Your AI Injection

Painting with AI: Interview with Pindar Van Arman

September 07, 2021 Season 1 Episode 9
Painting with AI: Interview with Pindar Van Arman
Your AI Injection
More Info
Your AI Injection
Painting with AI: Interview with Pindar Van Arman
Sep 07, 2021 Season 1 Episode 9

This week our guest was Pindar Van Arman. Pindar is an established artist who uses AI-driven robotic arms to paint. We talked with Pindar about his creative process and background, and  broke down the AI technology that he uses for his paintings. We also discussed the role of the robot as both a tool and a creative collaborator. 

To look at Pindar's work and learn about his creative and technological process, check out his website here:

Show Notes Transcript

This week our guest was Pindar Van Arman. Pindar is an established artist who uses AI-driven robotic arms to paint. We talked with Pindar about his creative process and background, and  broke down the AI technology that he uses for his paintings. We also discussed the role of the robot as both a tool and a creative collaborator. 

To look at Pindar's work and learn about his creative and technological process, check out his website here:

Automated Transcript

Deep: Hi there I'm Deep Dhillon. Welcome to your AI injection podcast, where we discuss state-of-the-art techniques in artificial intelligence with a focus on how these capabilities are used to transform organizations, making them more efficient, impactful, and successful.

This week, our guests cause somebody I'm really excited about Pindar van Arman. Pindar is an award-winning artist who uses AI to paint. So what does it mean to use AI, to paint? I'm going to just talk about what I'm seeing right now. So I'm seeing robotic arms that have paint brushes on them, and then the arm basically grab some paint and starts putting paint to canvas. And typically the robot's got a camera on it and is making incremental decisions based on things that it's seeing. And so the result is just amazing. So many of you probably know about style transfer techniques or these generative adversarial networks that are quite popular, but to actually see it in paint kind of dripping and all their color and vibrancy is truly taking it to the next level. We're not just looking at it, you know, in pixels on the screen.

Tell us a little bit about, you know, how did this happen? Like, were you a painter before? What's that evolution?

Pindar: Yeah, no, I was always an artist always been an artist, but I always think that everyone is an artist and I was painting and I, and I always painted and I was selling work on eBay, you know, it's just something worry. This is the only way I could figure out how to sell work. And it was actually kind of fun. And then I got involved in robotics and the, in the DARPA grand challenge, which was a contest held by DARPA to like, you know, make a self-driving car. And I didn't really enjoy involved in that.

Deep: Back in 2005.

Pindar: So you're familiar with it.

Deep: Yeah. Yeah, for sure. So did you get a car together with a group and like, yeah.

Pindar: We did pretty well. So we got a group of 12 guys together, guys and girls together, and it was T men SKO. And we went to both the first and the second. And then when we joined team case where the third, the first one, we like, we hit a bank and like we're out at the, like a couple hundred feet. It was horrible. Second one, our car drove 82 miles and we had the fastest time at the split time when we something went wrong and we, we went off road and hit a, I don't want, it actually made one of our team members, very sad. We hit a Joshua tree and then it sort of took out our wheels. And, and then the third one was, you know, it was, it was much more structured, but you know, the one that I was most involved with when was the second one and it was, it was racing desert. It was like one of those things you'll never forget, but you know, I got home and I was just so into AI and like, it was like all the things that could happen and with the possibilities where I was sure that everyone has self-driving cars in a couple of years, but I wanted to keep on doing work with AI, but I didn't have the money or the time or the expertise to make a self-driving car. Right. So I started looking around at like personal things I could do. I was like, oh, I'm just gonna, and I was like, I'm just going to make a, I'm going to make myself a painting assistant, like a super smart printer. And the time, you know, I had a job, I had a growing family and, but painting, you know, even though I was telling me, but I'd take up the whole eight hours, 12 hours on a weekend. I was like, oh man, it'd be so much cooler. If I had a, a very intelligent printer that would actually wield a paint brush and paint most of the painting for me, and then leave me the last hour, you know, turn my 12 hour task into a one-hour task. I was like, then I can do a painting every night, you know, this was going on in my head. And so I started building painting robots. So I would have assistant that like, you know, would do the tedious work and then leave the creative work for me. And I would just like, you know, just walk in, do an hour of art direction and be done with a painting. That was the fantasy a little bit.

Deep: So you kind of started off with this efficiency goal in mind and I'm, I'm curious, like what's the, what was the boring part of the paintings to you, you know, back then, or when you, when you thought about them.

Pindar: I do a lot of portraiture and to do a portrait, right? You have to get the proportions. Perfect. And I always thought it was like, you know, I do plenty of cheating, but one of the things you have to do is either got to get a projector. You got to put a grid down, you know, and like, or you have to be really skilled, which I'm not a, you know, like, you know, you see some savant portrait artists, they just like, look at you once, close their eyes, go into another room and repaint you. I can't do that. I would have to like, get the proportions right. Spend a lot of time getting everything just perfect. It's just, it's a robot gets proportions. Perfect. And we lay down the eyes in the right place, the nose in the right place, the mouse in the right place, you know, when I was doing a portrait and I would just go in and add the finesse. And so it would do the very difficult. I mean, let's, let's tell you that. Cause you talk about, you talked in a kind of, you said, well, I cheated and you know, a lot of people have that perception, but the truth is like, you know, going way back, you know, there was a lot of pretty well-known renowned artists that use like the camera Obscura back in the old days to like paint off because once you've done a, you know, a portrait, the, you know, the painful way from scratch it, it can feel, you know, pains to do that, you know, over and over again. Or you can be like my wife who does portraits all day and she's, you know, maybe not quite the savant you pictured, but pretty close to it. And, and you know, for her, it's all like terrible to cut a corner on anything.

Deep: But I'm, I'm curious, like, how do you think about like, what does that even mean anymore? Like cheating. Cause clearly people have to be asking you, like, what does it mean for you to get robots to pain? And you know, like what's the line of the artist and what, you know, and when do you cross it? I mean, clearly you you're comfortable crossing whatever other people's lines are, but like what's your line and.

Pindar: It would be cheating. It would be cheating people. If I, if I was to use, if I was to make these paintings and then not tell people there was a robot involved, I think that would be like, if you're an artist and you present yourself as more skilled than you actually are, you're cheating now of course, you know, I have, my secret advantage of is, is that when I tell people how I'm cheating, it's, it's actually more impressive than if I wasn't cheating. You know.

