A hesitant look back at the disturbing world of Google DeepDream

14 Dec 2015

An image processed using Google’s DeepDream code

Google’s DeepDream code unleashed a disturbing photo trend on the internet this summer.

In June this year, a Google Research Blog post titled Inceptionism: Going Deeper into Neural Networks described how an artificial intelligence (AI) program created by scientists at Google ‘reads’ and interprets images.

This neural network AI analysis – dubbed DeepDream – can be keenly tuned to recognise specific images to the point that it will find them even where they don’t exist. Just think of how the human mind can perceive various objects in cloud formations and you get the gist.

Google DeepDream

Left: Original photo by Zachi Evenor. Right: processed by Günther Noack, Software Engineer

Google researchers decided to turn the DeepDream code on its head, and use it to generate new pictures based on the visualisations of artificial neural networks’ attempts to identify details in an image.

Google DeepDream

Top: Input image. Bottom: output image made using a network trained on places by MIT Computer Science and AI Laboratory

They even used DeepDream to process a random-noise image, so that the result becomes purely the result of the neural network, and the results were quite beautiful.

Google DeepDream

Neural net ‘dreams’ – generated purely from random noise, using a network trained on places by MIT Computer Science and AI Laboratory

In July, Google made DeepDream code publicly available, and that’s when things got out of hand. A swathe of custom images, GIFs and videos started to spread online using the hashtag #DeepDream.

At first, it was just a little trippy, but still quite pretty.

But, this being the internet, things quickly took a turn for the weird.

Those networks trained to recognise animals were finding hybrid puppy-slugs everywhere.

The NSA HQ got weird.

Exploring got weird.

Jungles got weird.

Cycling got weird.

Mad Max got weird.

Art got weird.

Dank memes got weird.

Stock photos from the corporate world got weird(er).

And that was just the start. Soon, unsettling DeepDream visualisations were ruining lives.

Dancing, ruined. 

by Ross Goodwin (http://rossgoodwin.com)

Happy Cat, ruined.

Donuts, ruined.

Dinner, ruined.

DeepDream: the first internet meme to give us all a regret-filled hangover.

Keep watching Siliconrepublic.com as we look back on the greatest memes of 2015. Let us know your favourite at @siliconrepublic on Twitter.

Main image from Google’s collection of DeepDream images on Google Photos

Elaine Burke is the host of For Tech’s Sake, a co-production from Silicon Republic and The HeadStuff Podcast Network. She was previously the editor of Silicon Republic.