What follows is part one of my full hour-and-a-half long interview Obvious’ tech lead, Hugo Caselles-Dupré.
When Hugo shared with me that the media had gotten his story all wrong and it was really frustrating him, I offered to give him a longer format on Artnome to tell his side of the story. I believe this interview is the most accurate and thorough description of how three friends with no artistic background went from failing to sell AI art on eBay to being named one of Time’s “50 Genius Companies of 2018” and being the first AI artists to have work sold at Christie’s.
In the first half of the interview, we cover how Obvious started using GANs (generative adversarial networks) and how their work ended up at Christie’s. In the second half of the interview, I ask what contributions they made in creating the Portrait of Edmond Belamy, why they exaggerated the amount of credit the algorithm deserved in making the art, and if they believe Robbie Barrat deserves credit for the Portrait of Edmond Belamy, since they used his code and based the project on his prior artwork.
If you are simply interested in the highlights and the general gist of the interview, you can read my summary, an article I wrote called “The AI Art At Christie’s Is Not What You Think.” That said, Hugo’s story is fascinating on many levels, and if you have the time, I do recommend taking in the full interview.
Interview With Hugo Caselles-Dupré
Jason Bailey: So you are the computer science member of the team?
Hugo Caselles-Dupré: Yeah, exactly. I'm actually the one who discovered the algorithm during my research. And I talked about this with my friends, and then we wanted to do this project together. Because, as you may know, it's my friends with which I live. We've been friends since high school. We live together, and we always wanted to do something together, and we settled down on this art project. And then rest happened.
JB: A lot happened really quickly, right?
HC: Exactly. Yeah. I guess the main point of what I was saying to you earlier is that, yeah, we were like struck at the proportion that this took for us. It's, like, really unexpected what started as a side project became so important. We were really overwhelmed with all the noise around this.
JB: I don’t think the press is getting the story of AI art right. It sounds like you also feel like the media is getting it wrong? It also seems like people from the AI art community are frustrated with you. It appears you are saying certain things in the press, but from what I understand, you believe a lot of that is because the press are twisting what you say?
HC: Yeah. For sure. I think if I was not involved in this project, if I saw these articles, I would be very skeptical. I would think that this is some kind of fraud. I don't know if you saw the recent Reuters article about this. It was pretty shocking because the whole article is quite nice, I would say, and the video is cool. So they were like, ‘We want to do this interview with you. We want to come to your flat. We want to see who prints the artworks. We want to see who is your mentor, the one who bought the first piece.’ So they wanted to do something really complete so we thought, ‘Hey, it's a great way to show what we wanted to do.’ And so when we saw the title is like The artist as an algorithm: robot-made Rembrandt for sale, it's like, ‘What?’ It's like, we don't work with robots at all; we have no affiliation to Rembrandt. It would be a disgrace to say that what we do has anything to do with what Rembrandt did.
So when we saw that, we met the journalist, and they were like, ‘Ah, yes, sorry, but that's how it works.’ And also, it's pretty hard for us to control this, because for example, Reuters, they set up all this information, so there is a news article, there is a video, and then they send it to their clients, and their clients just take it as it is. So we had many other articles with the same title and with the same information that we we didn’t really want to show because it doesn't represent what we really wanted to do at first, and which is pretty simple, and that's really presumptuous. I guess it's just like we all three guys … thought that GANs was something that people should know about and really wanted to show what AI really is today. It's even harder for me because, as a researcher in machine learning, I really want the people to know what AI is right now, because there is a lot of hype around it, there are a lot of stories about it, and it’s all influenced by the culture. A lot of this cultural background makes it so that it's really easy to have this misconception about what is AI. And so we figured out GANs were a great way, an easy way to show what AI is.
Right now it's like a computer program that can do something that we find impressive, but it is not really that impressive in the end, because if you think it's some kind of robot with a human-like mind, okay, it's really, really impressive, but it's not -- but still, it's pretty impressive that a computer program can create really interesting visuals and realistic photos from only examples. So we think it was really a great way to show what AI is, and then it really turned out in a really different way. And we totally understand that the whole community around AI and art took this as a really bad thing, because for them, I think that they feel that what they do is not represented as it is, as they think it is. And now, we agree with them, but we can't really convey this message because we are being a big misrepresented.
