 Ddod ydw i'n ddod, Alexander. I'n ddod, mae'n fawr oedd oeddol. A i ddod i'n ddod i gwybod i'ch gweithio'r pwysig o ddod y ddod o'r artistig, yn ei fath! A dyna, yna, mae'n ddod i'n ddod o fy nghyrchu am ysgol, ar hyn ar gyfer fy modd, sy'n ddod o'r cyfrafff honno pethau a'r ddatau anhygoel iawn. Felly, mae'n ddod i'n ddod i'ch gweithio'n rhan o'r arwad. I run a meet-up in London that promotes various artistic projects that use AI in some form in art, music and design. I curate various exhibitions, free community conferences. This is one I did in Cambridge a couple of years ago and then in London last year where we put some printed versions of AI generated works up on easels. Then I also curated a media festival in the Netherlands called Impact, which looked a lot at some of the critical questions around AI and art. In the technical community I am maybe a little bit known for organising the SNPs on ERIPS Creativity Workshop, which I've been doing for the past two years and which I really hope to do again in December. If you're working in machine learning and doing something creative with design, art and music, you're very welcome to submit papers to this workshop, provided it happens again. Last year I made a collection of AI art submitted to this workshop, which is now available at AI Art Online. If you want to know more about what types of artists exist in this space, you can go there and have a look. Now I will ultimately tell you about what artists are currently doing in this space. I always start with Deep Dream. How many of you are familiar with this? Yes, very many hands go up and this is very good. I give a lot of these talks at art conferences and there maybe 10 or 20% of the room put their hand up. It's very nice to show these images to the technical community. If somebody didn't put their hand up and is not so familiar with Deep Dream, this is basically how it works. This is an artist, Mamma Acton, and he has run through this algorithm that emphasises the features and comes out with all these crazy shapes and colours. I think at this point is really when this latest wave of interest in AI art has started. I think it's four years ago now and when this Deep Dream came out into the public, a lot of people were kind of curious because you can see that it's got its own aesthetic which can be considered artistic and interesting even if you're not an art historian. In terms of artists who have been working with this technique, I'd like to mention Daniel and Brozy. He is a computational photographer so he takes lots of photographs and in this case he has incorporated Deep Dream. If you look closely you can kind of see how some of the rocks and kind of the river, they have the deep dream colours and the shapes but you can still recognise that it's seen by the river with houses and trees. The technique isn't too dominating which is what I like about this example because I think with these tools it's too easy to get carried away and make the work all about the technique and in this case it's nice that it's kind of in the background and the realism of the scene kind of still exists somewhere. Then you also probably saw this also around four years ago and that's probably more recently in the talk that Alexander Gay, right? But yeah, style transfer was great for turning various landscape scenes into works that look like they were made by Picasso or Monet or an Impressionist. I think it became really popular with Prisma and I remember when I started going to all these technical conferences and I was telling people that I look at this AI art space, everybody thought that all the artists were doing style transfer because maybe to the general public this was kind of an easy way of making a photograph or an image look very artistic. But if I told this to the art community then they would be shocked because many consider style transfer to be a bit of a pastiche because ultimately with this technique you're converting a photograph into something that looks in the style of another famous artist. You're not really developing or innovating so much because you're trying to replicate what has been done in the past. So the art community has some criticisms of it. But yeah, there are some more interesting examples like Gene Cogan who has rendered Mona Lisa in Cubist, Impressionist and Pointillist ways. And then also he broadened the definition of style to change it into astronomy, Google Maps and calligraphy. And I think this is one of the more interesting ways of using style transfer is to go beyond artistic style and think what else can act as a style. And then Sophia Crespo also has some really cool works where I think she minimized content and somehow basically she got these really cool sea creatures. And yeah, I also like to show you this way because I think it's very aesthetically pleasing. And then came the GAN, so the Generative Adversarial Network. And I think now it's very much synonymous with current visual art and AI practice because yeah, the GAN models have gotten pretty good. And in the process of getting there they have also produced some quite interesting results. And yeah, here is some works by Mario Klingerman that are about two or three years old now. And you can see that they, I think yeah, they're very beautiful artistically. So they might remind you of artists from the 20th century like Egon Schiller or Francis Bacon because they have I guess mashed up faces and sometimes weird limbs. And if you follow the GAN developments or have been experimenting with the technology a lot, you will know that a couple of years ago there were a lot of issues with making sure that the eyes were in the right places and that humans or cats had the right amount of limbs. And this is kind of also you can see how in some ways it's also reflected