Deep: Part of that, it is the art, right? Like so much that, so much of it is fundamental to interpretation of your work.

Pindar: Yeah. You have to know that there's a robot working with me. Otherwise it's one 10th is interesting.

Deep: So the disclosure parts, it makes sense to me, but there's like walk me through like, cause I started thinking a little bit about how your machine must work, but I, I realized for listeners benefit, maybe walk us through a time, a sequence of everything that happens from subject matter selection to like, what is the bot looking at? Like, are they looking at a photo or, or are they looking at an actual subject? And then two, like, you know, then it, it, it generates them an, a priority like concept image or something. And then, and then you're in a feedback loop, everything from like how you dip the, the, how the bot chooses to dip the brush in the pain, to the brush selection. Like just walk us through that. Like, you know, like, like how that is. And maybe wherever there are some forks in the road, you know?

Pindar: Okay. Okay. I, I always, I always wish I was more, I'm going to try and be concise cause I can babble on and on. But this, the, the algorithm that runs my robot and I typically don't name my robots, I just named the project. And right now I'm naming the project. Autonomous is, is as much I'm trying to program the robot to be as creative and autonomous as possible. It's not an artist because you know, it won't be an artist until it's a person, no robot will be an artist until it's a person or, you know, has an identity in self-awareness. But, but I think creativity is possible. So this is 20 it's 15 years of many, many, many AI algorithms put together. And the fundamental thing that's happening with my robot is, is feedback loops. And the feedback loop is you got to think of it in the terms of an artist, not just in terms of, it sounds like a computer science term, but I first heard of it from artists, Paul Clay and Paul Clay said like, you know, every artist when they say that every painter, when they create paintings is in a feedback loop, they step to the canvas, make a couple brush strokes, take a step back, analyze those brush strokes, and use that as feedback on how to make the nest brushstrokes. And they make more brush strokes, and you take a step back, how did it come out? And it's back and forth. And they have these feedback loops every brushstroke, every thousand brushstrokes. And they might, you know, halfway through a canvas, might step back and have a different, a higher level feedback loop. How's this whole thing. So it's, I think what's really important to like really achieve AI is to not run an algorithm and let that algorithm do its thing. It's to like run many, many, many small algorithms and try and analyze how those algorithms are working. And here's where I'll get into a little more detail. So that's the framework.

Deep: Okay. I want to just kinda jump in there for a second. Cause I think that's, there's like an interesting distinction, right, right off the bat here, not everyone is doing the same thing when they take a step back and look at their canvas and the amount of time and the thought process, and maybe even some socialization that goes on when you step back can vary, you know, like, you know, I, I, you know, I, I paint a fair amount myself and I know like, you know, my wife and I are both are both painters and we've, you know, we've done some residencies around and we've always had this, like this rule where someone else can yank the canvas away from the other person, just because, you know, you will like step back and over paint, sometimes Bork something and paintings, like I think painters can sort of appreciate this. That paintings are not just at all about the finished product. Like there's an entire process and evolution and many paid for every painting that winds up being perfect. I would argue this, I don't know, five or 10 or maybe even 50 failed paintings along the way. And sometimes potentially amazing paintings that got painted over a gut gut and got torked up. I just throw that out there to just kind of like, just sort of, sort of throw out the question, like, what are you doing when you're stepping back? You know, like, cause that's actually a very big question. Cause sometimes you're thinking something technical, some artists, you know, have like a thing they're trying to get to and they're trying to steer other artists totally reject that notion of trying to get to a destination that's a priority determined and go and kind of follow where the work's going and seeing it as a duality. I'm curious like for you, how are you thinking about that part?

Pindar: You know, that part from the robots vantage and. I can tell you're a painter and you're painting a very, I don't know how similar your art and painting, but I mean, I could, I totally feel what you're saying is like when you step back, you never know what the reason is. And also, I don't know, one thing I say, let's just jump into this and I'll jump back into your question is like, I always think to myself, you know, like I love that, that you, and your way can pull paintings away from each other. Cause sometimes people ask me like, you know, what's the hardest algorithm is the most difficult algorithm. I, if I could write this algorithm, be great. Cause there's so many times the robots painting way. I was like, that looks great. I should stop it. But then I don't stop it. And it destroys something. And I was like, the greatest algorithm is if this robot can just decide when the painting is done, is it, doesn't ultimately, I'm the person that decides when the painting is done and in December. It's gotta be one of the hardest things. And it's one of the hardest things for the artist. I feel like it's very easy for me to yank other friends of mines paintings off and be like, you got to leave this thing alone. Like I have no problem doing it for someone else's piece, but for my own, there's just this temptation to just keep going and keep going. It's all brown and ugly and you get to start over.

Deep: Yeah. And then you, you know, like we're spoiled with, can, you know, a, what do you call it? Control Z or like using Photoshop. We just want to undo what there's no one doing. There's no one doing, cause you're talking about physical pain. You got to just keep going.

You're listening to Your AI Injection brought to you by is That's Check out our website for more content. Or if you need help injecting AI into your organization.

Pindar: Here's, here's where I got inspired by. Like, I love it that you mentioned is like a bunch of different people do different things. When they sit back to the feedback loop and a technical aspect, maybe someone is like trying to find, you know, follow a very exact line. So they're, they've gone all technical in their brain. I just have to follow this line and get their proportions. Right. And another level of someone's just taking a step exit. Is there enough contrast? You know, another person might take a step back and saying, are the colors, you know, complimentary of each other. Okay. So there's all these different things. And, and it's funny how this is how I approach it. As I was saying, there's about 24, 2 dozen, a couple more, a couple less. It changes algorithms that whenever it takes a step back, they all start working and, and they all give themselves scores. And, and then, you know, and like, what are.

Deep: They? Tell me, tell me what those are.

Pindar: I'll tell you a bunch of them. One is one is, is there a contrast? It just does a quick contrast read. Oh.

Deep: And it's trying to score the like quality of contrast or something like that.

