Trump drives a Mario Kart, Darth Vader mows the lawn: thanks to artificial intelligence, absurd thoughts become images. What is behind the new technology?
Bizarre series of pictures have recently appeared on Twitter and Instagram: worlds that don’t belong together, distorted faces, comic-like arrangements.
For example, there is Donald Trump doing his rounds in the game “Mario Kart”:
Or this weird mix of the kiwi bird and the fruit of the same name:
Or the “Star Wars” villain Darth Vader mowing the lawn:
A real competition for the weirdest combinations seems to have broken out. Conceived in the minds of humans – executed by an artificial intelligence (AI).
What’s it all about? The Dall-E application, named after the artist Salvador Dalí, makes it possible. She was trained to translate written commands into visual form.
We already know AI from many areas, such as face recognition in photos. But now that she becomes a creator of works is a new dimension.
Who is behind this? The research group Open AI, funded by Tesla founder Elon Musk and by Microsoft. Their application Dall-E has been around for about a year and a half, but its use is limited to a small circle.
Since there has been a free, freely accessible, but somewhat simpler version, users have been experimenting with it. It was called Dalle-E Mini and has now been renamed Craiyon to avoid confusion with the original.
How does Dall-E or Craiyon work? After the command, it usually takes a few minutes for the AI to spit out nine suggestions. To do this, the machine reassembles the puzzle pieces from the network. She has to invent missing information – which sometimes works better, sometimes worse. The free version uses around 15 million images as a basis, the professional version around 27 times the amount of data.
Are the results convincing? The images in the free application are often distorted, miscomposed, or spongy. Sometimes the images look like realistic photos, but mostly more like surreal paintings. This has its own appeal, but the AI still needs to learn.
On the command “Bruce Springsteen with Billie Eilish on the stage” or “Pope on an e-scooter” you can see the desired figures in the result, but the quality is far from photo-realistic. “Heidi, painted by Van Gogh” works quite well, but “Blocher by Ferdinand Hodler” doesn’t.
The author’s examples
What else can be expected from the technology? The latest, powerful version of the Dall-E 2 can apparently not only imitate styles, but even imitate the complex requirements and characteristics of works of art. The example of a sea otter instead of the girl with a pearl earring in the famous picture by Jan Vermeer caused a stir. The quality is amazing.
The difference from the free version is obvious. This fails at the task.
How will the Dall-E be used in the future? The professional application Dall-E 2 will soon be offered to customers who have to pay. Google is also working on its own system. In certain areas, the rapidly increasing quality could become serious competition for illustrators, image editors or photographers who produce stock images, for example.
When AI creates photorealistic images, does that increase the risk of misuse? This is one reason why the technology has so far only been accessible to selected test users. Open AI and Google assure that they artificially limit the AI in terms of human faces and nudity. This should prevent deepfakes, i.e. faces from being misused in other situations.
In fact, no faces can be seen in the examples of Open AI so far, only animals, landscapes or other arrangements.
#creates #art #Elton #John #painted #Picasso #problem