in the artistic practice of that time. So the person in the middle has a random I think arm that's like sticking out at an odd angle and so on. And Mario Klingerman is a really good artist I think to follow because he's very prolific. And every time there was a new kind of GAN model he would kind of play around with it and also kind of add different layers. And yeah, these are some more images of his work. And this is one from last year when I think GAN's got very realistic and he created some installations that were kind of generating very realistic faces. And I don't know if any of you have made it to the heck Basel kind of museum or exhibition. Has anybody been there? A couple of people. Very good. Yeah, because I went there actually just before this talk and I can mention it now because you're all in Basel and you're here listening about AI art. And if you do become interested in it enough to go to the exhibition then yeah, there's an exhibition around AI art featuring actually somewhere by Mario Klingerman. And there's a mirror that you can go up to and front and it will kind of render your photo based on kind of the data sets or the people that it's kind of seen throughout the exhibition. And yeah, it should be open I think tomorrow and the rest of the week from 12 until 6. So that could be interesting. But yeah, back to the art. Yeah, the GAN's got quite realistic and there are also some artists like Libihini that are interested exactly in the techniques because they can now produce such kind of high quality and realistic results. And of course the media likes to pick up on this kind of controversy with the deep fakes and with all the kind of ethical implications of making people say or do things that they never did. But this is more of a playful kind of version of that. And it's called Eurorevision and it features two politicians, Angela Merkel and Theresa May who were kind of a thing dancing and also speaking some poetry in this kind of Eurovision style. Also maybe to highlight the current state of UK politics. But yeah, this is more a fun example of using these techniques. And Scott Eaton has some very beautiful works that I think render the human form in a very kind of beautifully realistic manner. And yeah, this is also like a short video that shows his process. So he comes more from I think the graphics community and oh dear, I don't know where there is coughing. But yeah, you can see maybe this poor person has been sliced apart and then has to cough. But yeah, this is a good example of I guess what the technique can do in terms of also depicting the human form. And now I will move on to a set of artists who are very much interested in the data part of this whole process. And I like to start with Roman Lipski and he is a Polish artist based in Berlin. And he took this photograph of a night scene in LA and then he proceeded to paint nine different versions of this photograph. And this became his data set. And so he fed this to the machine and it generated kind of some images based on his nine photos. And yeah, these were the images generated. And what Roman Lipski did was he painted some more works in response to those generated images. You can see them here. And then these went into the machine again to come out with this which is quite a bit more abstract. And the colour choices are a bit, well maybe clearly not human selected I think. And then Roman Lipski kind of painted some other works in response. And if you kind of look at these, I think it's quite interesting to see how the artist has developed stylistically through the help of this AI algorithm. And yeah, it's also I think a great example of how you can use these tools even if you don't care so much about making digital works because a lot of artists are also very much interested in continuing to work with traditional materials. And this is a great example of how a painter who's always been I think quite traditional in his choice of medium, he's always painted landscapes. And then he decided to get some inspiration by working with a technical team who kind of trained the AI on his paintings. And then he kind of proceeds to stick with a painting medium. So I think it's a good example. And next, there is a work by Anna Riddler called Fall of the House of Usher. And Anna is very much an artist who is interested in getting in her own data sets. And she watched this film called Fall of the House of Usher. And then she proceeded to make lots of, well, not lots. She made 200 or 300 drawings of black and white ink drawings of this film, which you can kind of see how that worked here. And then she proceeded to train pics to pics on her 200 ink drawings to produce an animation. And yeah, this is one of the versions, one of the early versions of the animation. And if you kind of watch it through, you will notice that there are certain kind of specific, specifics that really link this work to the artist like here in the big face. The algorithm drew the eyebrows and the eyes in the same way because that's how Anna automatically draws those. And then sometimes you have a chair that appears and reappears because to an artist maybe a chair is a background element. It's not really that important so sometimes you can miss it out. And so it's very interesting to watch how the algorithm amplifies the artist's particularities. And like Roman Lipsky, Anna also engaged in a feedback loop with the AI. So the first animation was I think the one I just showed you before. And then Anna drew more ink drawings based on that one, which she then trained pics to pics again to generate the middle video. And then she made drawings of the middle video. And the AI generated this last version on the right. So you can kind of see how it becomes a little bit more blurry because I think each