In a landscape where everything seems automated, simulated, augmented, glitched, or ai generated, how will we users be able to identify any humanity in a sea of robots? We’ve all seen a tweet or a discord server become overrun with bot replies, like ants ruining a picnic. We know there are folks with 30+ profiles to game whitelist mechanics or, worse, scam people. Mr Musk can tell you how difficult it is to determine which profiles are and are not bot operated. Yes, it is wise these days to trust no one and nothing until we have done our due diligence and formed our own conclusions.
Since August 2021, I have been asking @sqrtofpi questions. The following interview has been pieced together from our many exchanges over the last 15 months. This interview and accompanying painting is my portrait of Pi. It is art as CAPTCHA, and I hope it will deem us both as ‘human’.
JL: I really could ask a million questions.
Pi: Feel free. I may not answer immediately, but I will answer.
JL: Perfect. That's just my style.
JL: I am a portrait painter. I have traditionally painted from live models in my studio, but I am having fun adjusting to the metaverse. What do you mean when you say “model”?
Pi: For the types of AI/ML (machine learning) that I use, you're creating computer models that learn how to create different art styles. They learn this through training, during which you give them hundreds or thousands of images in the style you want them to learn. Over the course of 20 to 80 hours, it learns to generate that style of art. After that, it can create images in that style that no one has ever seen before.
For my genesis, I have one model that I've trained on nude paintings. For my second piece, I used a model I trained on forest images. And they can continue learning on similar topics very, very quickly. This is called transfer learning. For instance, the left side of ‘Winter, the longest season’ is the forest model.
And the right side is a winter forest model I created by having the regular forest model learn from about 60 pictures of snowy trees. It took a much shorter amount of time because it didn't need to learn the concept of a forest from scratch. It just needed to learn the concept of winter in the context of a forest.
JL: What did you use AI/ML for before making art?
Pi: I did work in a field that used machine learning to make predictions on large datasets. I also did machine learning projects for fun. It's such an interesting area of research.
JL: So, walk me through your creative process. Do you have an idea and then start training your AI?
Pi: Sure, I have an idea, and the first thing I do is find images to train the model with. I use exclusively public domain images. After collecting anywhere from 300 to 1500 images, I give those to the model to train for 20-80 hours, basically for most of the week. While that's happening, I write the image editing code I'll use to create the final piece. I write almost all of my image-editing code. I use an open-source image editing tool from time to time to do some stuff that's easier by hand, like certain cropping around the curved parts of an image, but otherwise, about 90% of the time, I'm writing my code to do it.
Pi: While this happens, I'm checking the model's outputs as it’s training to track how well it's doing. Once that's done, I combine everything, and the first phase is done. Once it’s all stitched together, there are always more changes that I want to make, so I'm writing more code to do those as well. Finally, if it's a video piece, I render the model outputs along with my image edits, frame by frame, and stitch those together into a video. The whole process takes at least a week, usually longer.
JL: Do you have an arts background?
Pi: Not a traditional arts background, no. But I've always felt creative. Coding can feel static and rigid, but it requires many of the same creative skills. The downtime that I've had during the pandemic has allowed me to channel those more into arts.
JL: Whose GAN art do you admire?
JL: I am impressed by the amount of skill that GAN art requires to be executed with proficiency. Do you think it is difficult to do?
Pi: Yes, there is a bit of a barrier to entry which is learning to use the tools and understand what they're doing. Then it's a matter of experience and going from entry-level to a level of proficiency.
JL: Do you have a day job?
Pi: I do some programming work. Unrelated to any of the art I create.
JL: Sounds vague but interesting.
Pi: Haha, you could say it's intentionally vague. I've purposely chosen to use a pseudonym for my artwork. I find the idea of being able to verify who I am based on just a wallet address to be appealing. And who I am in the metaverse is sqrtofpi.
JL: I respect anonymity, what can I know about you?
Pi: I'm based in the Eastern USA, and I fall into that 25-34 age range that is usually on forms, haha.
JL: Can I know gender?
Pi: Yeah, I identify as male. Also, I'm Black.
JL: As an anon, why hold on to those identities (black American, male) instead of just going full anon?
Pi: I thought about that a lot when I made this pseudonym. The decision to be anon while still identifying as black came down to asking myself what a younger version of myself would have found to be beneficial, and that was to have an example of someone else that looks like them/could relate to them doing this. There is terrible representation for certain populations in all aspects of tech. I was hiring for a role once and interviewed 50-60 candidates and did not see a single black candidate. I've worked in tech organizations with hundreds of people where I could count the number of black co-workers I had on one hand with fingers to spare. So I decided to have that part of my identity visible to anybody it might benefit.