Pindar: If there's, if there's white whites and dark darks and you know, and how I know. So crisp edges between these light lights and dark darks. Cause the human eyes seems to like crisp edges. Okay. This, this is my call. Just like, okay. Another one might be another one is like, are there complimentary colors? I really like it. I've noticed a complimentary colors next to each other pop. So it will give itself a good complimentary color score. You know, another one might be another one might be is the face. You know, like I do a lot of portraiture. So are the eyes symmetrical and the eyes in the right place. And I'll give you the simplest one. The simplest one I have is I learned from my daughter and my, my sons, my young kids is it just looks at the overall composition and says, is there an even balanced composition? Is the composition complete? And something as simple as looking at us like, Hey, there's nothing over there. There's a bunch of stuff happening over here, but there's nothing happening in that corner. I better go add something to that corner. You know? So it's nothing that evens out the composition and, and these, these algorithms fight with each other for like, give me the brush or, you know, here's what I want to do. And, and I even have some animations where, where I've shown the, a painting being made and I show, you know, like windows into six of these algorithms and what they're seeing and what they're fighting for. And then, and I have one simple thing, have two algorithms wanting to do something that's similar. The, the robot will actually go and say, Hey, you know what, both the complimentary color and the high contrast algorithm think that this area over here needs some help. Therefore I'm going to focus on that area. And so these things are like, you know, I always think of, I always describe it as Marvin Minsky is society of minds. You know, he theorizes intelligence, isn't a single intelligent, but our minds is a society. Our mind is a society of mind where we have many, many, many intelligences and some of them just sitting dormant, but when a problem comes up, all these different intelligence attack it and the most efficient intelligence will actually rise up and solve the problem. And in, you know, like, you know, you can almost imagine this as you got a little, you know, Victorian times you have a devil on one shoulder telling you to do one thing and an angel on the other, telling you to do another. And you're in just have these multiple minds trying to push you in different directions and you never know which one went out.

Deep: How does it, how does it win for your 24? I mean, I'm picturing like a normalized score. Yeah. I don't know between zero and one or whatever for each of your different metrics, contrast, et cetera, something has to dictate the next brush stroke.

Pindar: Yeah.

Deep: How does that decision get made?

Pindar: There's a confidence. They each, when they make a decision and this is totally arbitrary in some of this, some of the AI is very, very sophisticated and other it's just like, you know, I fucked a lot of numbers, but they have a confidence is like, I want to change this area. And I'm confident that I want to change this area. That I'll give you the simplest example of the one that, that looks for contrast. It says this area has horrible contrast and I'm very confident. Right. And then, and then, and then it'll, it'll just scream, add contrast to this area. And in the, in the, with a high, with a 95% certainty, I need to do this. Right. And then that might win out because if nothing else is certain of what to do, like, okay, let's go add contrast. And what's cool about this deepest. Not only do I have these, I have weights. I've like imagined 24 levers where when I'm making a painting, I might say, ignore the contrast. And I will take that lever and turn, contrast down and say, I don't care how loud conscious is screaming. The loudest to can scream as 10%. You know.

Deep: Are you there, like when these decisions are being made and can you intercede, like according to your own personal rule set, so you, you do, and are you kind of approving on each brushstroke basis or are you letting it run for a set of brushstrokes or are you messing around with that. More directorial?

Pindar: More like, I'll let it go for a couple hours. And like you, I come back and look at the painting. Is, is it going the direction I want? And then I'll be like, oh my gosh, I'm like, turn off. Oh, I'm going to be like, it's so boring. I'm going to really go tell the GaN generative adversarial network to, to like, just turn everything else down and turn the generative adversarial network up to look at the thing and then start recalculating faces and try and generate a new face, find new face and start painting a new face, you know, and I'll turn everything down. But the GaN, so the GaN runs or, or if the colors are really boring, I might say, I need more complimentary colors and I'll turn everything down. But the complimentary colors, stuff like that. It's funny. Cause there's, that's kind of what I feel like I do when I'm painting. There's like, sometimes you don't know what you're doing. You're just like, you just get in there and like, I'm just going to change, shake things up. Sometimes you get very, like for the style of painting, I do like, I'll start doing like really kind of freeform brush strokes. Sometimes you'll get obsessed with texture. Sometimes you start getting really close.

Deep: Like you feel like you're going somewhere and then you take a, take a break and then you come back with that very directorial mind that you're kind of describing this, like where, where I'll sit down and I'm like, where do I want this thing to go? Cause it could go, you know, and a lot of times there's like a deletion going on, like, because you know, I don't, I don't do portraiture, but you know, I'll paint stuff where there's, you know, a lot of competing potential focal points and you'll start stripping away.

Pindar: I mean, I think the way you're thinking about this feels very similar to at least what's going on in my mind when I'm paintings, I feel like that's an interesting Vantage. I'm trying to model it after I'm trying to teach it to, you know, I'm trying to teach you creativity. So I'm trying to, the only way I know to be creative as how creative I am. So half these algorithms are something I watch myself do, or I see my kids do like that one that I called horror evacuees. Like if there's an empty space, how do you solve that? Put something in it. What do you put, I don't know, go to one of your other algorithms to find that, but put something there. And what's what what's fascinating is those, those are the lower level ones that we haven't even gotten into. The deep learning. The deep learning is.

Deep: That's what I was curious about was like, cause I had a theory of how your stuff worked when I first saw it. But I'd love to hear it.

Pindar: Well. So are one of the things that I thought is is that you would sort of, you take the subject matter and then you know how, you know, you, you you've got style transfer. So I would, and if you think about an artist, you know, like from an era or from a movement, they have like, you know, like anywhere from a handful to 20, 30, 40, maybe different influences. So I'm kind of imagining like, you know, a style that evolves that's in essence, like within your favorites world and you know, at any point in time, and then you take that, that rendering is going to take your, your, your subject let's let's take, you know.

Deep: So you take like a subject and then now you gotta like render it. So I don't know for, for, for, just for our audiences simplicity, let's say, you know, you've got like a ton of Cubist, Picasso, he stuff going on. So now, now you've got the image itself that you're trying to, like, this is like the, the like kind of concept of where the artist wants to get to it's this thing. And that thing doesn't probably change to match. Now you go off and you start making strokes. And you're probably like mid, like, I would think that you're trying to like minimize different, like do the stuff that you're saying, which makes it interesting. But at the same time, you're trying to like minimize differences between that a priori kind of conceived target that you're trying to get to, like in this case like that, you know, like that like cubistic kind of style thing or whatever that style may be Is that anywhere close?