time Anna has less and less to work with. And these are some stills. And Helena Siren is also worth mentioning. And what I really like about Helena Siren is that she also sticks very much with her own data sets. And she works a lot with I think her own drawings and photographs. And also with kind of flower and mosaic themes. And yeah, these are some images of hers that I recently curated for a show in Switzerland. And I think the one on the right, it combines newspaper and photography. And yeah, the other one also, like yes, her works frequently combine different types of mediums that you wouldn't normally kind of see together in one image. And yeah, I definitely recommend you to follow her practice because it's very beautiful in combining these mediums and seeing what beautiful aesthetics AI can generate. And there's also Yegor Kraft who is an artist not working with his own data sets but thinking about the data sets we have from Antiquity because we have so many antique sculptures from that early period that sometimes have the noses that has fallen off through age or so on. Or sometimes it's only a fragment that we find. And he thought, how can we further develop these sculptures? Maybe we can use an AI to kind of generate the remainder of the sculpture based on the fragment. And he did that. I think these are some possible generated versions of these faces. And then he also 3D printed some of the components and I think complemented them to the existing fragment. So these are some examples of the sculptures. And yeah, now I will look at some artists who look a little bit more critically at this topic. And I like to start with this one. It's good that you like this one too. And it was made by Scott Kelly and Ben Folkinghorn. So these are two guys from New Zealand. And yeah, I'm not sure if I should call them artists or good advertising executives because I think, yeah, they work a lot in advertising. And this was, I think this wasn't an artwork in the end, but it's kind of very witty. And yeah, I think it's obvious what it's trying to get at. So this is a national park in New Zealand. And you come there and then there's this billboard that suggests the other national parks you can go to because clearly that's what you want to do when you're in nature. You want to be reminded of technology and you want to be recommended more stuff. And yes, there's another one with the seaside and also one with the slide. And yeah, this one is particularly good because I guess the point of the slide is to slide down. And if you do, you would bang your head on the sign. So it clearly shows how technology can also hinder your enjoyment of natural spaces if used wrongly. And yeah, then there is Gretchen Andrew who calls herself an internet imperialist. And she is very much concerned with kind of what comes out in Google Results and whether you as a human or an artist can influence that. And she's got a series of projects where she picks a set of words like in this case it's Freeze Los Angeles. So it's a big famous art fair that was just kind of starting out in its edition in LA. And she made some paintings which you can see on the images, actually on most images there against the white background. And she put her paintings in this kind of, I think, in a generated gallery space to make them look as if they were being exhibited at a gallery. And then she knows how to really make search engine optimization work for her. So yeah, if you did search Freeze LA at the time of the art fair, this is what would come up. So these kind of like imagined exhibitions of artwork as opposed to what was really happening. And she's done that a lot, I think, looking at also her hometown in Bow New Hampshire in the US and also looking at what comes up, I think, for products for women. So yeah, I think it's quite admirable kind of what she does in trying to change what is displayed by this huge global search engine herself only one person artist. And this is one of her works. And yeah, next I will mention Shiv Integer by Julien Desweff and Matt Plummer Fernandes. And yeah, there's this platform called Thingiverse, which I think you can use if you like 3D designs and printing objects. And these are some examples of the types of models or objects you can find on this thingiverse website. And what these two artists did is they created a bot that made mashups of all these different projects. And these are some of the objects that came out and the titles were also I think bot generated because that is a plastic action car and that is an open overlord nozzle. These are a bit of a, I guess, strange objects. And what I particularly found interesting about this project was the feedback from the community because what this bot was doing, it was kind of creating all these mashups and it would post them onto the website. And I think the website has some sort of a news feed and so it was kind of dominating it. And then you had various sorts of feedback from the community. Some people were not sure if it was a bot or a human that was kind of making all these models. And others were annoyed that it was dominating their news feeds so human projects got less exposure. And then the last one found it strange and was happy that it helped, that this person helped with the model. So, yeah, very different types of feedback. And although, yeah, I don't think this project explicitly uses kind of anything AI related, I always like to show it as an example of what maybe can happen in the future where you will have a lot of this kind of AI generated art fighting for space with artists. And ultimately, some of these crazy shapes were 3D printed and they were exhibited in a gallery in the Netherlands and you could vote whether it was Autospam. But remember, this was in a gallery, right? So you're