How three French students used borrowed code to put the first AI portrait in Christie’s

On Friday, October 25th, Christie 's performed a very unusual sale. 3 days As part of the Prints & Multiples event A portrait of Edmond Belamy, Auction of a canvas with a mottled look that looks like 18th on the canvas. Century shrine. It is expected to bring reasonable prices between $ 7,000 and $ 10,000, but the feature of the work is "created by artificial intelligence," Christie says. "[it] If you go under the hammer, you will notice the emergence of artificial intelligence art at the world auction stage."

But for the fast-growing members of the AI ​​community there is a portrait of . Edmond Belamy : It was noticeable.

The fingerprints were created by a 25-year-old French student, Obvious, whose goal is to "explain and democratize" artificial intelligence through art. Over the past year they have produced a series of portraits depicting fictional Belamy family members and expanded their work with interesting press releases. But insiders say that most of the code used to generate this fingerprint is the work of other artists and programmers. Robbie Barrat, 19, is a high school graduate who shares algorithms online with an open source license online.

Obvious members do not deny that they have borrowed heavily from Barrat's code, but they have not revealed it until recently. This has taken the first step towards open, cooperative and mainstream interest to some members of the AI ​​community. Seeing AI portraits sold on Christie 's is a milestone for the entire community, but is this event taken by an outsider?

In order to understand the concerns about attribution and professionalism that cast shadows at Belamy Auctions, you need to know about the tools used in the AI ​​art world. The most important of these is the Genetic Deficit Network (GAN), a type of algorithm originally designed by Ian Goodfellow, a researcher working at Google. The name "Belamy" chosen by Obvious is a tribute to him.

Goodfellow's work on the GAN is the story of AI legend. The story asked a simple question that one night in 2014, we would be drinking beer with our friends. What if neural networks can compete with each other?

GAN's basic idea is to train the network to find patterns and create copies in a particular set of data (such as a picture of a kitchen or 18th century portrait). Then a second network, called the discriminator, determines the action and sends it back if it can find a difference between the original and the new sample. The first network then tries to sly the discriminator too much by modifying the data. This process is repeated until the generator network creates a meaningless fake. Think like a security guard in a club. Spend your drunken friend away until you are not drunk enough to get in.

This basic idea proved a duel network. It is incredibly powerful and GAN is the cornerstone of modern machine learning at present. They were particularly beneficial in artificial intelligence, and pictures created using GAN have a unique aesthetic that reflects how algorithms process information. Networks know how to copy basic visual patterns, but they do not know how to get along well. The result is that the boundaries are unclear, the figures fuse together, and the anatomical rules come out of the window. This aesthetic has even a pseudonym. Google AI engineer François Chollet: GANism proposed .

Barrat has become a leading light in the world of GAN art by creating headlines with his surreal nude and landscapes. He also shares the algorithms GitHub uses to create these images, helping fellow artist coder to configure and run neural networks. This is a way to find Barrat's algorithm Belamy for Obvious's technical leader Hugo Caselles-Dupré and PhD students in Paris.

How much work can Barrat or Obvious offer is a difficult question. Creating an image using GAN is a multi-step process. First, the network collects training data as a "scraping tool" to replicate. You then configure the generation algorithm. This is the most time-consuming and difficult part. Next, run the algorithm and sort the output to select the hundreds or thousands of best examples generated.

According to The Verge Caselles-Dupré easily recognizes that Obvious borrows Barrat's elements. Like the "scraper" used to collect images), but he says he has modified the code to produce his own portraits as he likes. Caselles-Dupré says, "If you're talking about code, it's not a big deal." "But when we talk about computer work (19459003) it's possible to make it, and there have been a lot of efforts." (Some of them seem to have harassed Barrat, see discussion on GitHub.) Caselles-Dupré , Alias ​​Caselles, and Barrat.)


Two GAN-generated nudes from Barrat.
Image: Robbie Barrat

Barrat told The Verge that Obvious approached him and used the "component" of the code. We then almost immediately started the same tasks as the results of the pre-trained people and landscape network hosted by GitHub.

