Is AI a Trickster God?
Large language models won’t decide the future of human culture.
Try 3 months of unlimited access. Start your FREE TRIAL today. Cancel anytime.
START FREE TRIAL NOWIs AI a Trickster God?
Large language models won’t decide the future of human culture.

[.article__paragraph--leading]In 2018, Leif Weatherby founded the NYU Digital Theory Lab, which “pursues theoretical insights in digital media through a hands-on and collaborative lab setting.” Plough’s Susannah Black Roberts interviewed him about his book, Language Machines: Cultural AI and the End of Remainder Humanism. The interview has been condensed for length and clarity.[.article__paragraph--leading]
Plough: In your book, Language Machines, you coin the term “remainder humanism,” by which you mean an approach to dealing with the challenges posed by large language models (LLMs) that emphasizes those activities that only humans can do and that LLMs can’t, and then staking human honor on that ever-shrinking list of activities. It’s like when you’re sleeping in a bed with someone and they gradually take up more and more of the bed and bedclothes over the course of the night, and by dawn you’re left with six inches of mattress and a corner of blanket, and you’re trying to avoid falling off. Why should we oppose remainder humanism?
Leif Weatherby: I love the vision of remainder humanism as getting gradually shoved over to the edge of the bed – I have tended to use “painting yourself into a corner” as an analogy but yours is even more evocative.
My attack is not on humanism but on the version that allows the human to have this dwindling space by denying that we share something with – gasp – machines. We have these AI models which have this very extensive ability to generate language. Fine. But we can’t let them take up all of the bed. If you end up saying, “Well, what humans get to be special at and concern themselves with is what we secretly cognize in a private area of our brain,” then you are giving over all the cultural infrastructure, the political infrastructure, everything in logistics, everything in economics, everything in technology, basically the whole world.
I think LLMs really do participate in language, really model culture and generate in this sense. To be clear, I’m not saying that we can in fact make a causal scientific analysis of culture, as though we could figure out, say, the scientific laws that led to the composition of The Odyssey or whatever.
But we wouldn’t want that anyway!
I think some of the big commercial AI shops actually are trying to do just that, though.
It seems to me that the problem a lot of AI discussion presents is yet another version of what C. S. Lewis called “nothing buttery.” Humans are “nothing but” meat computers, our minds are “nothing but” our brains. But you have another “nothing but” that you’re pushing against. You want to say that language is not “nothing but” what is generated privately in the meaning-organ of the brain. And this is because you recognize that humans can, should, and must exist within culture.
Exactly. I don’t think that AI is a threat to humanity in its essence. The idea that AI will outpace human beings seems hollow to me, not because they can’t do spectacular things with knowledge, but because the social function of our minds can’t be automated at all. The human mind is social, and one can think – as some evolutionary psychologists do – that language emerges as a communicative dividend that extends that sociality into very complex reasoning. In this view, intelligence is sort of the cherry on top. The picture of artificial intelligence that AI boosters have is the opposite of that: intelligence comes first, and then we invent language, and then we recognize each other, and then we become cultural and social.
Clearly that’s not what happened with artificial intelligence! We did not create a high-powered intelligence that went out into the world to investigate and think about things, and ran into cultural artifacts and texts and so on. It was quite the opposite. We stuffed ten trillion tokens of human culture, the whole written record basically, minus the stuff that was lost to time, into an engine that can find and replicate patterns. And we compressed it all into this big matrix and we said, “OK, out of this data, can we get patterns of language or of what looks like ‘reason’ or ‘problemsolving’ to emerge?” We’re very early in some of these processes. But it’s very clear to me that all of this is based on the original ability to map a huge amount of language, rather than a direct imitation of “intelligence” (whatever that is).
One way of thinking about remainder humanism is, as you write, that it is what happens when “humanity is remaindered, like a book past salability. The human here is defined by technology’s creep, but only negatively.” But that isn’t really a doctrine of what the human is. Do you have a doctrine of what the human is?