Pindar: Yeah. Yeah. Totally. I mean, like down to the Cubist, so here, let me, let me, let me, it's almost there. Okay. So the first, you know, when I first learned about deep learning, it was like a, was it five years ago? Maybe I never get, sometimes I say I did something sometime and I'm off by a couple of years, but when I first learned about deep learning updated for you, it was right when AlphaGo beat Lisa doll at go. And I started reading the articles and the articles, and I've done, you know, I'd done AI in a done feed. I'd done like, you know, neural networks. But then I started reading stuff. Like some of the moves were creative. And of course I was interested in making creative pianists. Like, what do you mean creative? And everyone was like, no, it's not a gimmick this time. These are actually creative moves. And I started looking into it and as I became convinced, I made it my first deep learning algorithms were style transfer and, and I started messing around with it. And as like, you know what, there's something here. And the first paintings and these paintings got a, you know, like got a lot of attention where were Picasso? I took up, I can't pronounce it. Right. The one with the five prostitutes, I fed that in. And I was like making these gorgeous painting portraits. And the portraits looked like Picasso's. And I was like, oh my gosh, this is like, how, you know, I've, I've, I've gotten this creative part, but like the, the robot slack style, if he can steal Picasso style, you know, what is anyone but a conglomeration of all the artists they love before them. So exactly like you were saying, I got Picasso, I got Saison, I got van Gogh, I got Degas. And I started using those to make a style transfers, but it always irritated me that I didn't want to be a Picasso knockoff, you know? So I would start mixing the style transfers together, you know, I'd run a Picasso then a day Gaulle and then it stays on and the robot would paint them. And then I found this cool thing to do is like, you can, the robot can paint, begin painting of Picasso. Then halfway through, I could switch the style, transfer to a day guy. We're actually more interesting to season. And, and it would be this conglomeration where I was painting like Picasso for a little bit and then started painting the Xs on. And, and it became interesting. It became its own thing. So I did that for a while. And then, and then it got more, this is, that was simple, deep learning. And then it got more to us. Like, you know what, even though it's like, what if, what if it started learning from itself? And so I started like, this is thousands of me. I painted thousands of thousands of portraits is a big exaggerated pain.

Deep: Started feeding its own.

Pindar: Yeah, exactly.

Deep: The actual paintings, like back in as things to style transfer to.

Pindar: Yep. He nailed it, that's it. And then it started developing its own style, which is what you see behind. Well, you can see when you look at my art, I would say at this point, a painted about 2000 portraits of the last 15 years, let's see at this point, a good, a majority of the work that it's using is inspiration. And it's in the neural networks that it builds is its own work. But what does it say about.

Deep: That's fascinating. So like how do you, so, okay. Like this so far, this feels like, you know, I dunno if I was like, you know, like the next Picasso picker and I had to see like, and sort of look at like, you know, the thing that would work. I mean, it does make sense, like an evolving painter has influences. Oftentimes those influences are actual paintings. Although, you know, I'm sure there's certainly other influences that come in, like photography, music, poetry, all kinds of stuff. But let's just say for now they're looking at pain. So they get attracted to a certain set of painters. Then they just start painting. They start evolving, you know, into something that somebody else on the outside calls a genre or something eventually then they start seeing works of their own or elements of works of their own that worked in the past. And they start learning lessons from it. Like maybe the way I popped or maybe the way they had a little paint release sort of thing, going on around a lip or a nose or, you know, or whatever. And then, but they have to like, like it's not so conscious, right? Like, you know, when, when, when, when you're an actual artist, you don't sit around and think like this, but I'm doing. Oh, you don't. I was wondering about that with your painting don't you have like a favorite painting of yours. And you're like, wow, I want to paint something like different, but I really like what I did with that painting.

Pindar: I guess I would, I would, yes, I do do that. And sometimes they do it very overtly. Like I'm just going to blatantly try to like, paint like a master for like a particular master for a while, just as an exercise, but other times. And so that would be like, like very conscious and you know, like energy directed modes of forcing your style or something. But then there's like all this other stuff, like much more subtle stuff that you don't really think about, which is like, I don't know, you go on a traveling trip, you just wander through museums. You know, like one of my favorite museums is the modern art museum in Madrid. And I just, something, you know, will stick in my head forever. Like there's an answer key for a piece that, you know, I remember standing in front of, for like an hour and a half or something, two hours. And like, those things just are sort of a combination of serendipity life experience. Like, I mean, I barely even wound up in that museum, you know, it was like we were on the trip and somehow, and my wife's like, Hey, you know, you might like the third floor. There's like a bunch of abstract expressionists that if you're kind of into that and check it out. And then I, you know, like where to deep go, he disappeared for like hours on it. And like, so it feels like getting back to the bot though, I'm curious, like how do you make that determination of what the bot chooses to like and how it chooses to like it and then how it chooses to incorporate it into its own style. That's my, you know, ultimately that's me, it doesn't know when it's made a good work, right? So ultimately it's me picking from my favorites. It's like, wow, this worked great. This worked great, this worked great. And then I'm making the datasets that it, that I feed into it. And, and it's, it's kinda cool. Cause it's coming up with a look. There's definitely a look and I didn't realize this until my, you know, when I was young, I wish I had, it's taking me so hard to figure art out. And I still am like, not even like scratching the surface, but I wish I realized this past is like, it seems like, you know, most successful artists have a look. You, you, you know, like you can look at a Picasso and you can know it's a Picasso immediately, even if you've never seen that Picasso before, same with Eric. So, you know, and, and I, I sometimes wonder is like, you know, is everyone is, this is I'm going to go on a quick sidetrack, yours. It's almost as if an artist comes up with a generative process to make their own vision that they hone over years and they get very good at this generative process. And they make painting after painting, after painting. Like, and, and, and the public loves it because it's a style that's been honed. It's a very beautiful style and it's a recognizable style. And it's like in the last couple of years, the robot's gone down that road. It's like, since I keep on feeding it, its own work. It's new work looks like its own work. I'm sorry, it's old work. And, and this is appealing to a lot of people. A lot of people have told me, it was like, you know, I love it. I saw this painting come out. I knew immediately it was autonomous. And you, and, and, and so that's, that's interesting because you know, it's made me really question, you know, there's a whole genre of generative art and John or generative art is art made with rules, but some randomness. So, you know, when you, and it could be a computer program that makes generative art, or it could be someone hangs paintbrush from a tree and then lets the paintbrush swing in the wind, doesn't have to be a program. You know, that's generative art, but it's like, I began to wonder, you know, I started applying this idea of generative art and I realized that a lot of the great artists were simply generative artists, but instead of using a computer to execute their algorithm, they're using their brains, execute their algorithms. I think of artists like Pawlik, Pawlik used to drip white paint. Then when the white paint was done, he didn't even evenly across the canvas. You would go get the black paint and he would drink drip that evenly across the canvas thing. He switched back to the white painting just all the time, making sure it was even that just feels like Jaron Navarre to me.