JL: Do you think anonymity will change the way we, as a society, think about diversity?
Pi: Maybe. There are still differences in how/when people are on-boarded to spaces that can have an impact. Crypto is a riskier space, and those with more wealth can afford to take more risks without affecting their ability to meet their basic needs of food, shelter, etc. That said, I think pseudonymity, where a person's actions, ideas, etc, are tied to a continuous entity, brings us closer to a state where you are judged by what you do rather than who you are.
JL: Do you have a comfortable lifestyle, or are things a struggle? What do you hope to accomplish with NFTs?
Pi: I live comfortably enough. I'm learning to be grateful for what I have. What I hope to accomplish with NFTs is multi-faceted. One is personal growth. I'm growing as an artist, and it excites me to finish a piece and put it out into the world. I'm going to continue chasing that feeling. Another is the act of creating something permanent. I think part of the human condition is a constant search for how to create something that will outlast you. Unless something happens to the Ethereum blockchain, the NFTs I create will outlast me.
JL: Do you make other art? Like music, cooking, writing?
Pi: I like that you included cooking. Making food for the people you love can be an art form. I wouldn't always call what I make in the kitchen art, though, haha. I spend a lot of time accidentally figuring out what flavours don't work together, but I like to experiment. I haven't tried my hand at music or writing, though writing has always amazed me. I think art can be broadly defined as anything someone has put the work into master.
JL: Okay, let’s fill in some fun gaps. Are you athletic? Do you eat gluten? Can you sing? Which of your siblings is your favourite and why? Does your mom like your art?
Pi: Athletic, yes, but I'm definitely a bit heavier and slower post-pandemic. I do eat gluten. I can sing in the shower, and that's about it. I have one sibling, so by default, they are my favourite, haha. My parents have actually not seen my art. I've kept that aspect of my life almost completely separate from everything else.
JL: This is interesting, the compartmentalization. Do you show your art to your IRL friends? Or is it just for you and the internet?
Pi: The only person who's seen it is my SO. Other than that, it's just the internet. I think the compartmentalization of it helps me disconnect my sense of worth from however well I feel like my artistic pursuits are going, for better or worse.
JL: So, only one person knows Clark Kent is Superman. That’s romantic. What is your workstation like? Something about you makes me think you are meticulously tidy.
Pi: I am meticulous about some things, but my workstation is not one of them, haha. I clean my desk periodically, but it tends to gather a lot of scrap paper with jotted-down ideas and becomes a layered record of things I've tried. Mail and books tend to gather there too. After a couple of months, it starts encroaching on my ability to work at my desk, and I do a purge, and the process starts over.
JL: Do you listen to music when you create?
Pi: I do. I'll listen to the same album over and over while I work on a specific project. If I'm just exploring ideas, I'll have a playlist going.
JL: Why the repetitiveness?
Pi: I think it keeps me in the same mental state. Kinda how hearing a song from a particular period of your life reminds you of it.
JL: What album was on repeat when making Framed Growth?
Pi: For Framed Growth, I listened to Camp by Childish Gambino.
JL: What movie star do you get told you look most like?
Pi: I'm not really sure, but I do get told that I look like the NBA player Chris Paul fairly often.
JL: What are you working on?
Pi: I'm just working on some base-level code that could apply to multiple pieces in the future. I've gone back to fundamentals during this lull. Like different ways of doing stippling or different ways to interact with a curve using code. It’s fun until the dip keeps dipping.
JL: During this dip/lull, I am just chilling, trying not to look at the charts.
Pi: Same. The intrinsic value for me of the things that I'm working on hasn't changed. It still holds value as a creative outlet.
JL: How does your coding art differ in the process from the ai stuff?
Pi: Unlike the AI art, for my code-based generative art, I have as much or as little control over the outcome as I want. With that type of generative art, I can control where I want lines and shapes to be, their colour, their interactions, etc. For the most part, though, I cede some of the control to randomness. I focus more on designing a system that can produce variations of a central idea.
JL: What questions do you try to answer with your code-based generative work?
Pi: I think the main question I'm trying to answer a lot of times is, "how do I use a rule-based system to make something that looks natural?" For example, this piece:
Pi: Everything on that image is created directly from code I wrote. The background is just a standard off-white color, but I added code to darken areas of it in a noisy way.
JL: I don’t appreciate the process fully, I think.
Pi: Okay, maybe this will help. Here is the same image with randomness removed.
JL: How do you remove randomness?