probably targeting a very biased kind of group of people, right? Who came there because they want to see art so they'll probably say, yes, this definitely was art. Yeah. And now I will look at some artists who think a bit about facial recognition and we can start with Constan de Lart and that is his real name, actually. And the reason why I say this is because this project is called Dull Dream and you probably remember that I started with Deep Dream which was all about emphasising kind of features in an image. So you got these kind of very colourful images at the end and with the project Dull Dream, it's about dulling or reducing the features in an image so they become a bit more blurry. Like you can see this image of protesters that is very blurry on the right and also same with Trump. And I think this project is still live so if you go on dulldream.xyz and upload your own photograph, this will also happen to you. So you will become less recognisable and this is kind of a way for the artist to look at how we can fight facial recognition because of course that is a controversial technique that is used a lot in the public realm and I think a lot of artists look at it very critically. But yeah, Adam Harvey is also one of these artists and he was trying to think of what can you do to your face to stop it from being detected as a face. So he has this project called CV Dazzle and it's I think three or four years old now so I don't think all of these techniques work anymore because of course the technical community has been working hard to make sure they can recognise faces. Anyway, at a certain point in time if you had two triangles on your face then it would be detected. Then the face would not be detected but if you had only one triangle then yes, you would register as a face and also if you had some crazy hairstyle that partly blocked your face that would also prevent the machine from seeing you but nowadays I think a lot of the techniques have improved and the artist has been working with a different type of technique and this project is called Hyperface and it's like a camouflage and I think it tries to blend the face into the background and therefore stop being detected as a face. There are some artists who also look at facial recognition in more of a fine art context and these guys, Shung Sung Baek, Kim Young-Hong two artists based in Seoul they found some painters and asked them to paint portraits which became non-facial portraits so what the painters had to do is they had to paint a portrait of a person together with a facial recognition system and the portrait had to be not detected as a face and you can see the final results in these images and then there's also a video that was showing the process because as soon as the face was detected the artist had to change what they were doing to fool the system and so on and you can see some of the results here and then also here and whenever I look at these examples I think they're particularly curious because the one on the left to me as a human it doesn't really look like a face at all so I can kind of understand the machine there but on the right I think it's very obvious that it's a portrait of a human but I guess the eyes, maybe the nose is missing so all very lightly painted so it's probably much more difficult for a machine to register that and Tom White has also some great work that looks at image recognition and kind of the way that works and he has a series called Perception Engines and he made some prints that when shown to most image recognition models would register in a particular category so I think the one on the left is a starfish and the one on the right is either a cabaret or a brain but I get this wrong because it actually looks like both but the image on the left is a starfish and I think if I asked one of you to draw a starfish you probably wouldn't draw this because a starfish would normally have tentacles that are poking out in different directions but if you wanted to talk to a computer and tell it to recognise a starfish from your image then you would draw that so I think it's just interesting to showcase how different humans and computers can interpret images and Joanne Hastie also looks at this image recognition topic and she made various colourful shapes that were abstract and she got an AI to come up with an arrangement for those images and these are eight different paintings that she made based on what the AI told her to paint her images at and then she ran those paintings through an image recognition model to give them titles and this one became titled Christmas Stocking but you can see some other options were packet, sock, bandaid or pillow so yeah it's maybe a good way to name things maybe not and just to also mention a different technique because a lot of this has been gandominated and then a little bit of facial or image recognition at the end so Harun Fundindorpel is an artist who's been working with genetic algorithms and he was interested in looking at how you can maybe create a supreme artwork and he made some prints of turtle drawings and actually some designs of turtle drawings and then he had two drawings that you picked as parents that then produced lots of different designs that became their children and then you could pick the two new parents from the set of children and then they would create more children so the process would then continue until they all converged into something very similar and then at the end the final artwork was printed and displayed offline so that was also very nice oh yeah sculpture so I don't think there's been so much sculpture in current artistic AI art practice and so that's why I was very happy when there was some work by Ben Snell that went for auction at Philips I think in April or March and what this