Pictures drawn by artificial intelligence created by Google

Google recently tested artificial neural networks designed to recognize and describe an image. In order to understand how the system they created "thinks", the development team "asked" it to create images based on certain requests. The results, demonstrating how an artificial intelligence (AI) network interprets the world around them, are bizarre and often mesmerizing.

In a new report titled "Inceptionism: Deepening into Neural Networks," the Google research team takes a deep dive into how artificial neural networks work, especially as image recognition software.

(Total 9 photos)

Source: factroom.ru

1. Mahogany

The research team tried to "teach" the network by showing it examples of images it was supposed to remember. For example, in order to teach the AI what the word "mountains" means, they showed it various photos of mountains.

2. Mount McCowen

However, they got rather strange results when they asked the network to create its own image based on what it had "learned". In one instance, when told to draw dumbbells, the network created an image that combined metal and human hands, probably because every dumbbell image it was shown included hands holding dumbbells.

3. Repeating places

In some other tests, the research team asked neural networks to find specific things in images that weren't actually there. The idea was to have the AI modify the initial image to get the desired object.

4. Image generation

In other tests, Google employees told the AI network to produce random images without prompts, based only on random neural memories present in static memory.

5. Google researchers talk about such randomly generated images as the "dreams" of the artificial neural network.

6 Weird Seattle

It turns out that the AI can sleep and see amazing sheep along with deformed birds and multiple eyes...

7. Waterfall

…and can also see a lot of other crazy visuals. This is what The Lord of the Rings looks like when you watch it on LSD.

8. If AI paintings are partly an example of modernist art, then it is probably quite acceptable that AI used several of Edvard Munch's tricks with which he created The Scream. But what about the persistent use of the eyes? And is it… the dog on the left side of the picture?

9. Scream

Other images turned out to be unusually beautiful and complex at the same time. The Google team plans to continue to monitor what images are formed at deep neural levels, as they decided to continue training artificial intelligence so that it can better recognize the necessary images.