Other members of the AI ​​art community say it is not important, no matter what the obvious changes are. Mario Klingemann has won an award for his work with GAN The Bellge says that "90% of the actual work is done by [Barrat] "Tom White, a scholar and artificial intelligence writer from New Zealand, says his work is very similar, and he even runs Barrat's code with zero adjustment to compare the results.

The results below are from White's experiments. The Bellge says: "We can not prove that the auction portrait is derived from Robbie's portrait GAN model, but it certainly doubts that it can be converted to an image that this model can produce."


The images are very similar to Belamy portraits.
Image: Tom White

But how important is the code I borrowed? The creative world is full of examples of budget counts ranging from hip-hop to sampling to Marcel Duchamp's preview. And obviously they have given their portraits a new status by presenting them as traditional art. They have signed using a segment of algorithm code hung on a gold frame: "min G max D x [log (D(x))] + z [log(1 – D (G(z)))]".

Jason Bailey, a digital art blogger who runs the Artnome site, said, "Why is it that they have declared their work with AI-readyymade and missed the opportunity to bring us the first digital duchamp?" Said Karlingemann. The Verge But the reputation of this auction and not the Barrat, Obvious has gained a reputation and attended, Bailey adds, "If you are Robbie, I will be quite disgusted, and Obvious is not going to do it. "

Barrat has no grudge at all, and most are worried that an auction can give outsiders a false impression of artificial intelligence." I am more interested in the fact that real artists using AI are not getting attention "

Belamy I agree that there are many technical defects in the portraits (for example, textures with low resolution and smearing). It shows that the maker is not the main character of GAN brush. "To me this is dilettante's work, equivalent to a five-year graffiti that only parents can read," says Klingemann. "

Given this fact, it may seem unlikely that Belémy portraits have been chosen by Christie. But part of the reason for the obvious reason was the intention to accept a special narrative about artificial intelligence art.

In the media the Trio sent earlier this year, they led to this message. Their motto is, "Creativity is not just for humans," at a January press conference. "Right after artificial intelligence, they sold their first portrait to a French collector, and the articles dealing with the news were" made by artificial intelligence "


Image: Obvious

For artificial intelligence researchers and artists, this is a misunderstanding, and it's a bit of a misunderstanding to readers Bailey says that machine learning systems are more complex and autonomous than they really are. "Bailey says," It's like talking when you're anonymous or when you do not think someone else is listening. "What algorithm is the tool (against active collaborators The problem has been discussed for a long time by the artist coder, but certainly the self-driven agent No, Barrat said, "Anyone who works on artificial intelligence and art will realize. [this] "Media descriptions that focus on non-controllable artificial intelligence and" robots "that replace human work often ignore these nuances.

The press blames journalists for this kind of coverage, Caselles-Dupré said, "To the people we sit on the couch, how can we meet these people?" "We tried a lot of foolish things and we fully appreciate it."

Christie's is definitely a " The company had approached Obvious last summer, and since then has refurbished the company's brand and created a profitable new art Richard Lloyd, chief executive officer of Christophe, said, "It is the same work that we have sold for 250 years, though not drawn by people with powdered sugar. (At the time of writing, Christie refused several requests for interviews with The Verge .)

In this sense, the auction was already a huge success and led to media coverage around the world. In turn, it was clearly given the kind of trust that previously gained the support of the agency that sold the world's most expensive painting (Leonardo da Vinci Salvator Mundi $ $ 450 million) .

Belamy The most exciting thing about the auction is that trio machine learning of students with minimal background, including borrowed codes, inkjet printers and some passionate press releases, is at the heart of a landmark event in art history.

"People are quick to paint others with heroes or villains [Obvious] People do not do terrible things, Bailey says. "Three 25-year-olds wanted something entrepreneurial and entrepreneurial. They knew AI was getting a lot of attention," he said. "Nothing makes people bad . They did not know that it would explode in the sense that they would be in Christie's."

Caselles-Dupré obviously shocked It seems to have been received. "We were just really, really, really surprised," he added, "relieved" to tell the whole story. "The best scenario for us is to get one or two articles, sell pieces, earn money, and run another project." Instead, go to New York and display artificial intelligence art on the map. . Hopefully, they will not be the last.

Add a Comment

User*

Your email address will not be published. Required fields are marked *

Open

Close