- Cardi B says she drugged, robbed men in her past on Instagram Live Monday 8:03 PM
- Twitter thread roasts bathtub tray ads for women Monday 7:21 PM
- Nintendo set to release two new models of the Switch—possibly in 2019 Monday 6:45 PM
- Viral cat video ‘Dear Kitten’ finds new life in TikTok challenge Monday 5:30 PM
- Here’s every show that was announced at the Apple TV+ kickoff Monday 3:53 PM
- ‘Shazam!’ embraces the spectacle and heart of the superhero genre Monday 3:45 PM
- How to mute Twitter’s suggested tweets on your timeline Monday 3:02 PM
- What you need to know about Apple’s new streaming service Monday 2:32 PM
- Text-message fanfiction is taking over Instagram Monday 1:54 PM
- Your Asus computer might have a secret backdoor Monday 1:06 PM
- Trump is already fundraising off the Mueller report—even though no one’s seen it Monday 1:01 PM
- Michael Avenatti charged with trying to extort $20 million from Nike Monday 12:51 PM
- Logan Paul says being a YouTuber is ‘wack’ Monday 12:14 PM
- James Comey posts from a forest in wake of Mueller report Monday 10:35 AM
- These are the only online dating sites worth your time Monday 10:29 AM
Drama ensues over first-ever AI-created artwork sold in a major auction—and its $432,500 grab
The collective admits to using another artist’s open source code, but says they modified it.
In a historic first, an artificial intelligence-created piece of art has been sold in a major auction, raking in $432,500 at Christie’s New York. However, a debate over whose work has contributed to this painting has ensued online.
On Thursday, Christie’s tweeted about the accomplished painting, “Edmond de Belamy from La Famille de Belamy.” According to a description of the work, the canvas print was created from a “generative adversarial network” and was one canvas of “a series of 11 unique images.”
— Christie's (@ChristiesInc) October 25, 2018
The canvas went for more than 43 times the auction house’s initial estimate of $7,000-$10,000, and went to an anonymous phone bidder, the New York Times reported.
French art collective Obvious created the work—or so it’s been told.
According to the Times, other AI artists have called the undistinguished portrait “unoriginal,” stating that generative adversarial networks have been used in art since 2015, despite Obvious’s recent growing attention. Furthermore, one of those earlier artists, 19-year-old programmer Robbie Barrat, called out Obvious on Twitter the day before Thursday’s auction, writing that he wrote and shared the code that Obvious used to make their 11-part series.
“Left: the ‘AI generated’ portrait Christie’s is auctioning off right now. Right: outputs from a neural network I trained and put online *over a year ago*,” Barrat wrote on Twitter, referencing two images he posted of Obvious’ canvas and of his multiple creations. “Does anyone else care about this? Am I crazy for thinking that they really just used my network and are selling the results?”
left: the "AI generated" portrait Christie's is auctioning off right now
right: outputs from a neural network I trained and put online *over a year ago*.
Does anyone else care about this? Am I crazy for thinking that they really just used my network and are selling the results? pic.twitter.com/wAdSOe7gwz
— Robbie Barrat (@DrBeef_) October 25, 2018
Obvious later responded, writing that they didn’t use “pre-trained models” from Barrat, just the code. It also shared a direct message of a conversation between the Obvious Twitter account and Barrat in April, in which Obvious told Barrat it had used some of his code and wanted to check if he was OK with it.
A discussion we had on April. pic.twitter.com/2r9YUoTi5l
— Obvious (@obv_ious) October 25, 2018
Barrat had responded to the message saying he was “100 percent OK” with the usage, and that he didn’t need credit. On Wednesday, however, Barrat expressed disapproval with the manner in which Obvious used his code, and he included a portion of the conversation from weeks after the initial messages in which Barrat asked for credit. Obvious responded saying that it credits Barrat in its official posts, and mentions him when it can in interviews. Barrat then accused Obvious of having not initially crediting him, and only doing so recently, to which Obvious pushed back, saying the credit came after his first request for credit in April.
Despite Barrat’s expressed disapproval amid the auction of the portrait, others online disagreed that the collective stole or misused his code, or needed permission to use it at all. Barrat had uploaded his code to GitHub and made it open source, thereby making a donation of his work to the general public for their use. And since Obvious didn’t actually use an iteration of Barrat’s art created from the code, just the code itself, they didn’t technically use work that he hadn’t already made open source.
when you post code on github and someone else runs it and prints the output and sells it to someone, nobody has stolen from you.
