The world of Generative AI is disorienting. What's real, and what isn't? With the power to create almost anything that crosses the imagination, the output has provoked awe and terror in equal measure. And as we've seen, even a dash of existential nausea. 

That is, it tends to untether us from our existing sources of meaning, leaving us like Sartre's character of Antoine Roquentin: confined to a purely sensory experience, in a world of items without depth or essence. 

There are many ways to think about this. One of the ways in which AI-Generated media does this is through toying with the distinction between the familiar, and the completely novel. We're fine with the completely new, and we're fine with what's entirely familiar. But what irks us is what lies in the middle—a kind of uncanny valley, but for our psychological schema. 

Let's dive in.

Connected Generative AI Media

Generative AI continues to create an incredible array of art and media - anything from 90-second pizza commercials, to fashion photo shoots with indistinguishable realism. Yet, despite this impressive diversity, consider the idea that they fall into two loose categories.

The first is AI-generated media which is directly derivative of existing psychological schema. We're familiar with the idea, now reimagined with generative AI. Let's call this connected media. Think "Savages," the generative AI rap song in the style of Jay-Z. On a purely sensory level, it's an enjoyable track. But there's also a deeper meaning, anchored in our existing semantic network for Jay-Z, the actual artist. 

So when we listen, it's not like Roquentin staring at the chestnut root. Instead, there's depth and connection. It's a broad spread of associations, that travels into all of the related nodes in the Jay-Z network of our temporal lobes: Marcy Projects, ROC, Brooklyn Rap, etc. This pre-existing schema gives the listening experience depth. 

Likewise, conceptually anchored media can also be a combination of existing cultural concepts. You know of The Pope. You know of HypeBeast Fashion. You add those two together, and you get the now famous AI-generated image of The Pope walking around with a white Balenciaga Puff jacket. There are countless examples of these creative combinations, including a complete and impressively realistic Star Wars trailer, done in the style of Wes Anderson, and all completely AI-generated. There are endless combinations and permutations which enthusiasts continue to explore. 

Whether it's a new rendition from a familiar source, or an AI rendering of a novel combination, the media relies on existing concepts. This takes the experience beyond the purely sensory, infusing it with depth and connection. 

Unconnected Generative AI Media

The other category of generative AI art is much smaller in scale. These media aren't anchored in existing conceptual meaning, at least none of which is evident to the viewer. Let's call this second category unconnected media. 

By definition, everything that Generative AI creates is somehow tied to existing concepts, media, and ideas. But every so often, the sheer complexity and convolution of this process will completely obscure these conceptual anchors. Our brains have no psychological schema to assimilate it. 

The result is an output that looks, sounds or feels completely new. For example, there are throws of AI-generated abstract art, which are such a dense amalgam of genres of previous work that they're not connected to anything in particular. This also includes faces of people who look like no one in particular, or generative AI songs that don't sound like anyone specific. 

In unconnected media, there are no conceptual ties for the viewer to catch on to. The novelty is turned way up. Experiencing it is akin to viewing abstract art. Because it resists existing categories, it becomes a small category in and of itself.

Generative AI Without a Psychological Schema

As we've seen, the brain naturally gravitates towards category certainty. We necessarily utilize existing psychological schema to "lump" new exemplars into existing categories. And when we can't, we "split" them, and create a new category just for them. 

This is exactly how we treat Generative AI Art: Connected media taps are lumped into an existing schema, while unconnected media are split into their own. 

It feels safe in either one of these. The nausea and disorientation stem from media, which lie ambiguously right between them. This type of media behooves our categorical perception. We may know that it's AI-generated, but it's not realistic enough to provoke the visceral feeling of what it's intended to represent. 

It's similar to the uncanny valley effect: the eerie feeling we get when a robotic face is realistic looking enough to look "human-like", but not quite real enough to look like an actual human face. It throws a curve ball at our social cognition. It's not quite a human face, and it's not quite NOT a human face. 

The eerie feeling arises when our brains are unsure how to treat it. When it tries to retrieve the appropriate schema, it draws a blank. 

There exists a similar uncanny valley for Generative AI. Here, it's not a question of whether what we're looking at is human. Instead, it's a question of whether its meaning is grounded in existing concepts. It feels like it may connect to something familiar, but it's not quite familiar enough. 

One of the most prominent examples is the element AI-generated media currently struggles with the most: fingers. We know they're supposed to be fingers, but they don't cross the threshold into full realism and familiarity. As New Yorker writer Kyle Chaya describes, "Looking at gnarled AI hands, we fall into the uncanny valley and experience a visceral sense of disgust. The hands are both real—textured, wrinkled, spotted, with more detail than most human artists could achieve—and totally at odds with how hands are supposed to be." 

 

This strange in-between effect with generative AI art produces a similarly eerie feeling. We know what the object should be, but it doesn't automatically invoke the underlying meaning that should be associated with it. 

The Co-Creation of Generative AI Media

AI-Generated Media is an exciting new frontier, and its output will undoubtedly surprise and enchant over the coming years. It allows users to instantly create anything that crosses the imagination - and beyond. 

At the same time, its worth recognizing that this process is co-creative. The thing exists, but we inject it with meaning and significance. It's us that infuse the art with meaning and greater importance. The feelings that a work of art provokes is as much about ourselves as of the art. Sartre's contemporary, Anaïs Nin, summarizes this perfectly: "We don't experience the world as it is; we experience the world as we are." 

What plagues Roquetin is the nausea he feels when things, as they are, become divorced from this deeper meaning. We feel a similar disconnect when we look at AI-Generated Media - especially those which don't easily fall into our existing schema. 

The future of this new digital world remains to be seen. But, whatever happens, this much is guaranteed: The future may be nauseating, but it won't be boring.

Photo by Matt Johnson via DALL-E

References for Generative AI and Psychological Schema

Chayka, K. (April, 2023) The Uncanny Failure of A.I.-Generated Hands, The New Yorker

Sartre, J. P. (1938). Nausea. Penguin UK.

Squire, L. R., Stark, C. E., & Clark, R. E. (2004). The medial temporal lobe. Annu. Rev. Neurosci., 27, 279-306.

Young, A. W., & Leafhead, K. M. (1996). Betwixt life and death: Case studies of the Cotard delusion. Method in madness: Case studies in cognitive neuropsychiatry, 147-171.