Deep: But. I see him as general.

Pindar: Yeah. Like, you know, politics is a good example because it's, you know, there's like the sequencing, but just the very drip thing is clearly him. And then the, you know, painting horizontally like hunched over with the candidate on the ground. Like a lot of it is kind of like maybe even I, you know, he, I don't know the exact story of how he got so obsessed with the drip, but clearly, you know, he, he takes that direction, but it's almost like he finds a, like a great artist, finds a path to walk down. It feels like, and the path is sort of unique to them and it, and they keep walking down, they might have a few paths, you know, like there's different areas in a, in a, in a masters' life. You know, you can look at, you know, there's definitely like the non Cubist aspects of Picasso's life, for example one.

Deep: So I kind of want to switch gears a little bit in the middle. We can come back and stitch it together, but I want to make sure that we get a chance to like, hear about the robot itself. Like tell me about the, the hardware. Like, you know, do you build the arms? Do you, are you buying some arms? Like, tell me about the brushes. Like just, just, just the actual hardware.

Pindar: The re I'll tell you about the hardware of the robots. The first is that, you know, like the reason I don't like to name the robots is first off, whenever someone names a robot, it's always a bad pun, like, you know, Picasso bot or, or there's actually a polic bot I've seen, you know, there, or, or whatever. And, or, and, and also it's, it's falsely attributing an identity to something that's a machine, which is a tool. So I avoid it. And also the other reason is right now, I always call my, my system autonomous, but I have four different robots running it right now. And my software works on many different robots to tell you about him. When I started, it was, it was X, Y tables, and, and it would take a brush. And I would just like pick up a brush and dip the brush and paint. And the X, Y table would be basically something. I went to X, Y coordinates connected the dots, and then simply a solenoid, which is a magnet that would lift and drop a brush. That's it. So go to an X, Y coordinate, lift, and drop a brush. If it was a paint, well, it would drop the brush into the paint, would lift it back up there and go and draw lines, connect the dots. And it was in the whole system back in the day was connected dots and paint by numbers. You know, I'd give it shapes. And I would say, fill in this shape, connect these lines. Really simple, no AI, even in the very first stuff, very first off, it was a printer, but as it got more and more successful and lucky, I realized that whether or not the codes identical for, for the X, Y tables and my robot arms, because there's a lot of inverse kinematics where someone has done all the math to turn your robotic arm into, you know, you just give it X, Y coordinates, it'll move to those X, Y, and Z in the case of robot, arm C. So first I got, I got some donations from seven bot. They were these small robot arms that were like, you know, had about a foot reach.

Deep: Okay.

Pindar: And then those were, I had three of those when I started messing around with those, but I needed to be longer. So I would actually put those on, I'd put an X, Y table on the canvas or on them, so they can move around and paint. And, and it wasn't that I liked them better than the X, Y tables. It was, it was that people did not. Half of my art is a performance art, right. And they don't respond to a X, Y tables, as much as they respond to robot, arms were overtimes are just so much more visually entertaining and interesting. I've had both, it shows a robot arm next to a table and I've had people and I've had, when the robot arms off in the X, Y tables painting, I've had people say, what's the difference between this and a 3d printer. And.

Deep: They're conceptually it's way less exciting.

Pindar: Yeah. Right. It looks like a 3d printer. It's a 3d printer. What am I? And so, so I've gone the way of robot arms because of the audience to please the audience. If I had my, if I was making these in the cave, what's that.

Deep: And they're also quite powerful, like these arms that you can buy off the shelf now are, I mean, you know, it's everything from putting a rocket together to like, you know, to, to, to holding the brush. I mean, you don't need to kind of reinvent that, that whole part of it. I imagine. Tell me a little bit about the, how you think about the brush. Cause like, when I think about a brush, I think about a bunch of them. And then I also think about pallet knives. And I also think about whatever random craps in my studio that I grabbed and start using, like, God knows what it is. And then, and then, and then, then, you know, there's like the paint mixing and like all that, like how much of that are you like thinking about in the context of the bot? Because I noticed, like, it seems like a lot of the, you know, the, at least the videos that I saw on the site, which I encourage our listeners to go check out for sure, because you're only getting this an audio, if you actually see it, it's a whole other thing. But, you know, I think there were like a handful of brushes and there was a, it was kind of like a, like a drippy or consistency of paint and then it dips and then it goes and paints. And so tell me, tell me how you think about that.

Pindar: All right. I'll take one step back and they'll go right into that. So the arms you're right. You know, the arms are better, not just because of the X, Y two, they're under can handle pallet knives. They're much more agile with the brush. If you imagine a 3d printer, holding a brush and go up and down and, and, and draw lines next. And why, so when you get to the arm, you can do swoops, you can do a, you can take a pallet knife and make like, you know, some very, very specific drawings. And so, so that is one advantage. I just, I just, I just have always been challenged. And here's where I got a weakness and I've seen other, I've seen other artists like other, even other robots do a little better. I have a hard time handling and managing paint. One thing I do is kind of my signature look is most people, they do paint horizontally when they're dealing with robots. I like to paint at a 45 degree angle. Cause I, I like to paint with really liquid paints. And every once in a while too much get pains gets put on a place and start dripping. And that looks like a, you know, some street art and it's like a mistake. And, and the more, the more abstract in the steak in my machines, the paintings look, the less robotic, it looks the less, it looks like it was printed by a printer. But most of the states are also what artists crave. I mean, you know, like artists are craving mistakes, you know, you're, you're aggressively doing stuff. I mean, you know, I'm always spinning the canvas around trying things down and trying to things, you know, and then, yeah, I'd be like, those are the stakes are what you want. I mean, unless you're doing kind of absolute perfect realism, you know, but I feel like that's even realist painters don't really do that much about it as much anymore because so, huh. Yeah. And it gets even cooler with that. It's like not only are the mistakes, what we want is like, you know, I was, you know, we were talking earlier about, is taking pictures constantly. Like what should I do next? If it has a, a priority or it's trying to get to an image and all of a sudden drips appear somewhere that really throws it for a loop. And all of a sudden it has like, it's not executing it. It's just like, it's getting feedback. The feedback is, did some ex it executed some commands, but the commands aren't executing the way it was thought it would be executed. So the visual feedback it's getting puts it into another direction. And I love that serendipity because that's the, because it's not a false, it's not a fake randomness. You know, a lot of these generative systems, I couldn't all randomness is fake, but a lot of generative artists systems the introduce randomness to get variety and variation. This one is not introducing any randomness to get variation.

