Parameters of Whimsy

Simpson Myers

Interview by L. Valena
January 21, 2023

Can you start by describing the prompt that you responded to?

The prompt was a sculptural piece made of reclaimed materials, and it was very bright and colorful. There were lots of different aspects to it, a lot of movement and texture, a lot of joy and vibrancy. It reminded me of a coral reef or a school of tropical fish.

Where did you go from there?

I really wanted to recreate the feeling the work gave me, rather than visually replicating it. I work in multiple mediums. I do embroidery, painting, and I'm just moving into emerging tech such as machine learning and artificial intelligence. With the machine learning visualizations, I had never done one that complex, with so many different shapes and colors. But I thought that if I could get a visualization out of it that worked, it would possibly bring that same vibe -- that whimsy, lightness, ocean vibe to it. I have a favorite algorithm that I use for most of my visualizations, which is from a lot of trial and error.

Is that like a secret recipe? Having a favorite algorithm?

Yeah! It's funny, when you're using this tool for an artistic visualization, in the end it's not necessarily what's going to be the one that's the most 'accurate' to the input. It needs accuracy, but it also needs something else, because you get a very static, boring image at the end if it's just the accurate input. You want it to have a bit of a challenge in its own processing, so that it can try some new stuff and take some new leaps. That's where you really end up getting a visual output. For me, it's not necessarily the most accurate algorithm, it's the one that produces work that is the most aesthetically pleasing to me. This kind of movement where you can see the visual leaps that it's taking -- that kind of thing.

Because this was a different project, I wanted to try the other algorithmic things as well. So I began this whole process in which I started with one set of inputs and tried a bunch of different algorithms, different loss-activation functions, which is basically how it decides if it's making a correct choice. I got a very mixed bag of results and ended up going back to my preferred algorithm in the end. There was one that nearly knocked it off the post.

My favorite loss-activation is mean squared error, which I find gives it almost an angular, geometric quality, but there's a lot of movement in it as well. You'll see directional movement from the inputs. It starts to lay out all of the inputs at once, and then it slowly builds on that. You can see it coming up on the page like that.

The one that nearly beat it is called Huber. It really starts with the majority input first. So things come up differently on the screen. If there's more green, it will show the green first and then slowly bring the other colors in based on their input weighting. That was really interesting visually, especially at the beginning. But it didn't really take it all the way through to a nice finish. The original one takes it through to this really interesting place, so that's why I made that choice.

I did around fifteen hours of training on the inputs. The video actually shows around two thousand iterations of it trying to learn. That's what we're actually seeing in the video. The beginning to the end is the machine trying to replicate my inputs around two thousand times. What we're watching is its learning process of putting down something, assessing if it was correct, and then making another choice. But that's happening every second.

Wow, that is fascinating. How did you get into this?

It's a little bit meandering. I do writing as well as visual art, and all of my work is centered on the experience of self. Not just my 'self,' but the human 'self'. What makes us unique, and what makes us the same. I recently did a project with Wunder Gym and Science Gallery in Melbourne, where one of the possible prompts we could respond to was connected to AI. There was this video that I loved from around ten years ago, where they plug in these two chatbots to talk to each other, and they have this mind blowing existential conversation. The first one asks, "Do you believe in God?" And they just go straight into it.

To me, it just raised so many interesting questions. It made me realize that AI and machine learning are an extension of understanding the human self, on two prongs. On one hand, our understanding of the self, experience, and knowledge does affect our approach to how we're trying to build these things. Even the neural network itself is based on the human brain. So we have a human-experience focus when it comes to creating these intelligent machines. Then on another level, it raises the question of what it can tell us about our real humanity. There are always going to be limits to the machine. So then what's left over? Whatever is left, past that limit, is humanity.

To me it made sense as an extension of my practice. Actually, it really was the cornerstone piece that helped me tie all the different bits together. Ultimately, what I'm trying to do is create an AI version of myself as an art project. My approach to that has been to look into machine learning, so I've learned coding.

I'm trying to build these two machine-learning frameworks. One takes all my data, which I've been collecting for three years: journals, habit trackers, period tracker, sleep tracker, emails, everything. I'm trying to put this data into a machine-learning context that should be able to look at that and draw out the connections and correlations that I may not see as a human being. The other project is a decision-making framework. A machine-learning program that can go through a decision as if I were doing it myself, but maybe better than me.

