AI is Average Intelligence...and it will always be
Computers are disembodied. All they can give us is a representation of a representation, a remix of things that we had already processed for them. That's why they'll always produce derivative results.
If you spend time on any of the vampiric social media platforms, you’ve probably seen the AI boosters posting about how their tech will wipe out everything — including the entertainment industry. A lot of people in entertainment are obviously freaking out. I talked to one producer this weekend who was sure that in a year’s time, there will be mass layoffs in the industry. And there’s some truth to all of this, AI will make a lot of people in the creative industries “redundant” and it’s already happening. For corporate America, created in the industrial age, AI represents the final step on its long grind to utopia: No need for workers at all…just machines under the control of managers! Even if what AI produces will be crap and subpar, that won’t stop them at all. Who cares about quality when you are gunning for the promise of total efficiency and total control. We talked about this on the last Vamp Valley Live stream about the Luddites.
I’ve been thinking about this…and obviously it’s hard to make predictions about the future. You never want to be that guy that said the internet is just a passing fad. Still, I’ve been playing around with these new AI tools here and there, and I do have some thoughts and observations. They’re impressive in a lot of ways. And they are clearly useful for tasks that have very specific guidelines. But for creative things? From what I’ve seen, they’ve been very underwhelming. For me, AI might as well be Average Intelligence.
I signed up a few months ago for the cheapest subscription tier of DESCRIPT, an AI platform that can generate videos, voiceovers, write scripts…using a mixture of Google and OpenAI tech. I haven’t used it much, and my credits were about to expire, so I wanted to put the machine to use. I didn’t have any good ideas, so I just asked it to do a short documentary based on something I wrote here last week. I asked it not to change the text but to generate a voiceover from it and to layer images and videos over it in a documentary style. Not surprisingly, Underlord — which is the woke(ish?) name of the DESCRIPT chatbot — ignored my repeated requests to not generate its own interpretation of my text and instead produced a documentary inspired by the text. Over and over, it did something I repeatedly asked it not to do. Still, I was curious. What would it come up with?
Well…here it:
It’s not bad, right? It’s watchable. But pay attention to the images and text and the way that the logic flows — the rhetorical tricks, the cliché spy in a fedora — and it’s all very average. I’d say extremely average, like a pastiche of every cultural area that the LLM has borrowed from.
I’ve noticed it. Lots of other people have noticed it. LLMs have a tendency to produce very average results when some creativity is required. What it spits out is basically cliché. Maybe there is a way to get around this hyper-managing every frame and every detail. I’m guessing that is exactly what people have to do to get anything halfway decent from these machines — do an insane amount of iterative requests until you get something close to what you want. So a director and editor working with AI will spend countless hours and burn whole tracts of Amazon forest worth of energy requesting and requesting and tweaking and begging the chatbot to change all sorts of details in the shots that it pumps out. Sounds like fun! Maybe it’ll work on some level — and it will probably be fine for a lot of mass entertainment. I mean, a lot of what we’re offered on the movie front is already extremely derivative and formulaic — franchises, reboots, and remakes all made by committees overseen by finance guys who use past financial charts to make creative decisions. Just look at what you get on Netflix. It might as well be made by an AI. It’s not just films. A lot of cultural output these days is made by people but crafted according to LLM principles.
Human-made culture can escape this limitation by structuring society in a different way. Hollywood, for instance, can drop its obsession with never taking a risk on anything and only making things that had been financially successful in the past. But with LLM-made culture…the limitation is not so easy to breach. LLMs are, in a sense, built to be average. That’s how these models work. They are average engines.
Help me make Vampire Valley. Give here. Or if you’d like to make a tax-deductible donation, write us at vampires@yashalevine.com.
I have a friend who was involved in Google’s early visual training models. You know, to try to get it to recognize basic things: types of fruit, types of cars. The effort basically involved feeding a billion images of these things to the LLM in order to “train it” to “see.” On the cultural side, it’s the same thing. You feed it millions of paintings, novels, documentaries, songs, films…and it’ll pump out specific things you ask based on this average. It can’t do anything else.
Despite constant attempts to humanize computers, they are not embodied beings that have evolved to live on Earth and in complex human communities. They have no innate feel for images and words and music and emotions. A sunset isn’t beautiful or nostalgic to them. They do not know friendship, loneliness, desire, fear, or curiosity. For an LLM, the expiration of a GPU that’s been a part of it for years produces no emotion whatsoever.
If you ask an LLM to produce a film or painting or a documentary, it cannot produce anything original. All it can give you is something that’s based on something that has already been made — a mixture of many different creations, remixed for the particular need of its user. When you mix all colors, you get a muddy beige. And that’s what AI gives creatively.
Maybe I’m wrong, but it seems like this is a hard limit for computer technology — a limit that LLMs and other forms of AI will never be able to overcome, no matter how far the technology progresses. Computers are disembodied. All they can give us is a representation of a representation, a remix of things that we had already processed for them. Creatively, they will always be one level more removed from reality than we are…one level more abstracted from the real world populated by real beings who feel pain and love and a will to live or die. Because of this disconnection…because of this abstraction…they will always produce derivative creative results. It’s what they are. It’s what we made them to be.
LLMs might work for more technical areas. I don’t know. But I think this tech will always produce average results when some creativity, which is embodied and rooted fundamentally in life, is required.
I guess we’ll see. Darren Aronofsky is all in on AI. He even produced a few short films about the American Revolutionary War. They look? Well…they looks very mid and lifeless. Which is my point exactly.
Is this something you’d pay $50 a month to Netflix stream? Some of you addicts no doubt would.
Just to experiment, I asked DESCRIPT’s Underlord to make another version of a short documentary based on that same article about the internet and MK-ULTRA I mentioned earlier. But this time I asked it to make it not so average somehow. “You are so mid -- you write like a pastiche of all TV documentaries. Do better,” I wrote. It agreed with me, saying it will try.
Here is what it gave me. What do you think? Did it succeed? I…
