what a certain object looks like, and then asks it to extract it from a different picture.

“In this case we simply feed the network an arbitrary image or photo and let the network analyse the picture. We then pick a layer and ask the network to enhance whatever it detected,” a post on Google’s blog explained.

“Each layer of the network deals with features at a different level of abstraction, so the complexity of features we generate depends on which layer we choose to enhance. For example, lower layers tend to produce strokes or simple ornament-like patterns, because those layers are sensitive to basic features such as edges and their orientations.”

But that was for simple pictures of animals, the sky and towers. For the picture above, it reversed the process. In essence, the neural network was dreaming about images and objects that it knew about and visualised them, without being asked to picture a specific object.

You can read the fairly technical explanations and research on Google’s Research Blog.

advertisement

About Author

advertisement

Related News

advertisement