Eventually, I want to be able to put these two together and say, "Yes, that's my AI best self." It's just an experiment. What can it do? Maybe it will turn out to be my AI worst self, we don't know. I think there's a lot to be explored there. Can we even do a new kind of self-portrait with that kind of approach? That's the kind of stuff that really fascinates me. This visualization is a part of my learning process. I'm learning the capabilities of it, how to manipulate it and get outcomes from these technologies. I think it's really interesting that, in this piece, we're seeing not only a representation of the machine’s learning, but also a representation of my learning. This isn't my end goal, but it's a part of the process.

I love the language you're using when you're talking about this. It really sounds like a collaborative relationship that you're building with this machine.

I think it really is collaborative and it offers a really interesting look back on ourselves. It works in a totally different way, and it's brought up a lot of different questions for me about how we even view consciousness. How do we view the self? We understand how we view consciousness and sentience through the lens of a biological being. We even have glimmers of being able to see and acknowledge that in other biological beings. As time goes on, we're going to understand that all biological beings have a kind of sentience. We'll get there I think. It raises this really interesting question for me. When we're looking at machine sentience or consciousness, are we wrong to look at it from a biological perspective when we struggle so hard to see it in other biological organisms? The fact remains that the whole way of seeing the world from the perspective of a non-biological entity is so different. It's something that, even if it existed, we may not be able to recognize. So it's very interesting ethically as well.

How could we? No matter what, we're seeing the world through our own little biological eyes.

That's why I see it as collaborative. My own limitations to understanding what I can do with a tool doesn't mean that's all it can do. If I'm a bit more open to it showing me what it can do, I might get further.

We have such a unique opportunity right now. I've studied human history and mythology; that's where my practice started. Reinterpreting mythological and religious stories from a new perspective, which turned into seeing the self from a new perspective is a thread running through my work. It's all connected, because even if we look into the deep past, humans have always been humans. There are so many things that time and context can't change. And then, at the same time, our experiences are totally unique. I really think we're going to be able to see these threads more than ever with this machine learning. What is humanness? And I think we're just at the cusp of discovering that. Or maybe just discovering what that might look like.

Right, I don't know if we'll ever know. And that's okay. We'll keep striving towards that. That is the thing that we want to know more than anything else, right? As humans, we're just desperate to understand ourselves.

There's an element of futility, right? I'm also interested in quantum physics, and there's a truth within that system that no one can exist in the time and space that you occupy. In the whole cosmos, you are totally unique. And that physical time and space cannot be experienced by anyone else. So even though we have a shared experience, there is this element of total isolation. Even though you can try to convey to me your experience, or I can see the time and space you've moved through, I can't ever see it through your eyes. I call that the spark of inclination. That's the individuality that every person has. If we can also find the commonalities, then it will be easier to focus on the individual and allow each other to become more fully realized.

How did you choose these colors?

The way it works with the inputs is with a dot, a single pixel of color. The way it reads the inputs is really interesting. If I put a single dot of color on the screen, it will just fill the whole screen with that color. It doesn't see it as a dot. So if you want different shapes, you have to not only draw it with the color you want, but you have to enclose it with another color. You have to give the proper visual structure to the thing. It not only reads the pixel input color, but also the colors around it, and creates a balance. So if you have lots of white around those colors, they'll be really washed out. It was a bit of a balance of being able to get the shape and structure visually that I wanted, without totally diluting the color or creating big blocks of space. These aren't colors I normally work with; I used colors from the prompt. It was a balance of what I found aesthetically pleasing in a structure that could be read by the algorithm.

I did over a hundred videos for this piece. I chose this particular one because it had the vibe of the prompt in tone, in how the colors blend, and in the speed of things coming together. Peaceful, joyful, light movement, and the colors aren't too vibrant. Some of the videos were very pastel, a melted rainbow paddle pop for instance. Some of them were super vibrant rainbows, which I loved, but it wasn't giving me the same vibe.

Do you have any advice for another artist participating in this project for the first time?

Feel it. Don't overthink it. I could go into concepts for weeks, but I really was glad that I just made the decisions quickly. It felt very exploratory, which was a very fun process. You're out of your comfort zone getting another work to respond to. It's not like your normal approach, so it's better to just throw it all out the window. Respond with what you feel.

Call Number: M65VA | M67VA.myePa

Simpson Myers is a multidisciplinary artist working on the unceded Boon Wurrung lands of the Kulin Nation in Australia. Primarily concerned with exploring the individual experience of self, the work is informed by human stories like religion and mythology, but also philosophy and science. Recently working more and more in emerging technologies, she is excited to question what the idea of self means in our modern, increasingly virtual age.