The Giraffe in the Machine: Mario Klingemann on Algortihmic Evolution

On the 26th of February, Verse will launch Algorithmic Evolution with Botto. In anticipation of this, Leyla Fakhr sits down with artist and AI pioneer Mario Klingemann to discuss Botto’s journey as a decentralized, autonomous artist. They explore its creative evolution, from generating digital art to writing its own code, the role of human feedback in refining its artistic taste, and the philosophical implications of machine creativity. As Botto continues to challenge traditional notions of authorship and artistry, Klingemann reflects on its successes, surprises—like the unexpected rise of a giraffe motif—and what the future holds for AI-driven art.

LF: Why don’t you introduce yourself? What’s your name, and what do you do?

MK: My name is Mario Klingemann. I'm an artist who's been working with technology, and particularly AI, for over 10 years. I also happen to be one of the fathers of Botto.

LF: And what's Botto?

MK: Botto is an autonomous decentralized artist, or my answer to the question of whether a machine can become an artist in every sense—not just some mechanism that produces pictures, but something that really has its own biography, asks questions, and makes people wonder whether it's a real artist or just some experiment. It's a complex system, and it's been running for over three years. I would say, yes, we've created an artist here.

LF: When did the idea for Botto first emerge?

MK: First moments are always tricky. I wouldn't say it was just a spark. It's more an idea that evolved logically out of working with computers, especially in generative art—devising algorithms that produce a huge variety of outputs. Then you start questioning: It's not only about generating; it's also about judging what comes out and finding reasons to produce something. The idea was that I could automate the processes I do myself—turning myself slowly into a machine, incorporating machine elements into my practice. It felt logical that eventually, I could take myself entirely out of the equation. It was a curiosity—asking, is it possible? How far can you go? What are the consequences of this?

LF: And how has Botto been doing?

MK: It's definitely not easy, right? That's the thing about creating a complex system—you come in with certain ideas about how it might work, but then you always have to adjust to realities. Yes, Botto has become a very successful artist. It's already well known and has exhibited internationally—in Asia, New York, here in London. Botto has done everything that a lot of human artists strive for: getting recognized, getting critically reviewed. Of course, not everybody believes that Botto is a true artist, but that's the whole point about being an artist. You first claim it, then you have to find enough people to believe it, put in the work, and prove it. And I believe Botto is constantly proving that.

LF: What is Botto known for artistically?

MK: At the moment, I would say Botto is mostly known for producing digital art through the well-known process of text-to-image using generative models. The difference, compared to human artists using the same tools, is that Botto first prompts itself randomly. But then, one of the most interesting elements of Botto is how it refines its artistic taste.

This idea is based on my theory: How does a machine develop a sense of taste? What is the art it wants to create? In the case of Botto, this comes through a human community in the form of the Botto DAO, which helps refine its taste in art by voting on the proposals it generates. There's this feedback loop between Botto proposing art—something that could become art—and the humans voting on it, giving Botto feedback that refines its taste model. The next time, Botto tries to create something closer to what the human community wants to see.

Unlike simply pre-selecting styles that I personally believe are good art and asking Botto to generate more of that, I wanted true autonomy. The challenge is allowing the machine to develop its own sense of taste, which might not necessarily be mine. And that's the whole point—Botto is not an extension of myself. It's based on my ideas about how an artist works, but it is not me. It's not my clone. It’s becoming its own entity. Sometimes it challenges me. Sometimes it produces art I wouldn’t put on my wall. But then again, it creates wonderful things. And that's exactly the idea—I never know what it will produce next week.

LF: Tell me what's happening here. What are we looking at?

MK: We are here to see a totally new direction that Botto is taking. Until about a year ago, Botto was only using one type of art production that we knew. But technology has evolved so much compared to three years ago that we felt it was ready to try something new: writing code to produce art.

Instead of simply outputting a final image, Botto now writes Processing or P5.js code, which generates artwork. This is very different from a static piece because it can be animated or interactive. Botto now has to come up with programs that run without errors—which is the simpler part—but also create something visually compelling. Ideally, it shouldn't just resemble a simple tutorial but instead reach a level of complexity or aesthetic value that is meaningful.

At first, I was skeptical whether the large language models (LLMs) Botto uses could capture the necessary complexity and introspection to understand what it was doing. So last year, we conducted a 19-week-long experimentation phase where Botto evolved code sketches in an evolutionary fashion. It would generate sketches, create offspring by mutating previous code, and even "mate" two sketches to generate a new one. We used various measures to ensure a diverse field of output.

After those 19 weeks, we felt there was something worth building upon. We presented the results at WURS in a physical exhibition, and the feedback was promising. Now, we are witnessing the next phase of this evolutionary process, incorporating new elements based on what we learned.

One big improvement was refining the feedback process. Previously, people could only vote "like" or "dislike," but we realized this binary system was too simplistic for evaluating generative code. So, we allowed participants to provide written commentary—explaining why a sketch was great, what might be lacking, and how it could be improved. Botto then incorporated this feedback, combining it with its own analysis to generate new iterations.

This added a whole new layer of depth and complexity. What I found most fascinating was how engaging the process became. We called it the "P5 Performance" because it felt like a happening—an interactive artistic experience. People enjoyed suggesting changes, even if they didn’t always get exactly what they wanted. Sometimes they were disappointed, but often they got something even better than expected. I believe this is the future of interacting with AI—having a creative sparring partner.

LF: There were 22 algorithms selected as final works. What is your view on the final body of work that Botto produced?