JB: Of the three of you, you are the technical one, so when there are reports that are technically incorrect, it looks bad and backfires on you?
HC: Exactly. Backfiring on me really bad. I was a bit worried at first, and then in the machine learning Reddit [forum], there was a post about this Reuters article, so I was like, ‘Okay, this is too much. I need to react to this.’
Because at first we were like, ‘Okay, let’s not react to things.’ We don't want to upset anybody. We don't want to make drama of it. We just want to do what we want to do. And so this post on Reddit, I feel that it was really conveying a bad message to the research community. And so I responded, and I saw that the whole community was upvoting my post and saying, ‘This sucks. This happens a lot on the general media about AI. You need to keep on working on pointing to the right stuff.’
And that's why, on the website, we have an announcement where we say what we really think and why we wanted to do that, and that no, we are not inventing algorithms. That's why we call the collection the Belamy collection, it’s a tribute to Ian Goodfellow, which is a scientist that I really admire. Because, yeah, all of the sites, they are really pushing that misconception around AI. And for my friends, it's also different, because now, even though they don't come from a scientist's background, now they do know lots about GANs and how they work. And so for them, it's so hard to convey a message and to really say what you want to say.
JB: So Mario Klingemann’s quote in a recent Art News article says, “Pretty much everyone who is working seriously in this field is shaking their head in disbelief about the lack of judgment when it comes to featuring Obvious, and of course, Christie’s decision to auction them out of all artists who work with neural networks.” For artists that have worked for five to ten years on AI like Mario, it is frustrating to see someone come to GANs with no artistic background, use a stock GAN, and end up at Christie’s. How would you respond to them? Do you understand why they would be frustrated?
HC: Yeah. And we totally and actually agree and understated how they can be frustrated by that, because they have been putting a lot of work in what they do, and we really admire them because some of their pieces are really, really interesting, really creative. And we were inspired by that, too.
I’ve done some artwork when I was young, I took some painting classes and things like that. We haven't been working with art and AI for a long time. And at first, it was just like a really cool way to present what GANs are. First, you are doing this portrait and making it really accessible, because the fact that you put in it a golden frame and put this signature on it, it's really easy to understand ... For us, we thought it was a way to start great conversation, great debates, very easily.
But yeah, we were very surprised when Christie's came to us and said, “Okay, we want to auction this.” For us, it was impossible to refuse that because it was such a great opportunity to show this technology to a much larger audience and lots of people, and to present to them what I do in my research and what we are really interested about. So it was a great opportunity. It was a way to put light on this whole community that we love and that we have been following for a long time that we think deserves lots of attention. We think that this community's really great and should have more people know about it.
So at first it was, ‘Yeah, cool, we are going to help them, we are going to make the community even greater.’ And if more people know about it, it could be a great thing. And for us, there was no competition or something. We were just like, ‘Let's do it.’
JB: I think early on the perception was that you had little interest in art and were pushing really hard on selling the work on Twitter, eBay, and other locations. This gave the impression you cared more about selling than making art. So when you showed up at Christie’s, people filled in the blanks that you must have just been good at marketing and tricked Christie’s into selling your work. I think many people are thinking, ‘Here are people who found an existing algorithm and immediately tried to find a way to sell the results.’ How would you respond to that?
HC: So at first we were like, ‘Okay, this project is cool and we want to do that.’ And then we also worked hard on this. So I was beginning my Ph.D., and my friend was working part time in his job and part time doing this project. So we had something to try. So it took us a lot of time. And the other person, the other member here was like full time on the project. So we also needed a way to make this work, and we needed to finance ourselves, to pay our rent and to pay for food. We're like, ‘Okay, if we want to pursue this, we have to find a way to make it work for us.’ So we were like, ‘Okay, if we make paintings, if we make artworks we're going to sell them, and eventually we can bootstrap and we can continue doing this and experimenting with all this stuff.’ Because GANs was just the beginning, it was the first project we were working on. And we think that since this technology and many other algorithms have so much potential, then we want to -- it's just our first project, so we want to make it work. And maybe if we have a little bit of success, we can continue.
