By Nick Sheeran
When tasked with creating the art for the fourth annual Forbes AI 50 list, it immediately struck me that we should use Artificial Intelligence to generate our deliverable. AI is making considerable strides not only in commercial applications, but in visual art as well. Its artistic endeavors are being auctioned at Sotheby’s, generating NFT collections, and expediting traditional production processes across media. What does this mean for the future of art, and its reception by the general public?
In most cases for artificially intelligent visual work, such as the main art for AI 50, a General Adversarial Network, or GAN, is “trained” on a large dataset and compares the individual relationships between each instance of data to understand what belongs and what doesn’t. The GAN slowly learns to filter out noise in that data, revealing the similarities, and finally achieving the ability to recreate the material received or to determine whether or not a new input matches it. It’s the same machine learning process that goes into loan approval automation or public health diagnostics, the main differences being the end goal and the dataset used. Once trained, the GAN’s knowledge is contained in a matrix of vectors, referred to as a “latent space.” The art you are viewing is a composite of 4 separate “latent spacewalks,” trained using two ready-made datasets provided by Runway ML, and two that were hand-curated by Forbes staff. We “walked” through the matrix, each step resulting in a frame of a video.
I find this process enjoyable because it renders any single output of the GAN less interesting than a sequence of them. My goal is to engage audiences with art in a dynamic format that is reliant on time and systemic relationships beyond the static perfection of a framed painting or statue. Rather than objects made to create an aesthetic experience and hold value, I like to think of art as a momentary output of artists’ practices, which are living, breathing, often focused not just on form but on investigation of the world. It’s more than giving the viewer a feel-something moment, though that’s great. It’s also meant to spur thought, influence opinion, and ultimately affect change. This activity is dynamic, purposely fuzzy, ill-defined, loose and inviting serendipitous meanderings and forking pursuits. It generates turbulence and optimistically criticizes, inviting you to do the same.
In the turbulence of this dynamism, there’s something else to realize: the so-called author of a work of AI art is no longer a lone creator. The AI, plus the sources of its training material, is her (often unpredictable) partner. Designers and artists have long discussed the idea of programmatic co-creation, see Sol Lewitt’s Wall Drawings or the Conditional Design Manifesto. But the impending widespread utility of artificial intelligence is going to bring this spirit of collaboration further into the mainstream. Everyone and no one can be a creator, and that’s great! It comes at a moment historically pivotal for other reasons that bear the need for collaborative spirit, such as climate change and the reframing of peaceful globalism in the face of a resurging Cold War. The idea of domination has to go away: it is time to rewild, reunite, and return to traditions of communal and ecological reciprocity that we have lost sight of. Although AI is high-tech, and can certainly be used for evil, it has the potential to reinvigorate organic relationships that are essential to a sustainable future. By no means do I see what we’ve made as very pretty or refined, but maybe that shouldn’t be the point anymore.
Let’s also take a moment to explore the idea of refinement in AI as it relates to kitsch and ingenuity. Researchers measure the accuracy of a GAN using a metric called the Fréchet Inception Distance, or FID, which basically quantifies the accuracy of the GAN’s output in relation to the data it was trained with. If you want to make a GAN that generates, say, leaf blowers, the lower the FID, the more realistic the leaf blower. If we tried to do this in art, we would immediately arrive at kitsch; it’s just a thoughtless reiteration of something preexistent. That’s the difficulty in creating something meaningful with AI—if it’s too accurate it’s meaningless, and if it’s too ambiguous it’s meaningless again (and yes, swirling images that make buildings look like dogs, Van Goghs or cheese balls almost always fall into the first extreme of this meaninglessness spectrum).
Not to mention, it’s pretty hard to find 5000 images of something, and harder still to curate those images in a way that isn’t betrayed by the limitations, and underlying discriminations, of humanity’s image-making apparatus. For example, we used a series of images capturing well-designed industrial products for one component of the illustration, and I question if it only results in a kitschy pastiche that reinforces the heuristics of that field, as opposed to investigating the universal structures governing them, and profoundly winding the world into a moment. It’s critical that as GANs are adopted by the public we avoid setting off on the wrong path. Look around you, what meaning is embedded in the built environment? How are you interacting with the cumulative expression of society?
Ultimately, we will forever continue fusing disciplines into others and discovering the intersectionality of our existence. Art will become a tool and tools will become art and hopefully, at some point soon, decentralization will shift the responsibility of creation and design away from a technocratic few to the cooperative many. When that happens, it will be more important than ever for us to define what’s essential to sustaining our lives. I would like to think they will be filled with individuality, curiosity, accountability and optimism.