— twiterated prisoner’s dilemma (@sneakdotberlin) October 25, 2018
I think he added it in April 2018, but yeah, after it was already in use. Yes, it stings when other people make money and not you, but that's allowed by the original BSD license. https:/github.com/robbiebarrat/art-DCGAN/commit/b475eb0b229b751e2ef25b4f2ad3510f755a52ac
— Ellecer Valencia (@ellecer) October 26, 2018
@ChristiesInc have made their cut. Opensource means Free to use so legal routes limited IMO. Art peeps will claim some added value / narrative on top of what you released. Keep making good stuff instead.
— Marcus Lyall (@lyallmarcus) October 25, 2018
for the record: i'm not 100% convinced that they are using your model, but they are using your dataset scraper + a similar architecture so very similar results would be expected. from this interview screencap (https://t.co/341SZSWjul), it's likely vanilla pytorch-examples dcgan. pic.twitter.com/UdAcLxIpAG
— tom white (@dribnet) October 25, 2018
of course they have better marketing – they aren't artists, they're marketers.
I put it out open source with no license so there's no legal stuff I can claim. I really just wanted to provide the AI + art community with some fun tools – never thought anyone would do this.
— Robbie Barrat (@DrBeef_) October 25, 2018
“I put it out open source with no license so there’s no legal stuff I can claim. I really just wanted to provide the AI + art community with some fun tools – never thought anyone would do this,” Barrat wrote on Twitter.
I've been trying to play it off and spin it positively – when news people have asked me about this event I usually just try and mention actual AI artists and say to check out their work instead, but this is *really* becoming incredibly upsetting.
— Robbie Barrat (@DrBeef_) October 25, 2018
Speaking to the Verge on Tuesday, Obvious tech lead Hugo Caselles-Dupré said that they did use Barrat’s code in the creation of the nearly-half-a-million-dollar portrait, but that it was modified it. On Twitter, Obvious wrote that it had used the same data, but “cleaned by hand (removed portraits with several persons and stuff), changed the number of filters in the discriminator (worked better for us), and [used] PyTorch (not torch, i.e. Lua code like in the original repo) code…”
In an official statement, Obvious thanked the AI community and Barrat, calling him a great influence” for the collective’s work.
“We would like to thank the AI community, especially to those who have been pioneering the use of this new technology, including Ian Goodfellow, the creator of the GAN algorithm,” Obvious said. “And artist Robbie Barrat, who has been a great influence for us.”
However, Barrat told the Verge that he doesn’t hold a grudge against the collective, and is “mostly annoyed” at the perception of AI art that outsiders might take away from the auction.
“I’m more concerned about the fact that actual artists using AI are being deprived of the spotlight,” Barrat said. “It’s a very bad first impression for the field to have.”
Update 5pm CT, Oct. 26: In an email to the Daily Dot and other news outlets, Obvious wrote that the collective “made a point” to credit other people who are using GAN before the auction, including Barrat and Ian Goodfellow, who created the GAN algorithm. Obvious also reiterated that they modified Barrat’s code “with his approval” to use for the Famille de Belamy series, and that they gave Barrat credit in their original Medium post about the series after he asked.
“Indeed, Robbie Barrat deserves credit, which we gave in our main Medium post as soon as he asked back in April. We also credited him right after the auction. Once again, we would like to take this opportunity to credit him. Moreover, we strongly advise those interested in the field to follow this artist, who continues to deliver amazing artwork,” the statement continued. “The sharing of this technology is something that excites us as we hope to learn from others the same way we hope they learn from us. We are influenced by many artists working in this space, as we believe is the case across the history of art.”
Update 8pm CT, Oct. 26: In a statement to the Daily Dot, Barrat said he doesn’t play to take legal action against Obvious.
“Money isn’t what I’m concerned about,” Barrat said. “My main concern is just that actual artists working with AI get more recognition, and don’t get shadowed over by this collective of marketers and their recent sale. Artists like @quasimondo, @glagolista, and @dribnet have been making incredibly compelling artwork using AI as an artist’s tool for a while—people like them need to be recognized since they’re actually pioneering and pushing boundaries in their art, not just using pre-existing code and putting it in a fancy frame.”
Barrat says he still plans to keep making components of his projects open source.
“OSS is incredibly important to me since it’s how I learned to program—by looking at other people’s open source code,” Barrat said.
Samantha Grasso is a former IRL staff writer for the Daily Dot with a reporting emphasis on immigration. Her work has appeared on Los Angeles Magazine, Death And Taxes, Revelist, Texts From Last Night, Austin Monthly, and she has previously contributed to Texas Monthly.