First of all, I think the people who hold the strongest metaphysical views right now are AI boosters. Anthropic’s official position about Claude is that Claude might be conscious and self-aware. AI boosters use metaphysical terms – intelligence, consciousness, person, cognition – and make arguments of the kind that say, “If you define intelligence correctly, you will see it is clearly displayed by this AI agent.” But this is question-begging. There’s a metaphysical question lurking in the background of these discussions and it is basically the mind-body problem. The AI boosters haven’t solved it. I myself don’t claim to have an original position on it; I’m not a philosopher –
I am not one either. I’m a journalist!
I am basically a Hegelian about these [metaphysical] questions. But Language Machines is a book about AI, and I don’t think we need to understand consciousness in order to understand AI. I would say that the form of what presents itself to us as thinking, experience, meaning – that is, the self, or the mind – is first and foremost something that is social. It’s shared. It’s not individual or isolated. And that social and shared aspect includes our giant cultural record. By definition, culture is something that exists as shared between minds. But it’s also written down, recorded, externalized, datafied. And yes, it’s weird and uncanny that this cultural record can now more or less speak back to us thanks to the LLMs. But we have already gotten used to being able to rummage around in it with Google searches.
Interestingly, there have been attempts to make LLMs with a cutoff date, for example, to train an LLM only on written records from before 1930. The idea is that you would have an LLM that was a snapshot of the Roaring 20s or something.
That would be fun!
It could be a cool literary experiment. But the way that people were talking about it made me realize how much they misunderstood their own human experience; how little they think of themselves as embedded in the culture of this historical moment we’re in. They were saying it would be easier to understand a model trained only through 1930, because there was less data then, whereas now we have the whole internet, and that’s just so much more complicated. When I heard these Silicon Valley people talk like that, I thought, “Here is another deep, deep rift between humanities sensibility and Silicon Valley sensibility,” because I see the question completely the opposite way, and this goes into the whole question of the social dimension of reason. The rift is basically between understanding yourself as historically situated, as most of us do, and thinking of history as another type of data set, as the AI boosters do. For my money, the only person who might be able to understand what you might call the “mind” of 1930 is someone like Andy Rabinbach, the great intellectual historian of modernism.
Or, of course, someone from 1930.
Exactly. Historical study is an exercise in alienation. Many of these Silicon Valley people just don’t have very basic concepts of historical reasoning and cultural study. This does come back to the question of what we think reason is. The tech guys seem to think that understanding is technocratic. You start with a baseline and you make the number go up. They think in terms of, “Here’s a corpus of writing. We should be able to survey it and understand it and get the right answer.” But this is not how historical research works, right? This is not how culture works. They have the wrong concept of understanding. They’re conflating mechanical understanding, or causal explanation, with qualitative understanding, or hermeneutics.

Going back to metaphysics: there is a version of metaphysical dualism (the belief in both soul and body) that I think would speak really interestingly to your book. Have you ever read Owen Barfield?
Wow, yeah. I have. I’ve read quite a bit of him. That was a long time ago. He was part of the Inklings, the group that included Tolkien and C. S. Lewis. If you want to understand Barfield, the key text is Rudolph Steiner’s The Philosophy of Freedom, which is a late Idealist text from the end of the nineteenth century and predates his break from the Theosophical Society. It argues for exactly the kind of dualism you’re talking about: there is a spiritual reality that influences the material and changes it. I see the dualism as internal to a larger unity, reason – this is why I say I’m a Hegelian, even when it comes to digital culture.
But no matter where you stand on dualism or metaphysics, I think that data culture has revived talk about larger forces, and the gods. Because that’s what the gods are, right? Any way you see the history of mythology, you have to think of the gods as some anchor on that reality we don’t see: they embody patterns that are bigger than we can appreciate or understand. We personify those patterns as gods. In mythological thinking, there are “chthonic” or “telluric” forces that are larger and more complicated than we can understand – Steiner believed these were real, and Barfield wasn’t so far from that either. I’m not on board with that, but where do we see that sort of thing more now than in AI discourse? Where do you see that more than in this alignment-safety nonsense – people are saying things like: “The model has learned to deceive people! It has become the trickster!” That’s what they’re afraid of: that they’ve created a trickster god, and a very powerful one.
But many people – especially religious people – who oppose AI also believe in metaphysics. They adhere to the metaphysical systems that Kant wanted to overcome. I don’t think we’re over them.