Deep: The physical world is introducing it. Yes.

Pindar: Yeah. And so then it has to react and it has to, and I think that's so cool.

Deep: But that's it, that's a really essential part of it. Cause when you see the pieces, you see that right away. Because I mean, you could imagine like putting that into your original, you know, upfront digital rendering of what the thing is trying to paint. You could imagine it just putting in drips, but then it's trying to like paint a Priore conceived drips. Like it just feels fake, you know, like artists would never do that or it wouldn't be like, I'm going to put a drip in there. And like, I mean, they might, they might, when they're kind of in that like less analytical mindset where they just start, you know, but I don't, I don't think they would like conceive of a painting at level of granularity of technique manifestations. Otherwise it starts to feel formulaic or Seasonal or something.

Pindar: Yeah. It just, I don't know. Yeah. And what's interesting and you nailed it. It's like nailed it earlier actually is sometimes I walk by the painting and the drips look so cool. I just want to stop. But of course, you know, like the drips have like in one painting, a trip has gone over the eye. And one of the algorithms, I have a lot of algorithms that really focus on the eye because that's essential to portraiture. So I have a lot of algorithms that treat the air. It finds the eye and treats the area around the eye differently than it treats the rest of the painting. So if a drip goes over the eye, you know, an alarm flag goes off. It's like that. I looks messed up, but sometimes I love it because the emotional, this is where I'll interfere with the, with the canvas. There's something emotional in a painting where the eye gets covered. It's like, you know, it's just like the eyes are the windows to our soul. So if you have someone, Half their face, the eye is smeared. That is, that is an evocative emotional portrait. The robot has no idea. It just made an evocative emotional portrait, but it's just messed up with the eye and you mess with the eye, you're messing with the person's soul and you're making a much more interesting portrait than if you're not.

Deep: Yeah, no, I, I, I, I, I was thinking that when you were describing those 24 kind of rules or, or, or objectives, if you will. Cause I, I was thinking about for every one of the ones you would say, I could think of some, an example, painting or artist or something that would intentionally do that.

Pindar: Yeah, no. Then that's where I step in and, and I've even thought to myself, I was like, huh, should I, you know, I've even had this occurred. And as like, it's so cool. And when I get messed up, should I like have some pseudo logic in there or another module that says if one eye's messed up, turn off the eye correction, you know, like just to know it's good. You know, we got some, we got something emotional going on, but that's where I really fail is I really can't find a way, obviously, because if I did, you know, I'd have a person I'd have, I can't find a way to add the emotional element. That's always comes back to me. I don't. And I try constantly to like measure paintings, you know, with like some kind of like, like emotional index.

Deep: Have you ever thought about just like leveraging some, some crowds to like give you some of that emotional metric and then training up some models to in essence, assess the emotions on, let's say various axes. Like you can think of, like, I can think of a number of attributes. There's like kind of the realistic on one side uncanny valley, like contorted on the other, you know, like one on one extreme, you know, you've got, you know, Francis bacon, like where things were like, there couldn't be any more crazy and distorted, which emotionally like you could, you, you could imagine building a, you know, a system pretty straightforwardly, just running some Turkey, experiments, feeding some images, engaging some emotional responses from them, training up some models and then applying it here

Perhaps you're not sure whether AI can really transform your business. Maybe you don't know what it means to inject AI into your business. Maybe you need some help actually building models, check us out at [inaudible] dot com. That's Maybe we can help.

So, one thing I wanted to ask you about is I, for, for years, one of my, I don't know if I'd call it a pet peeve, but one of my observations of industrial arms is they have mastered one type of movement of humans, which is, I would sort of describe it, like when you're maybe three years old learning how to write the English alphabet or an alphabet at that age, you know, you draw like very slow and carefully, but w you know, after you've learned to write for, you know, a while, you know, we follow this much more ballistic motion where it's just like quick, you know, fast, you know, movements. And I don't really, and then there has to be some industrial bots out there that, you know, focus on, on ballistic movements. But I'm curious if you thought about the role of that in art.

Pindar: You're right. And I, and I like how you put that you're right. It's like it is mechanic, but it's almost intentional. So I've held myself back in this place, in this, in this regard. And I'll tell you what I'm in. I'll tell you how I've held myself back and I'll tell you what I was interesting. I've held myself back and like, you'll see a lot of my work. One of the signature things is this crosshatching, you know, it's just like these very straight brush strokes. And that appeals to me because first of all, you know, all these artists are always looking for a signature look. So it's one of the signature things you'll find on my paintings. And if you look at the art, you'll always see very straight brush strokes in at 45 degrees. And the reason why I think this is fascinating gets people's imagination is when they see these paintings. So like, oh, look, this is hand painted, but.

Deep: It feels is.

Pindar: Yeah, this artist has really straight lines and they're perfectly 45 degrees. And it's kind of does this cool thing. And I love that effect.

Deep: I like the honesty of that. Like, that feels really honest to me. It's like, it's like, I can do something that humans can't. I shouldn't go out of my way to like, to make straight lines, you know.

