AI and the Threat of Nostalgia Culture

What happens if the spread of AI slop crowds out the market for original writing and art? It’s easy to read that as the goal of OpenAI et al. Sam Altman would much rather you pay him $20 a month to have his bot tell you about what new albums you might like than have you pay Rolling Stone and its stable of writers $20 a year to do the same. And given that most of us have a lot more we’re interested in than just new albums, and ChatGPT will readily tell us about all those things too, while Rolling Stone output is rather more limited, it’s easy to see how that $20 a month can be tempting. If enough people are tempted enough, then every subscription that might’ve gone to a publication goes to Sam Altman instead.

The trouble, of course, is that to know which new albums to tell you about, ChatGPT needs to be able to find content about new albums, either in its training data or by searching the web and ingesting what it finds. And while plenty of people will LiveJournal for free, to get good original content to ingest, that good original content needs to be paid for. Professionally produced content needs professionals, and professionals need salaries. Or, at least, freelancing checks.

You can maybe replace some of the lost market for media subscriptions by publications signing deals with OpenAI et al. to give them access to whatever they publish. And then they use the fees OpenAI pays them to pay their writers and artists and other contributors. OpenAI takes what those writers and artists and other contributors contribute, and spits it back, or spits back the information it contains, to people paying OpenAI $20 a month to have it answer all their questions.

This is basically ChatGPT as a bundled subscription. Except it’s more like if you bought a bundled subscription that sent a ton of publications to a warehouse in your backyard, along with a strange little man who lives in that warehouse, reads everything that arrives, and then answers whatever questions you might have when you knock on his warehouse door. But, regardless of this Lynchian third party, the publications generating all that new content are still getting paid. (Let’s set aside, for just a moment, the oddness of experiencing new writing and art not through the words or images of the writer or artist, but instead through the second-hand lecture of that weird little Lynchian guy who lives in a text prompt.)

Where this could all go wrong in a way that destroys the flow of money from all of us to the writers and the artists, even the flow through the OpenAI middleman, is if it turns out what most of us want isn’t really new information, new styles, new expression, but a remixing of the old.

Even if we’d had super powerful computers 50 years ago, we couldn’t have built ChatGPT, not like it exists now, because what also exists now but not 50 years ago is the internet. And not just the internet in terms of a globe-spanning computer network, but the internet in terms of vastness. We needed decades of people putting stuff online. To train a frontier model, you need a lot of material to train it on, and training it on a lot of material is possible in a world where the sum of human knowledge and output exists in digital form at publicly accessible URLs. It’s not possible in a world where that knowledge and output is tucked away in physical objects (books, tapes, scrolls) spread out in countless collections in buildings all over the planet.

So what an LLM knows is a snapshot, at the time of its training, of everything it could access on the internet at a given moment, supplemented, as you use it, by what it can find through traditional search. This means an LLM knows a ton, but it also means an LLM is like those photographs the Zoomers pass around of a 90s bedroom. It’s a freeze frame, or at least most of it, and while you can introduce new ideas when you run the prompt, it interprets those new ideas in the context of that frozen moment.

Of course, you can retrain the AI. You can say, “Okay, let’s look at the internet again, now, and base your new knowledge on what it looks like in this updated state.” You’re creating a new frozen moment, like a photograph of a 2000s bedroom. The fresh version of the model can then have that extra bit of freshness–until the world moves on and new content is created and tastes change. If the company waits long enough to retrain it, it’ll be like Disco Stu: hip and with it in a moment that has passed it by.

Assuming people paying that $20 a month for their favorite chatbot want something a little more fresh than Stu, assuming they want what’s current, what’s at culture’s novel edge, then maybe the bundling economy works. It won’t be perfect, because OpenAI isn’t going to sign a licensing deal with every creator who wants to create and get paid to do so. It’ll sign deals with the big publishers, who have legal departments for such things. The market for new creative works will look more like three channels of network TV than the long tail, but it’ll still be a market.