And so, yeah, we felt like, ‘Okay, we need this money, so we need also to sell them.’ So we got in touch with all the people that could help us meet galleries or meet art collectors. So at first, people in Paris were just like looking at us, and we went to actual galleries, saying, ‘Okay, we are doing this. What do you think about this?’ And they were like, ‘It's worthless. You can just stop it here. It's not interesting.’
So we pushed and continued to meet people. And then we met with Nicolas Laugero-Lasserre, who was first a bit skeptical, and then interested in what we do. So he was interested, and then, like, ‘Okay, I'm willing to help you. I'm willing to finance you and to buy your piece, and to start this adventure with you. I will put this first artwork in my gallery in Paris.’ So with this, we got a little bit of press coverage. People were starting to take us a bit more seriously. They were like, ‘Okay, this may be interesting.’
So we met a few more people. And then we did a lot of events in Paris, actually. We went to a lot of different cultural events where we presented what we do. We were talking with people presenting GANs, saying, ‘Hey, we are doing this. This is similar technology. You should hear about it. It's really cool,’ and so on and so forth.
And then, it's actually Christie's that Tweeted us saying, ‘Hey, we've done this blockchain event.’ And we were a part of the SuperRare website. So SuperRare was a really good way to work this out. And so when [Christie’s] saw this, they thought, ‘Hey, these portraits could really fit into what we do at Christie's, so let's do it.’ We were like, ‘Okay, sure.’
It happened so fast for us. First, it was really unexpected that so many people could get so excited about this. For us, it was a just a way to create something together, have fun, do something creative, presenting a new technology. We are interested in technology; we are interested in art, also. And so it was really a fun experience.
And now it's beginning to be less and less fun because we have no control over what is happening to us. We can't really control it now, so we are just passive at the moment. We are like in the middle of the storm. We're just waiting for someone like you to just demystify this whole story and say, '‘Okay, this is really the story.’
It's not that strange or malicious. It's just the story of a few friends that just wanted to do something cool together. And it just went out of control because of the way AI art is in the world right now, I guess.
And that's something that really struck me when I went to ICM, the International Conference on Machine Learning. I went to it in Sweden this summer … and I really understood at this point how big the AI bubble was. And so when I saw that, I was like, ‘Okay, this could really go wrong. People are really, really crazy about AI.’ I thought, ‘We are doing this innocently,’ I guess, and now we think that there can be real consequences and real misconceptions.
And some of the comments, we really agree with this. It's like, what we do is not that complicated, or it can be qualified as not really very original compared to what is done in the AI art community. And we totally agree with that, because it was just the first project. It's the first project. You've got to start somewhere. And we started there. And we didn't know that it would be the thing that journalists would be crazy about, and that Christie's would be willing to do something like this. I don't know. It's the choice of Christie's, so I don't want to blame them, because for me, it was a great opportunity. I think that it would have been a great thing to take not one piece, but a few pieces from several artists that would be representative of what the AI arts sphere is right now. But yeah, they wanted to do this, and they want it to go fast. We totally understand why [they wanted] to do this fast, as they have lots and lots of other things to do. But yeah, much of the comments we just agree with. We are just a bit sad that we don't really have a way to talk about it, and I think people are seeing us and believe that we are people that we are not really. We are just, like, three 25-year-old friends that wanted to do something fun.
JB: So it sounds like everybody has to start somewhere. This was was your first project; you were excited to get some funding because that meant you could keep learning and working on the project. When people don’t have the backstory, they assume you were the computer scientist and the other two members must have been marketing full-time.
HC: I think that's really a good point you're making. And I think it's a bit sad, also, that you need to have this art background to have respect for art. I don't really feel like I don't have any respect for art. I've been around art since I was born. In my home there are lots and lots of artworks. I had the chance with my parents to travel the world and to see the greatest museums. I have a lot of respect for art. But because I'm a computer scientist, that doesn't mean that I can’t come to something related to art, also. And also, I think maybe Christie's just didn't know about this AI art sphere? And I think that's something that you feel in the article, too, is one of the first sentences is, I think, ‘With the ongoing talk about Christie's and blockchain, we wanted to do something with AI.’ And so that's kind of where this started.
