AI art doesn't lack 'soul', it lacks signal

s AI art “soulless”? If you search Twitter for the word “soulless” right now, you’ll see (mostly) a discourse about ChatGPT’s latest image model and the “Studio Ghibli” trend. This is all “soulless AI slop”, say the people. There’s a clip circulating of Hayao Miyazaki, co-founder of Studio Ghibli, reacting to a demonstration of AI-generated art and calling it “an insult to life itself”¹.

There’s something to this. Soullessness isn’t just a generic insult that people throw out when they don’t like art that was created in a certain way. It’s a real phenomenon. But I don’t think AI art is inevitably soulless, and I also think there is a time and place for soullessness.

In order to get there, though, we’re going to have to get a little more precise, and a little metaphysical.

Soulnessness: beauty without expression

In my quest to find something intelligent to say about this, I found an essay “What is Art” by Leo Tolstoy. Tolstoy doesn’t use the word “soul”, but we can still use his framework to try and give a precise meaning to soullessness.

Tolstoy thinks that most people are confused about what art is. Most people conceive of art as the creation of beauty. There are different schools of thought as to what “beauty” is, of course: some say beauty is just subjective taste, other say beauty is objective, a reflection of the divine perhaps. No matter, Tolstoy entirely rejects the notion of art being beauty creation as circular and meaningless.

Tolstoy, would rather conceive of art as an act of communication, like speech. Art is a means of transmitting feelings and experiences.

To evoke in oneself a feeling one has once experienced, and having evoked it in oneself then, by means of movements, lines, colours, sounds, or forms expressed in words, so to transmit that feeling that others may experience the same feeling – this is the activity of art.

It is possible to create beauty, but not to transmit a feeling or experience. Tolstoy would deem this non-art. The Internet, I hypothesize, would deem it “soulless”.

AI art can be beautiful: I don’t think there is much doubt about that. But is AI art a transmission of feeling and experiences? Often no. The AI itself doesn’t have feelings or experiences – at least not in the same way we do – so if there are feelings and experiences being transmitted, the AI cannot be the origin of them – they must come from the people prompting the AI.

And often, the people prompting the AI aren’t really trying to express anything. They’re just trying to be responsible for the creation of something that looks or sounds good. Twitter right now is full of people taking memes from yesteryear, feeding it to ChatGPT with “in the style of studio Ghibli” appended, send post. Call it non-art, call it soulless slop, I do believe it is useful to distinguish this sort of thing from true artistic creation. When you want to experience true art, you’re not just seeking to please the senses, you want to feel like there is somebody on the other end, with something worthwhile to say.

Sam Altman posted the other day something he called “literary meta-fiction” that he created by prompting an internal OpenAI model. It was on the theme of grief. The writing seemed technically excellent, it had a type of “beauty”, so to speak, but it bored me out of my skull. I very well might have been able to find enjoyment and meaning in it, had I perceived it to have been written by a renowned author, but knowing it to be the product of an AI given a generic prompt, I could not find it within myself to care.

But not all AI art is “generic prompt + send post”. For some AI artists – they do have something to express, they have a vision for their piece, and then spend hours iterating on their prompts, rejecting, accepting, enhancing, stitching together generations, and tweaking the result until it matches their artistic vision. This is proper art in Tolstoy’s sense. Nevertheless, if you create and publish something like this, people will not be able to get past the fact that it is AI-generated, and still dismiss it as “soulless”. What gives?

Signalling artistic inspiration

What art like this lacks isn’t soul but signal.

With traditional art – say you’re a painter – if you have something to express, some feeling or emotion, you will invest days or even weeks of your precious, mortal life putting a brush to canvas. You wouldn’t do this unless you (or your patron) believed something was worth expressing – call this “inspiration”. The beauty of your art is evidence of the effort you invested in it. The effort you invested in your art is evidence of how favorably you judged the inspiration. So, creating beautiful art – traditionally – is a way of signaling the strength of your conviction that you and your art have something worthwhile to say.

There’s a meta-level to this, too. If you’re a painter, you don’t only invest effort on each painting, you’ve also invested years of your life ahead of time learning to paint beautiful things in the first place. Making art your life’s calling is a way of signaling the strength of your conviction that you are the sort of person with a certain sensitivity to experience, who discovers inspiration, who often has something worthwhile to express.

Of course, effort doesn’t make meaning – it just signals meaning. No signal is perfect. Artists have always been able to create beautiful, technically excellent things that are meaningless and uninspired.

Now, though, everybody can.

AI obliterates the ability of beauty to signal effort. AI makes it possible to invest next to zero effort and create something that appears (superficially, to the layman, at least) nearly as beautiful as something that took days or weeks of human effort, and years and years of training. There is no easy heuristic to distinguish a piece of AI-generated art that was some normie’s momentary whim, and a piece of AI-generated art that was created by weeks and weeks by a poetic, tortured soul iterating on a prompt.

Because beauty is no longer a reliable signal of effort, it is also no longer a reliable signal of inspiration. So, the Internet dismisses all AI art – however beautiful – as soulless. A lot of voices in my social media feed – even the AI-friendly ones – are expressing ennui, a sense that something has been lost with the introduction of the latest AI image generation capabilities. But I don’t think what we’ve lost is our artistic soul. What we’ve lost is simply the ability to rely on beauty as a signal of artistic inspiration. And I think this is fine. We will just need to find new signals.

Scattered thoughts

• Social signals are more important in a world where beauty cannot reliably signal good art. Be more active in how you enjoy art. Talk to your friends about what you like. Write reviews. Comment on things. AI art hasn’t really become too competitive with traditional art in terms of the ability to entertain, but I don’t think this is far.
• Time will improve things. Models are getting better at following instructions, giving the user more control. This is certainly true of the latest image generation models from OpenAI and Gemini. It’s also true of the latest generation of speech models (quick plug for Hume Octave!) The better that models get along this dimension, the more AI artists will develop their own style and techniques that make it possible to signal high effort.
• Not everything beautiful is art – but not everything beautiful needs to be art. I got a lot of joy yesterday seeing my family photos ghiblified, or done in the style of religious art from the Byzantine Empire, or done in the style of Francisco de Goya. I enjoy it in the same way as I enjoy a snapchat filter, or a funhouse mirror (albeit more intensely). We should be celebrating this ability of this technology to easily create beautiful, personalized, joyful experiences for people. At the same time, we don’t need to celebrate it as art. Art is a different thing, that brings people joy and richness in a different way. And we should take care to elevate the AI-generated content that actually is good art.

Last, I will shamelessly plug a piece of art created by me (plus AI). I wrote this poem in 2019, and a couple weeks ago I used Midjourney to give it a picture, and Octave to narrate it.

Multimedia is an interesting case. The poem itself I hold to be art in Tolstoy’s sense – I wrote it to communicate a particular feeling (a certain ironical delight in the English language and the juxtaposition of software culture and apocalyptic imagery). The AI-generated image and narration should probably not be considered art. My goal for the image and the voice was simply to depict the text of the poem and to be beautiful, not to realize any sort of independent artistic vision.

Soulless slop? You tell me.