Pi: In the nonrandom image, you can see the basic structure of the piece is there in terms of where lines are placed. This looks fairly static and computer-generated because it's so perfect. The lines are exactly straight, the background is exactly uniform. The real world is not like that. There are imperfections everywhere, mostly due to small random processes. So, to make that image look more natural, I add randomness in selective ways the background is just a standard off-white color, but when I add code to darken areas of it in a noisy/random way, it looks more natural and paper-like for the lines, I add a few layers of randomness:
The path should have small variations, so it's not exactly straight.
The weight of the line should vary in thickness as the line is being drawn
The line should skip in and out randomly, similar to a pen that's running slightly low on ink.
So when I say that I'm removing the randomness, I removed all those variations I added.
JL: Can I see the code?
Pi: Sure, this code says, at each step, to vary the weight slightly between 25% and 100% of a target line weight I've chosen.
weight_offset = map(noise(i*0.05),0,1,0.25*weight,(weight)) input_pg.strokeWeight(weight_offset)
JL: Do you think your work is more targeted to those who code?
Pi: I think it's just targeted to anyone who also finds it interesting. The things I post or release are works that I made that I like. Someone with an experience in code-based art may have a deeper appreciation of the process that went into it, but I don't view that as a requirement to enjoying it as a viewer.
JL: When did you decide you wanted to be a dev?
Pi: About 7 years ago. I was working in a career that was technical but not very creative. I picked up coding as a hobby and found that it scratched that itch that I had in the middle of the Venn diagram of creative and technical. I realized that I was excited to leave work every day and come home and work on my projects. So I decided to make it more of a full-time thing.
JL: So, I am not a techy person at all. I am more of a big-sky thinker. Can you explain a time when you had to problem-solve creatively? What was the problem? How did you go about solving it?
Pi: There isn't any big problem I can think of now. It's more little things that pose issues and need solutions. For example, I was working on a contract for a project last year, and the artist asked for the ability to have minters select specific tokens and then, later on, be able to release the remaining tokens, in order, in a public sale. It wasn't anything I'd done or seen done before, but it required a good amount of trial and error and trying new ideas. My creative process is best categorized simply as persistence. It involves a lot of making things that don't work and continuously pivoting and trying new things until I find something that does work.
JL: From working with you on @Forms, I have found you to be a clear communicator. Surely, you must write?
Pi: I write for myself in terms of journaling and getting my thoughts on a page and out of my head. I haven't written anything that other people would read. It would be fun to write a fiction story, and I'm sure I'll try my hand at it sometime in the future. And I'm glad to hear that you think I'm a good communicator. I find text to be easier for that since you can self-edit before you send. Now I'm thinking about what the in-person speech version of texting would be. Something like text messaging where the words show up immediately on the other person's phone as you type, and there is no backspace/delete ability.
JL: Do you enjoy playing video games or watching television?
Pi: I play Zelda Breath of the Wild sometimes, and some sports games like NBA 2k or FIFA. Other than that, I don't game a lot. I do watch a lot of television/movies. I just watched Big House and Patriot.
JL: Do you wear glasses?
Pi: Yeah, tortoise shells.
Hey JL, I had an idea and wanted to bounce it off you. You and I have talked about being anon in this space and how different people approach it. There is part of me that doesn't want to be the Pi symbol forever, but I still want to stay anon. I'm considering rotating through a series of AI-generated PFPs using a description of myself fed into an image generator such as stable diffusion or DALLE. I'd include a notice that all of my images are generated as well. What are your first reactions to that?
JL: Wow. I have a lot of thoughts. First, I look at DALLE portraits, and I do feel like I get more of a sense of a person. But I know it’s not really you, and that’s scary because people will assume it IS you. Which is cool but then why say you are anon? I mean, this isn’t so different from those filters on IG. It will help people feel more connected to you. Do you want that?
Pi: I'm hoping to feel a bit more human than fully Anon while still being Anon. There is a piece of it that I find interesting, which is that this is just an algorithmic transformation of a text description of me. The painting you're working on follows a similar pattern in that it's a painting of your interpretation of my text description. Both are based on the same information and source, so are they just different representations of me?
JL: I love that question. I want to believe what I do is very different from DALLE, but AI is probably just faster, and I can’t add or remove randomness. I have to just roll with it. Do you think I am naive for thinking we can still be ourselves in this new world?
Pi: I don't think it's naive at all. When I think about what I see from you online between personal posts about travel, posts about art, memories of Jane, etc. I think it gives a much more complete online version of a person and feels real.
JL: What do you want people to feel when they look at your profile?
Pi: I'm having a hard time coming up with an answer. I guess a sense of my progression with my art.
JL: So, what do you think?
Pi: You want to hear a really funny coincidence? I don't normally wear a jacket over a hoodie, but I did yesterday and thought, "this is a nice combination".