Pindar: But then, you know, but I both always tried to like, think about how could I improve this stroke to be more human and I've lots of tests. And one of the things that's just so hard, it's like you is painted a painter. It's like, you don't realize you do some of these movements until you really study it. Like, if you're drawing, if you have your brush, for example, and there's any painters going to say, oh my gosh, I do that. Or if you haven't noticed you do that, you're going to notice next time is when you're drawing a very straight line and you've got your brush and your brushes wet with paint, you are ever so slightly spinning the brush between your fingers to. Keep that. You know what I'm talking about, for sure. Nice and sharp. You're like almost putting the brush very subtly. I'm talking like a degree. Every, you know, like a degree is second almost just to keep the head of that brush really sharp and a lot's going on there. Your eye is watching that brush, watching how the lines getting laid down, lots going on there, that the robot is not doing a rope. My robots cannot do. You know, they would have to have a camera. They'd have to have feedback. They'd have to see how well the brush was going down, how straight the line was. They'd have to look at the, the, the, the shape of the brush had all of those so much skill. That's just missed out on the, by the robot that a robot arm should be able to do, but it can't do, that'd be an extra spinning access and a lot of feedback.

Deep: Yeah. I mean, it feels almost like, you know, like the S similarly with like an actual, you know, like with the, with the human artist, you know, you've got your toolbox, you know, not like a literal toolbox, but like, you've got your toolbox of techniques independent that evolve, right? So like, you know, there's, there's plenty of painters who maybe, you know, never really spent much time in the palette knife and they start mastering it, they start getting into it. And maybe at some point they're just painting with the palette. And then, you know, there's, there's like, you know, there's, there's, there's folks that, you know, use a lot of mixed media. They start, you know, maybe you have a lot of drawing elements or something. They starting to getting that there's others who maybe never done a portrait before, and they've just done like cityscapes or something. And now all of a sudden they start adding the ability to draw, you know, from portraiture vantage, which causes them to pull in some more tooling, like, you know, figuring out face proportions, you know, nicely figuring out how to deal with the nose, which is, you know, every portraiture is nightmare, por portrait, portrait paintings, painters, nightmare. So I'm curious, like how you, how do you divide your time between giving, you know, the robot more technique and like another tool in its toolbox, if you will, like a palette knife or an understanding of something versus like, you know, continuing to, you know, use what it has, but maybe focused on different subject matter or other stuff.

Pindar: Yeah, that's good. That's, you know, I, whenever I make, I make changes, it's usually in leaps and it's always, there's like 10 things I really want to do. Like I want to make, I wanted it just start changing brushes, but that's like, you know, when things are going good and the pins are coming out good, and I'm pleased with the direction, I just keep on doing, I, you know, let's keep on going deeper and deeper and deeper into this. But then every, so often I challenged myself. There'll be some kind of show and someone will ask me to do something better. Like, for example, I'm doing a collaboration with, [inaudible] an artist on the east coast. And, and she, and we're doing this collaboration where we want to really find a style between ours, but doing a style between artist means I have to get very high res images of a time-lapse with the robot while it's painting right now, it's always taking photos of what it's doing. Right.

Deep: I don't give much attention to those photos being a color.

Pindar: Correct. I just want to know that they're the colors of proximate and that the, and so like right now, I'm doing this big and I'm doing this big change where I'm trying to get the studio and get the lighting. Perfect. And I wouldn't have done that unless I had the challenge of like doing a collaboration with an spotter. And for some reason you would think doing lighting is easy, but you know, when it comes to machine vision, every time I changed the lighting, so I get good pictures, I screw up one of my algorithms, my algorithm gets too sensitive and start seeing too much content or whatever. It just, it's a very sensitive system. So it's usually a fun project I'm doing with someone inspires me to improve it a little. I'm trying to think of some others. I have, I have a, another project where I'm working with the children's hospital and the children's hospital. The kids are so demanding. They, they want, they can take a tablet and penal onside with the robot and the robot will, will mimic what the kids are doing. But the kids, you know, they have all these, they, they, they just, you know, they have, their imaginations is awesome. And they keep on asking for a big, big improvements, which I'm trying to keep up with them. And the improvements will be like, I want to be able to draw a thin line and a dark line. They want to, you know, mix colors better. And, and, and so that's pushing some developments, but I don't know. I guess I rambled a little, but I don't know when and why I improved stuff. It's usually from external forces. We're big project.

Deep: Kind of made me think, like, when you were describing your collaboration, it, it sorta made me think of printmakers. I don't know if you're familiar with like, printmakers, like lithographer ERs, or, you know, other types of prints, but, you know, historically, like, you know, the printmaker from the craft side, or I shouldn't say that, but like the, like when they're not acting like the, the, the artists, but they're in collaboration with artists. So like, you know, like a Ken Tyler collaborating with a Picasso to put together a piece, the, you know, the print side, the printmaking side, it's not like just the laser printer and a poster like, you know, it's, it's, it's very like a very intensive, high, high, skilled process for them to collaborate and like work with the artists to realize the vision and almost curious, like, do you see like, what's the future of what you're working on? Cause it feels like one potential future that it feels like you're mostly doing is you're like, you know, the printmaker, who's not working with an external artist, but actually doing their own work. And, you know, just happens to like run a, a set of, you know, nine or 10 or 10 or however big the edition size is. But there's another side where, you know, some, some printmakers are less focused on the, on their art and their personal art and more on the, like enabling of the artist side. And wondering if there's like a worldview of the robots, that's similar because it's not like if you're not really into this, I don't think he would get very far, you know, with a system as complex as yours as a, as an artist. Like if all you care about is the art, but I'm curious if you see maybe an emerging world that way.

Pindar: Yeah. There's so many answers. There's so many answers. A lot of things you're saying is with the collaborations. It's interesting because you know, there's been some very famous artists that have gotten in touch with me and they want to collaborate. And, but I quickly realized that they don't really want to collaborate with me. They want to use my robot as their printer. And, and it gets into these awkward situations where all of a sudden it's like, well, you know, I see what you're trying to do, but it's not really a collaboration. You know, that's, you're trying to use my printer. My robot is a printer and, and it always gets to this awkward situation for me because me called my artists ego go, I don't want to be someone else's printer. You know, like I want, if you're going to collaborate with me, I'll paint with you and we'll come up with something that, you know, we both work on, but it's not going to be your art peanut by my machine. And so I, so it's, it's, it's almost, but like when I have really good collaboration is like the one I'm really enjoying with Ann spotter right now. It's more of like, Hey, I really dig your style. You like mine, let's try and find a way that someone looks at it and does a double-take it's like, wait a minute. Is this a Pindar van Arman or an ANZ falter? Because I see a little of each in there, that's the ultimate collaboration. And, and so like, you know, when I, when I approach it, that's interesting. Like, I didn't even think about that from a printmaker because, you know, sometimes the big artists like Picasso going in with a printmaker, I'm sure they would want the printmaker. Maybe not to be anonymous, you know? And it technically.