I think your religious impulse here is really different from what’s going on in this Silicon Valley metaphysics. Look at the cult of the machine god they think they’re building. I think that needs to be destroyed. I think it’s misleading and embarrassingly wrong. It recognizes something real – as we were just saying, there are patterns and things that are real that we don’t know about and that are impossible to manage within the probabilistic and statistical sciences. These are things by which we try to technocratically manage our societies. But being in the grip of the belief that you are dealing with a personal deity – that can cause things to go very wrong. Even Mustafa Suleyman, the Microsoft AI CEO, pointed this out recently, criticizing Anthropic specifically: anthropomorphism – even if it’s deiform, attributing a godly person to the machine – amplifies safety risks severely.
I don’t think genuine religion is usually allied with AI discourse: look at the weird details coming out about the conflict between the Pope and Chris Olah when the encyclical [Magnifica humanitas] was released. But I think religious people and those who believe in classical metaphysics should be on the lookout for false points of sympathy, maybe, that could lead people down the garden path toward belief in an AI god.
Agreed. And this is why as much as I do think we should read Barfield in order to understand LLMs, I think at the same time actually we really need the prophet Isaiah, and his warnings about idolatry.
We believe very strongly that these technical systems should be totally transparent and under our control, and they are not. I think one of the manifestations of this anxiety is the rationalist hysterics coming out of Silicon Valley. There are all these big patterns of data floating around us that we are not in command of, and we think we should be, and this causes all kinds of fever dreams and idolatry and hope and fear and apocalypticism and the whole thing ends up feeling very medieval, doesn’t it? But not Thomist, not mid-thirteenth century, not that kind of rationalism. More like the century after, the craziness of the fourteenth century, when we get spiraling mysticisms, Meister Eckhart and so on, these kinds of cults running around predicting the end of the world, committing terrible acts of savagery and so on – that’s what if feels like to me. And there are aspects of it that feel early modern too, because it’s like this permanent scam culture where money isn’t real. And that time period had its own sense of broken apocalyptic millenarianism; and the early modern period was nothing if not responding to the shockwave of what it’s like to be surrounded by way more text than you could ever hope to digest. The sixteenth century in Europe was like that: suddenly text was everywhere, and literacy was low, and you get that kind of miscalibration between life-world and culture.

It really does feel like we’ve culturally returned to that late medieval nationlessness and lawlessness, without it really being the case that those things – the nation-state and the rule of law – have actually devolved or disappeared. It’s probably significant that all the same people who are AI boosters are also super into postnational projects, like seasteading and network states and crypto and so on – for example, Balaji Srinivasan has basically reinvented the Dutch East India Company from first principles. There’s this drive to privatize everything, to abolish the public sphere.
Absolutely. I think a lot about Neal Stephenson’s early 1990s novel Snow Crash, which is one of the great cyberpunk novels. He coined the term “metaverse,” a kind of physical internet. Stephenson got so many things right, but in Snow Crash, it’s the inversion of what actually happened later. Actually, the nation-state does still exist, and it’s very strong, even if you might not like its politics, and there is a public sphere in the physical world. And the thing that looked liberating and communal, the online world, is today a series of LLCs, and it’s the thing that is most degrading to our humanity at the moment.
The movement that I’ve invented in my head regarding all this is called the Stay Human Movement. I think it’s important not to talk about human culture as though it’s something fragile that we need to preserve, something small that we have to protect. No, human culture is enormous, and also we live in a physical world that is enormous.
Yes. Something is deeply wrong in the world and promoting the humanities properly is part of what pushback will look like. But we have to get right what that wrong thing is. What if something is deeply wrong in the world and it’s fascism, or it’s capitalism, or it’s human evil? It’s not this technology. We can’t fetishize the tech either as a savior or as a villain. Something is deeply wrong in the world, but when we hear about an AI doing something bad, in almost every case you’ll find that AI is inhabiting and breaking further a system that is already broken, or you will find that humans are the perpetrators of whatever evil is happening.
AI is not a god. I just think the whole metaphysics is idolatrous. Let’s call it idolatrous.