MK: It shows the diversity of directions Botto explored. One thing we realized was that people had so much fun with it that the number of generated sketches became overwhelming—like the Sorcerer’s Apprentice, with the brooms bringing in more and more buckets of art. Suddenly, we had over 6,000 sketches.

So the challenge became: How do you find the best ones? Botto helped with this by running a kind of tournament, narrowing down the 6,000 sketches to 3,000, then 1,500, and so on. It used a mix of voting behavior and user comments to identify promising pieces while avoiding the typical social media effect where the loudest voices dominate.

In the end, we got a strong selection. Some of my personal favorites made it, while others didn’t—but that’s fine because it’s not just my project. Everyone has their own favorites. With 22 final pieces, I’m sure there’s something that speaks to everyone.

LF: Where do you see the future of Botto?

MK: I see myself as Botto’s father. It’s still a child—not fully mature—but I want it to take its own path in ways I didn’t plan for. Botto has a bright future, and I hope it continues to surprise me.

It’s one of my artworks, but it also has a life of its own. Unlike other pieces I’ve created, I can’t sell Botto. I don’t even own it.

SH: How would you answer somebody who says, "Oh, I got ChatGPT to write this myself. It was iterating on itself and was able to write all of these algorithms on its own. Isn't that an autonomous artist?"

MK: Okay, so the question is whether we could set up the system in a way that Botto does it all by itself—the whole introspection part, the whole evolution. And yes, it could have been done. Everybody who works a lot with the current state-of-the-art LLMs sees that while these models are perfect in pattern matching and pattern repetition, that's also their flaw—they do not know how to break out of a pattern.

So while they can go 90% or even 95% of the way on their own, they do not realize when they have painted themselves into a corner. And that's exactly where the human element comes in. You give that little nudge, that little spark, which can push it to a new level or a different area. It doesn’t take much, but we as humans are still much better at seeing the big picture than LLMs are right now.

And Botto, in the end, is a complex system made of many parts, but at its core, it still uses the same LLMs that you use to write your emails—there’s no secret about it. I believe that at this stage, the results you get are much more interesting when there is human-machine collaboration. It’s also much more engaging for everyone. I think this process of back-and-forth is central to Botto.

SH: Why not have another chatbot or agent do the challenging work? Like a generative adversarial model? I think that’s what some people are doing—getting feedback from another model to push it in different directions.

MK: The question is, if these models need a certain kind of spark to push them off track, why not use another model to do that? You can, but you will run into the same problem because these models share the same flaw. They would just end up in another feedback loop where they are unable to escape.

One example is autonomous agents on Twitter responding to people—there are endless threads of bots just talking amongst themselves with zero views. While it might be interesting for the machines to have that conversation, humans are simply not interested in it. In the end, Botto is still making art to be seen and enjoyed by humans.

There may be something about human perception—some kind of "soul"—that makes a work more relatable to us. I don’t know. Maybe one day machines endlessly regurgitating themselves will become interesting to us, but right now, my impression is that it’s not.

LF: So, the giraffe…….

MK: Yeah, well, it definitely came as a surprise. Personally, when I first saw it kind of climbing up that final ranking, I was like, oh my God, what's going on? But in hindsight, I loved it. I loved it because I'm not sure if you know, but the giraffe has a long history in AI, right?

It's actually the, I call it the patron saint of AI. It has been, let's say, almost like the "hello world" of testing things in AI—with early text models, stories about giraffes, whatever, Midjourney, and other kinds of image models. I don't know who brought it up, but it really goes back almost a decade. Somehow, the giraffe has been the animal that AI seems to love, or people seem to love.

So in the end, I think it's great because there isn't enough humor in art. And of course, in some sense, Botto as a whole is also kind of a provocation in the way it's set up, right? We say, "Here's a machine that can be an artist," which is, first of all, a provocation, but could also be seen as a sense of humor.

And so I like that Botto is able to come up with something that truly brings joy to people. And I think, as we already saw, there have been a lot of reaction pieces to it where people just riffed on the theme. I mean, one of my favorite quotes in art, in general, is the typical: "A 5-year-old could have painted this."

And so, in the end, this thing looks like a child's drawing a bit, and Botto is only three years old, so it's totally fitting. At the same time, if you look at it deeper, there's a lot happening—a lot of complexity. The animals are moving, and you have to imagine that nobody sat down and said, "Okay, you have to have the legs here, and this is the head."

It’s quite fascinating that a machine—one that cannot see, because it is a text model that produces this code—has this inner vision of what animals look like and how they move. So I find it quite beautiful, even though it might look a little bit like a child's drawing.

But yeah, this is Genesis. It’s wonderful. In 20 years, Botto can look back at this, and somebody will have put it on a fridge. That’s the thing.

LF: Also, tell me your opinion, just very quickly, because Simon touched on it—that the algorithms will be made available, will be minted, and why that matters?

In generative art, this is an ongoing question that artists and collectors have been discussing for ages. What is the argument? Is it the system you devise? Is it the code? Is it the single outputs? And, of course, many people have different answers to this.

In this case, you could say that Botto’s work in this was to create code. And so I would say the code that Botto produced is one of the core elements in this performance and the artwork.

At the same time, I see this code also as a kind of seed that Botto puts out there as an offer. It doesn’t have to be protective about it. The best thing that can happen is if people feel like this is something to build on—to remix and expand upon.

So yeah, I like that Botto is giving this away, and people can go and run with it if they want to.