DALL-E AI art tool expands access in beta stage The lab behind the artificial intelligence art tool is giving access to up to a million people on its waiting list, just as worries grow about possible abuse.

Surreal or too real? Breathtaking AI tool DALL-E takes its images to a bigger stage

  • Download
  • <iframe src="https://www.npr.org/player/embed/1112331013/1112579680" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript

JUANA SUMMERS, HOST:

A new tool powered by artificial intelligence is creating pictures that wow the internet. It's called DALL-E after Salvador Dali and the Pixar movie "Wall-E." Until now, only a relatively few researchers, artists and others have been able to try it, but the company behind DALL-E says it's going to give access to a million more people starting today. NPR's Bobby Allyn joins us to discuss the art DALL-E can make and why some people are really worried about its possible uses. Hey there, Bobby.

BOBBY ALLYN, BYLINE: Hey, Juana.

SUMMERS: OK, so first, this is all incredibly visual - and we are on the radio. Can you describe how DALL-E works?

ALLYN: (Laughter) Yeah. So I think the best way to understand it is just to use it. And, Juana, you and I were talking yesterday, and I said, OK, what better way to get how this thing works but to try it, right? So I asked you to give me a few prompts of images you'd like to see. And what did you tell me again?

SUMMERS: Yeah. So the first one that I gave you was a taco playing pinball in the middle of the road. I guess I must have been hungry. And then another one was Satan wearing sunglasses, parachuting into Mount Vesuvius.

ALLYN: I love it. OK, so I took those descriptions, and I entered them into the DALL-E system. Now, the system is built on this enormous database of images - right? - and then this really fancy computer code makes connections almost like the human brain and spits out images in seconds. So those prompts you just gave me - I put them into the system. I hit enter. And the results were pretty whimsical. Why don't you take a look and tell me what you think?

SUMMERS: OK, so I'm looking at the first one now, and it does not look like DALL-E understands pinball as well as I do, but there's a couple of animated tacos. They are indeed in the middle of the road. One of them is skateboarding down a street. They're all really cute. And then this other one I might actually have nightmares about.

ALLYN: Yeah, it's ghoulish.

SUMMERS: It is very ghoulish. There are some sunglasses on a parachute going into Mount Vesuvius. Quite literal.

ALLYN: (Laughter) Yeah, so most people using DALL-E right now are doing just this - are having fun. But, you know, if you take a look at them, it doesn't, at least to me, look like a computer generated them. I mean, they're quite impressive. So people now who are testing it out are doing things like this. They're doing realistic photos of silly situations. So it's all fun and games, right? But I spoke to Joanne Jang. She's the product manager of DALL-E, and she said, you know, Juana, some people have some pretty interesting ways that they're using it now.

JOANNE JANG: A surgeon told us that they've been using it, actually, to help their patients visualize post jaw alignment surgery results. Filmmakers have told us that it has cut the time that it takes to storyboard from weeks to days. Restaurants run their menu items by DALL-E and got whole new plating ideas and menu inspirations that they've incorporated.

SUMMERS: OK. But, Bobby, as I understand it, there are also some concerns about how DALL-E might be used. How are they accounting for those? And what are those concerns?

ALLYN: So DALL-E will not let you generate images if you're using the name of a real person, like President Biden. But if you type in, say, a 79-year-old white American president, you do get some photos that vaguely resemble him. So, you know, there's no way to get an exact photo of, say, our president, but there is a way to get sort of some kind of photo that kind of looks like him. Right? It will not let you make pornographic images, violent images, images that are hateful or any photos showing things like ballot boxes and protests. You can't do that. They're really afraid that people would try to generate them and launch disinformation campaigns.

SUMMERS: So, Bobby, you have told us about DALL-E, how it works. You have shown me how it works. But who is behind it?

ALLYN: Yeah, it's a company called OpenAI. It was founded by Elon Musk - although he's no longer involved - and tech entrepreneur Sam Altman. It's financially backed by Microsoft. And DALL-E is ahead of the game when it comes to AI image generation. But tech giants - Google, Facebook, Amazon - they're all trying to develop similar AI tools right now. And after this trial period, you know, we'll see what happens. Maybe more will join, or maybe OpenAI will just stop inviting people.

SUMMERS: All right. That is NPR Tech reporter Bobby Allyn. Thank you so much.

ALLYN: Thanks, Juana.

Copyright © 2022 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.

NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.