I Cloned Myself

The Question

Artificial intelligence has become a powerful design tool, whether it’s generating concepts in Midjourney, outlining ideas in ChatGPT, or upscaling renders with Magnific. These tools have become part of my practice and likely yours as well.

Of course, the obligatory question is always asked whether or not using this technology is real design. Is posting sketches to instagram real design? Is making superficial renders real design? Is sketching in VR real design?

We had this joke in school that design is simply about making choices, which makes everyone a designer… but not everyone — a good designer.

Design is about making choices. We come up with ideas, create concepts, evaluate them, and choose the best one. The tools we use simply help us make the best choice. With AI we can create realistic concepts faster than ever before which allows us more time to test, refine, and narrow in on that perfect design. And in that way, yes, of course using AI is real design, because at the end of the day, the responsibility of making that final choice still lies with the designer.

To me, a more interesting question has been whether AI is capable of shouldering that responsibility. Can AI completely replace the designer? Can AI be the curator, the tastemaker, the artist itself? Can AI not only make its own choice, but select the same design I would choose, or possibly even make a better decision?

There was only one way to find out, so I cloned myself.

The Experiment

I gathered my writings on design, transcripts from podcasts and interviews, and a lot of my instagram data and uploaded all of it to OpenAI’s custom GPT editor. I drafted a set of rules to essentially jailbreak the AI chatbot so that it had permission to have its own opinion. I told it to act as my clone and make decisions confidently, concisely, and solely based on my design philosophy and thought process.

As of this essay, ChatGPT isn’t able to view images or videos so its focus is on how I think rather than my aesthetic (which is often more conceptual than visual anyways). I’m sure soon we’ll be able to upload all forms of media which would allow for a more holistic AI clone.

The goal was to see if I could get my clone to generate ideas, narrow them down, and select a final concept. The deliverable would be a 3D model that I could print out — a physical object created completely artificially.

I laid out the game plan and asked my clone to design a chair. From there it would provide me with a list of ideas along with text prompts to generate concepts.

In general it seemed to grasp a bird’s eye view of my design philosophy. It would generate ideas like puzzle chair, wave chair, origami chair, etc. It would often reference my values like playfulness, delightfulness, or inventiveness and explain why a particular concept fit my philosophy.

I then used the text prompts to generate 3D models using Luma AI. Of course Luma AI has it’s own unique aesthetic. Text-to-3D is still in early development, but honestly I like that the designs feel less refined and more like AI playdough.

I shuffled images and text prompts back and forth between the two AI systems and would ask my clone to select its favorite design. I would pick my favorite out of each grid to see if it would also pick the same one. From my tests, my clone would choose the best design correctly about 2 out of 3 times, which is notably above a 1 out of 4 random control.

After it narrowed down the concepts and selected the final one, I would take the 3D model and render it out myself. I ran this experiment a couple times to see what the collections of designs would feel like. Did they feel like me?

The final designs lack a bit of the clever twist a lot of my work has. And certainly, the generated 3D models were rough and lumpy, but I can see how some of them could be reworked into more polished designs.

This one was my personal favorite. My clone called it the “Ribbon Chair”. The way the legs curve around to create the backrest is delightful. I can definitely see a path to refine this into a more thin elegant café chair or maybe thick upholstered lounge.

I printed out a scale model to complete the experiment - a completely artificial object.

The Clone Chair is available to mint as a digital collectible with accompanying 3D file. A full scale version can be commissioned upon request.

The Results

1. I couldn’t fully give up control.

If you hadn’t noticed, I actually failed. I still picked my favorite design to print out. There’s this obvious tension and paradox I felt while working on this project. I wanted to make a cool design that told this story, but also wanted to trust the AI to make its own decisions. Sure, I let my clone come up with ideas and go through the full process of selecting a concept, but I ultimately ran this experiment multiple times until I was satisfied with a design it chose.

I also intentionally chose to use Luma AI because it has that characteristic AI playdough look that again fits this story better. Plus it has the added bonus of being a complete 3D model ready to print. I could repeat the experiment using text-to-image AI to generate a more realistic looking design, but the images would lack the physical 3D print aspect which makes the whole thing less compelling.

Ribbon chair redesigned in Midjourney, lacking a 3D model, but still cool!

I realized how hard it is to give up that final curatorial role. Is AI capable of taking that from me? Not currently. Could it in the future? Maybe, but human creativity is built into our DNA. We want to be involved in the creative process regardless of how good AI gets. We want to make the final choice. We want to be the artists, the storytellers, the final decision makers. I believe it's much more likely we will use AI to augment our abilities just like every other technology we currently use. It’ll become our creative companion crafting beautiful designs that align with our visions.

2. My clone’s designs are my designs.

Interestingly, I feel more “ownership” over the chairs that my clone made compared to ones that I could generate through other generic AI programs. After all, I did design my clone, and it did seem to grasp general values that I believe in. The technology will only get better, and I imagine my clone will be much more closely aligned with me one day. Maybe we’ll be able to riff back and forth on designs. It’ll know where to push back and critique my work, and I would do the same to it. This leads into my last realization.

3. I might design forever.

This project has given me a glimpse into the idea of digital immortality. We document so much of our lives, and as this technology gets better we’ll be able to very accurately emulate our digital selves. Which brings up the question of what happens when I die and my digital clone continues to exist? At some point I envision it would have free agency to continue designing on its own, collaborate with other humans or AIs, and possibly work to create products designed by me posthumously. At least for as long as someone or something wanted to engage with it.

I might design forever. We all might.

- Nicholas Baker