artist did was he got a computer to generate a design of a sculpture based on a big data set that included I think old sculpture and also some more contemporary and modern designs and then he killed the computer so he kind of mashed up and ground it up into this powder and this is kind of where the sculpture came from so yeah a little bit brutal maybe but I don't know I think it just makes it special on so many different levels because if you look at it you think that it was AI generated even so yeah so it's a really cool project and another recent project called Aida Robot and this one was designed by an art dealer in Oxford who actually always used to sell I think modern art so you know like Picasso and so on and then he decided that AI art is the future so clearly a robot AI artist is also the future so she's been making some works and I think she can also draw portraits of people kind of based on facial recognition and then using some robotics to draw that and yeah it's a curious project that highlights how trendy AI art is now if kind of art dealers are also kind of jumping on the bandwagon I mentioned this already but I seem to have still left a slide of this here so yeah you can go to AI art online and look at more projects and if you are working on something that is relevant I have a workshop in Seoul at ICCV so the international conference on computer vision so the workshop will be on the 2nd of November and we are accepting papers on computer vision for fashion art and design and then also art so if you are working with computer vision techniques to create some sort of art then yeah you should submit by 10th of August and we'll create another online art gallery and there may even be a prize and if you have any questions or anything you can email me or maybe ask in the 5 minutes or so that we have now thank you thank you very much questions please line up on the microphones and while you do let me start with the first question how much of an artist is actually the data set? I think it's always quite difficult to define the artist because art is such a broad definition and it can encompass anything from a Picasso painting to a pair of artworks left on the floor and I guess if you're looking at this generative AI art field that is composed of the data set and then kind of the technique and then the curation and then also somewhere along the way the concept then I think it's one of the stages and it also probably I think it depends right also a lot on the concept and on your technical ability and yes I think it's just one of the components but I think actually an artist might want to have a bias even in the data set actually a bias we try to avoid in data science an artist might want to have so I think that's a nice learning I think Yes it's possible I mean particularly if it's an artist that draws or paints and they have a particular kind of style the way they do things then maybe they want like in Anna Riddler's case to kind of draw the eyebrows and the eyes in the same way because that's what makes their practice special even though in theory eyebrows and eyes should look different Yes sure sure Can you show the AI art gallery? Okay Seems everybody's blown away Oh Danielle, next question Thank you that was a really interesting tour One of the things I noticed is that much of the art seems to be about subverting things or taking things that exist and then putting off this kind of machine learning spin on them so the example of the film the follow the house of Usher or the did you like this we recommend that sign boards so they're playful and they're subversive and they're playing with things that we already know even the example of the artist mashing up his computer to create a sculpture but I didn't see many that were that were not of that kind of art that didn't take this ironic and detached stance I didn't see anything that seems like it will be striking out in a brand new direction in the way that Braque might have done 100 years ago or so or other artists who found a new way of looking at something rather than taking something and playing with it and do you think this is going to come oh wow what a question you ask yeah so I mean others tulips I think that's a positive example the tulips project no I think actually the project I would mention is I mean I really like Deep Dream to me that's kind of the technique that was really kind of unique I mean I haven't researched into kind of psychedelic art or like any any related movement so but certainly I haven't seen much like that in the mainstream but with Deep Dream there was this very sudden aesthetic that was linked to this algorithm and to me that was kind of very creative that was that was new and special I think some of the early guns were interesting too but they did kind of create works that were somehow similar to maybe what 20th century artists did and yeah I guess in some ways it is tricky in this field because if you want to become a contemporary artist then like subverting subverting things or taking a critical look at what's happening in society is what the curators want so that's what they will fund and what they will exhibit and yeah I guess in the final sense or I guess looking at kind of yeah a generative art process if you kind of have the given that a lot of it is based on the data set which is kind of stuff that has already been done so it will create something based on this stuff so I think maybe if you move away from these generative techniques and use other types of techniques then maybe you will get something that is a bit different I don't know Thank you very much Thank you very much Lluber for coming by I think you are around if you have questions just feel free to ask and discuss so yeah thank you thanks again for coming by Thank you very much