Deep: Yeah. I mean like my, my wife was a printmaker for years and, you know, nothing gets her goat more than somebody calling a poster, a print, you know, like, and you get all of her friends together and they just lose their shit completely. But like, but, but yeah, I mean, some, some do like, you know, she's, you know, she's, she's printed a few, you know, like worked with [inaudible] for example, the principal. And, you know, in those cases, it's very much like, you know, the big name artist wants to do their thing, but I think, I don't know the details, but I'm pretty sure they always have the, the, the studio, the print studio is always on there. And, and, and most printmakers, like similar, I guess, to like, glassblowers are like this too, but there's, there's the part where they're enabling somebody else, but then that same artist has their own work. And there's some kind of like wall where they'll do both, you know, usually because they need to pay the bills, you know, like, and, and so they, so yeah, if somebody thinks of it as using your bots as a printer, then, you know, the printmaker would just be like, you know, piss off. I'm not going to do that. But if it's, if they're thinking, but the reality is like, at least with respect to printmaking is that the collaboration part is not really optional, like to do it well, requires the two to stand in the room together to stand at the layer level, to like have feet, you know, to like really work closely to construct the piece and maybe, you know, the artist, like the actual master or artists or whoever isn't there the whole time. But it's very much, I think if it's done well, it's almost always properly seen as a collaboration, but it's not quite the collaboration that you're describing, you know, where it's, it's, that's like, you know, like a, like a true, I don't know.

Pindar: Yeah. Those are the ones I'm interested in. I'm definitely. And you know, and this is, this is what I've learned. Like everything, this is what I've learned. I just did a piece called a collaboration with past present and future. I realized that, you know, I'm collaborating with every artist in my dataset, you know, like, you know, if I have my say, I mentioned earlier, Suzanne Picasso, and I'm the guy, you know, I'm collaborating with them in their work. I'm collaborating with all the art, all the code that I've, I've swiped, you know.

That's an interesting way to look at it. Yes. Like the algorithm itself is collaborating when you're doing style transfer, you know, when you're, when you're like taking multiple seed images and like for sure. Yeah. And then there's like different levels of collaboration. Now, I'm almost thinking you use your robots in conjunction with another artist to do a collaboration piece. And now you take that and put it back into the bot for future work, Ends up being really cool to see like where those lines are. And even, and even like, you know, like I go into GitHub and I grabbed half my deep learning algorithms from there. And I, I, I, you know, I make a mine and I modify them and I make them, like, I tune them to my stuff, but man, I'm, I'm collaborating with the, the, the engineer that came up with that deep learning, you know, with that neural network. For sale. And then, and then what would be the fun. And then of course, I'm collaborating with the robot going back and forth, seeing how it's doing and it coming back to me. And then here's the, here's the other one I realized just this is, this is outside of robotics. I realized that also every portrait I do is a very, very intimate collaboration with the subject. The subject will make or break a portrait, how open they are, how much, how honest they are, how, how much they cooperate and, and like, you know, especially I found that on COVID times a lot of these portraits, I'm very, very strict about the sources of where, where I get my portraits. I won't paint other photographers portraits, for example, even, you know, if someone comes up to me and says, oh, I had this professionally photography photograph, I own all the rights to this photo. I'm like, yeah, but you know what, the photographer that came to your house or met, you took those photos. And I don't feel comfortable taking his vision. Now we have to get our own photos. And if we're going to work from photos.

Deep: How do you do that? Yeah. How do you choose subject matter? And do you, do you always work from photos and do you shoot them? And like, how does that, how do you.

Pindar: I try and shoot them? I can't always shoot them. I'm, you know, I act a little high-end high and mighty. And on this high horses, I can't always shoot the photos. So, but I definitely make sure that if someone's taken an artistic photo of someone that's off limits, you know, but if someone's taking a headshot, a professional photo, and I might use that as part of my GaN, but I don't want to paint a headshot. That's boring. Right. But, you know, I definitely don't, I don't mind using a bunch of this is interesting. You haven't even made me, I haven't thought about it too much detail. I love, you know, if someone wants to distorted face, I love getting like 10 of their photos together and throwing their photos into a GaN. And I have some tricks that, you know, gains typically need thousands of, of images, but there's some tricks where you can get around with just using 10, you know, you can like, it's, they're actually interesting tricks. You know, you flip them, you have rotate them a little, you, you, you, you, you can, you can make a good data set out of just a couple of dozen photos. And the other thing is I have like, you know, I have all these online galleries where I like have the instructions. Here are the instructions. I need you to take a photo of yourself or have a friend take a photo of you straight on. And then from like 10 different angles. And, and I'll have like, you know, and I just, I just wait for people to get me a lot of images. And once I have a lot of images, I can run with it. I always loved the idea of building a photo booth and having someone like in the photo booth, I haven't done this, but like, this is how I described. I want, I want a bunch of photos of you making different faces from different angles, from different lighting. And then I'll throw it into my AI and my AI, or do something with it. And then that's that's. And then the other thing I do is I collaborate with, with portrait photographers. And so there's a very, very good portrait photographer. I love for art, kitty Simpson. And, and if you go to our autonomous, you'll see my collaborative work with her. And I just try and get my robot to paint her photos as emotionally as possible with as much. And then she gives me feedback and we changed the algorithm and I have a whole separate algorithm just to deal with, with her photography. It's so beautiful. So, and people on the people listening to this can't, can't see it, but you can, you can see it deepens the pictures behind me on the left. And like I said, you'll see them on Awesome.

Deep: All right. Well, I, I want to, I think we're out of time here. It's been, it's been totally awesome talking and, and just catching up on your work. I mean, it's, it's, it's just super exciting for me, you know, just being a machine learning guy and also being, you know, independent for a while. So thanks so much for coming on. It's been great.

That is all for this episode of your AI injection, as always. Thanks for tuning in, you can find more about Pindar's work and his creative That's C L O U D P A I N T E R.

Dot that's all for this episode, I'm Deep Dhillon, your host saying, check back soon for your next AI injection. In the meantime, if you need help injecting AI into your business, reach out to us at xyonix dot com. That's X Y O N I Whether it's text, audio, video, or other business data, we help all kinds of organizations like yours automatically find an operationalized transformative insights.