But we also don't understand why they didn't do something with Robbie Barrat, too, because he was there. He had 300 different nudes at the blockchain event.
JB: I actually worked with Robbie Barrat and SuperRare on the project where we gave out the 300 frames from a single AI nude in London. In terms of whether or not you have to have an art background, I think anyone can make art or become an artist, and that is a good thing. But this is more nuanced than that. For example, if I took someone’s Processing.js code and made a few tweaks and won the Turing Award, you would be offended as a computer scientist. Anyone can be a computer scientist, just like anyone can be an artist, but I’d have no business getting the highest honor in computer science for tweaking some code as an amateur programmer. You know what I mean?
HC: Exactly. Exactly. I feel like the most important people in AI and art, yeah, I really am interested in them. If I'm trying to get in their shoes, I would be surprised, I would be shocked. I really understand.
We do have respect for all this community, and we also hope that this whole event will shed light on all this community and have them bring them more good attention and have them get to where they want. Because we don't feel like it needs to be a competition. This whole event has turned this into a competition, and it's really sad.
But also, it's just a choice. They made a choice and they just have to deal with it. Christie's made the choice, and we just have to roll with it. And hopefully it will bring the best for every one of us.
JB: So we have talked about how you ended up at Christie’s. Let’s now dig into what it is exactly that you do in making your art vs. what the machine does. What aspects do you think are creative where you are making contributions?
HC: Exactly. So we're doing exactly what people working with GANs are doing, which is to curate data sets. I don't know if everyone does this, but we manually go over every piece of data in the set and we want to change it so that we get to the specific data set that we want, and then we train the model. We try to find the best hyperparameters and the best moment to stop training in order to have the best result. So we did that with our project, and with the projects we did on SuperRare, too.
So it's the three of us iterating together, trying out things, trying to find the best result, and using GANs as a tool. So this tool is like -- as we said, lots of time before, GANs are like your paintbrush, but it's like kind of an enhanced paintbrush because it automatically creates images. And so, yes, it's an artist's tool or an artist's medium, if you want, but I would say also … it's not directly comparable to brushes because the whole image created by the tool is kind of interesting, too.
I think also something that is potentially worth noting is that we are all in this community, and we are really used to this tool, but [most people] are not used to this tool. So I think that's what we did… So this part of creativity where there is a small part, but a significant part, of the process where the algorithm creates the images, even though [there are many other] layers, and [without them], there couldn't be any artwork. But this part is also interesting, and it's very important, I think, for the general audience to understand this part -- in order to understand what is AI right now, it is a computer program where at this moment, it can create something automatically. A robot’s not anything like that. But there is something a bit creative by the algorithm, which is pretty interesting, and it is worth noting. We are not saying it's the only part that is interesting of the whole work, but it's worth noting.
So that's why we wanted also to have the formula on the canvas. That's just like a way to show, ‘Hey, it's also pretty interesting.’ Because most of our time, we did it doing events in Paris and meeting people, and we saw how you have to present GANs in order for some people to understand them the most easily. And so with this back and forth discussion with actual people, we said that this was interesting, and that by doing this, with this portrait with the frame and showing the signature, it was really easy for people to understand that, ‘Oh, okay, it's the computer, it's not a robot.’
It's not like the whole portrait is made by machines. There's, like, this particular part that is done by the machine, but the whole rest is done by the artist and is the most important part.
And also … since there is a significant conceptual part to AI art, I guess because of all this AI stuff, it was, like, the most easy way to show these conceptual parts to the general audience. And once again, we totally agree that for people that are really used to it, it's really not that well said, it's not really nicely put. But I guess for other people and for [the] broader audience, it's an easier way to comprehend what GANs are.
JB: So somebody else wrote the GAN code you used, correct? Was it DCGAN that you used for the Belamey painting going to auction at Christies?