Three Reflections w/ Van Arman

lorepunk.eth:

There’s nothing out there quite like Reflection—it’s not hallucinating the next word like ChatGPT or building on an aggregate of images like Midjourney or Stable Diffusion. It’s also not an autonomous artificial person like the AIs of science fiction. It’s an extension on your lifelong experience and practice as an artist.

There’s been plenty of writing out there, like your wonderful statement on the sovrn site for the recent Reflections drop, that goes into the technical details of Reflective AI, but I’m intrigued to learn how you approach your creative system as an artist, every day as you switch it on.

Pindar Van Arman:

My tools are paint, canvas, a robotic arm, and algorithms.

My art is the synthetic creative system that uses these tools to paint.

My hope is that I have achieved making a system that is reflective within itself but also gives viewers the ability to reflect on how its creativity works.

I have built this system by studying my own creativity and constructing algorithms that embody how I create. I try to understand what constitutes creativity, break it down into components, recreate them as algorithms, then embody the process in a robotic system.

The canvas that the system paints serves as a window into the process, giving the viewer a vehicle to imagine how the synthetic creativity unfolded. It contains all the marks as a visible artifact of the process. Every expressive stroke, each technically precise line, reveals what the system was thinking at that point of the creative process.

Furthermore, the digital element of each artwork makes it possible to witness exactly when and how the system performs the various creative decisions it made, like pulling a face out of latent space, balancing a composition, enhancing luminance differentials, among dozens of other algorithms.

The system leaves a trail of marks on the canvas as it reflects on itself, offering the viewer a way to reflect on the process. AI art has to go beyond the shock of the Uncanny Valley and give people ways to understand and talk about the underlying structures, because it's these structures that are currently shaping the world.

lorepunk.eth

What captures my imagination when you describe this is that Reflective AI offers us artistic merit and value in so many ways—the canvases and data it outputs, the algorithms challenging each other for primacy. But the invitation to reflect, to interrogate our own consciousnesses and examine the nature of creativity, adds another artistic discipline: that of systems art. I don’t think I’ve heard of an AI art project that becomes conceptual art in this way. Why is this significant for AI and Generative Art?

Pindar Van Arman:

There are many artworks made with AI, but only a handful that use AI systems themselves as a medium. Furthermore most artworks that are currently being called AI art are better described as digital works created with the aid of AI.

For example, the most popular application of chatGPT is people using it to enhance their writing, whether it be essays, stories, emails, or whatever. But no one thinks of these outputs as “AI Literature.” Lots of pop songs are currently being made by AI, but no one considers “AI Pop” to be a music genre. Likewise we should be cautious of putting all images created with AI in the genre of “AI Art.” Allow me to temper that statement, however, with the acknowledgement that if the digital images are dealing directly with the medium, then yes, they can be considered AI Art. But if they are instead dealing primarily with aesthetics, however surreal, they might be better understood as either digital or surreal art.

The fact is that we are so early in the widespread use of generative AI that there remains confusion and disagreement over what exactly the AI Art genre is. And before we can have something that can truly be considered an AI art movement, it’s first necessary that viewers are able to understand the underlying processes well enough to engage with them and the discourse surrounding them.

What I hope to achieve with Reflection is to provide insight for the viewer to better understand this. To better understand AI systems. It takes the AI out of the black box and allows people to conceptualize it, see how it works in detail.

lorepunk.eth

As I have learned about Reflective AI, I’ve been delighted to see that it comes to its creative decisions the same way I do when I write a poem—by sense, feel, vibe, and most of all, continual reflection on what I’ve been doing in a piece, with reference to my years of experience and memory of what I’ve done before.

I find it humbling and hopeful that an AI system can do the same thing—and rather than feeling intimidated by this, I feel kinship with a system that makes art via a raucous Parliament of competing ideas and inclinations.

At the same time, the whip-crack speed of AI development, and the rapid integration of LLMs and other AI into our society, sometimes feel far more inscrutable. Is designing innately reflective AI systems a possible resolution for social concerns about the tech?

Pindar Van Arman:

AI is now moving faster than even optimistic experts have predicted. We are at the inflection point where AI is not only getting better, but getting better faster. I personally would not be surprised if general AI (the AI we see in Hollywood) announces itself in the next couple of years.

When many of the leading experts describe the key elements of the next phase in AI, what they talk about is reflective AI. Most notably, the founding lead of Google Brain, Andrew Ng, recently presented a broad framework that he believes will carry AI into its next phase. His broad framework describes “iterative agent workflows” that use reflection, multiple agents, and collaboration. He has noted that GPTs wrapped in such agents increased their accuracy from 48% to 95% merely by reflecting and improving their responses iteratively.

So aside from similar names, how close is Reflection, my synthetic creative system, to the reflective AI that experts are predicting is the future? My robots already operate with multiple aesthetic algorithms, the algorithms already collaborate, and the system has been reflecting via feedback loops for more than a decade. It was already using this framework before Andrew Ng presented it in March. And I do take pride in the fact that my first Reflection marks its provenance from February 7, 2024, a month before Ngs announcement. The collection is in a place similar to if I had released a collection called “GANs” in 2014. My art is 10 years ahead of this trend and the evidence is on the blockchain.

But aside from being forward-thinking, what I am trying to do with Reflection is give people a way to experience this idea visually, and to bring it into contemporary artistic discourse. I’m also building tools that make it possible for other artists to integrate multimodal, reflective processes into their art. I am already being asked to showcase Reflection in multiple shows and galleries from Shanghai to New York, so excited that my work with Reflection is just getting started.