The worry is what if people don’t even want that? Everything, after all, is a remix. We endlessly recycle. The youth of 2025 long for the bedrooms and CRTs of 1995. Marxists complain that capitalism keeps bundling the past and selling it to us as the latest product. We revive folk, and swing, and new wave, and whatever 80s video the TikTok kids have found in the last week. Ours is, in other words, a nostalgia culture.

And this is what AI is really good at. You want something in the style of something? And that style is an established one, well-represented in whatever moment of the internet the AI was built on? That bot has you covered. It can remix with the best of them.

There’s clearly a market for this stuff, too. Stranger Things convinced a generation that what they really wanted wasn’t the future, but endless recycling of an idealized past. Old sitcoms find new life with people too young to have ever seen them when they aired. Etc., etc., etc. The worry, the way the market doesn’t just contract, but becomes small enough to drown in a bathtub, is people, or enough of them at any rate, is that most people in fact just want the old reflected back to them in ways changed just enough to not feel like a rerun but to still feel nostalgic.

I think there will always be a market, from the perspective of the companies making the bots, for breaking news. People do want to know what’s happening in the world around them. We need someone to tell the bot when Russia invades Ukraine or who won the election or what’s up with the price of eggs. The bots need that kind of information to answer “What’s happening?” questions, and so even if the newspapers go under, the corporations gathering those $20/month fees will want to hire enough people to answer them. But the rest of it? At a scale where we can consider it a sector of the economy? At a scale where someone can say, with a degree of reasonable hope, “I want to be a writer or an artist when I grow up?” It’s easy to tell a story about how all that could collapse.

In this world, original content becomes like oil: It’s what powers our existing technologies, but it’s only ever the product of the dead and gone, and we’re not getting any more of it. That sounds bad. I wouldn’t like that world, even if it comes with a cheap bot who lives in my phone and can tell me about everything there was to know at a moment somewhere behind me. And even if that bot is quite clever, and can give me writing of a kind I enjoy or music of a kind I enjoy, the fact that it’s “of a kind” means there’s something pretty important missing: namely, the whole point. What the tech bros who say they’ll never read a novel again because they can have ChatGPT tell them any story they want miss is that it’s not just about the story, it’s about the fact that someone is telling it to you. Art is beautiful and exciting and all that, but it’s also connection. It’s knowing there’s someone else on the other side of that prose or the other side of those notes, and if the novel or the song are meaningful to you, it’s because that other person on that other side gets you. Or gets something important about you, or about something you share.

A world without that is a world I wouldn’t want.

Which is why, even though the above story makes a certain sense, even though we can see how the economic incentives and the cultural preference for nostalgia could pull us in that direction, I’m optimistic we won’t actually head all the way there. Or, if we do, we’ll turn around and head a good deal of the way back.

Because this sort of society is undesirable, and people really do strive for what’s new and interesting, and not merely what is remixed, I am skeptical that such an economy would last. I think people will dial back their use of AI to generate endless slop because the market for slop will collapse. I lived through the era of reality TV. It was crap and it was everywhere, because enough people liked it and it was so cheap to produce compared to scripted shows. But the era of reality TV gave way to the era of prestige TV, and not because the relative costs shifted, but because people decided they actually liked watching shows with a bit of narrative quality. You can still find reality TV, of course, just like you’ll always be able to find plenty of AI slop. But reality TV didn’t eat the world, and, for the same reasons, AI slop won’t, either.

Markets, when left to do their thing, can run off in stupid directions. But they also have mechanisms to correct course. Nostalgia culture is real, the desire to endlessly reduce the cost of content is real, and they can interact in ways that very likely will cost some people the ability to earn a living. We likely can’t avoid that, because LLM technology is here, and people like it, and it has enough genuinely useful applications that we wouldn’t get rid of it entirely, anyway. But it also can’t do novelty, it can’t be a cultural innovator. And plenty of people want both. And are willing to pay for it.