Ken Liu on AI and Freedom

May 06, 2026

Ken Liu graces ChinaTalk with his presence. He is the author of the Dandelion Dynasty silkpunk fantasy series and a brilliant short fiction writer — one of his stories was recently adapted into Sam Altman’s favorite show, Pantheon. We all know his translation work on the first and third volumes of the Three-Body Problem trilogy, but even better was his absolutely brilliant translation and commentary of the Dao De Jing. As much as I hoped that project would get him fully on the classical Chinese translation train, he followed it up with a very different direction — a techno-AI thriller, All That We See or Seem, released late last year. Irene Zhang of ChinaTalk joins us to co-host.

In a wide-ranging conversation, Ken Liu argues that:

Technology is the most human thing we do — humans have always externalized our minds into the world and then allowed those creations to reshape who we are.

AI “slop” won’t stop humans from making art that matters, and the real distinction isn’t quality versus slop, but between desire-fulfilling machines and artists who draw from the collective unconscious.

The deeper danger of AI isn’t machines replacing humans, but systems that train humans to behave like machines.

Science fiction isn’t prophecy, but mythology — and ideologies are just mythology’s cheaper, hack cousins. Orwell, Shelley, Tolkien, and Le Guin endure not because they predicted the future, but because they gave us metaphors powerful enough to think with across generations.

Large language models are intelligent, but can’t be wise. Drawing on Laozi and Zhuangzi, Ken explains why everything that truly matters lies beyond language.

Listen now on your favorite podcast app.

Technology as Human Expression

Jordan Schneider: We’re living in the age of Claude Code, and I want to start with a passage you wrote. Why don’t you set it up and read this vision of future coding and writing?

Ken Liu: Let me start by saying what the book is actually about. All That We See or Seem is a techno-thriller in the sense that none of the technology mentioned is really speculative — it’s all either already here or very possible, just needing to be scaled up slightly.

Julia Z is a hacker, a hero in the mold of Clarice Starling or Jane Whitefield — someone with a very strong moral compass and a very dark past. She’s trying to escape that past, but events keep pulling her back in, and she realizes she cannot overcome external threats unless she confronts the demons within her. In this novel, she has specialized skills with AI and robotics and is tasked with finding an artist who has disappeared — an artist who works with AI to help large audiences dream together.

The passages I’m going to read are reflections on Julia in the Age of AI. Here’s the first, which is about what it’s like to be a programmer — something very close to my heart:

The hardest part had been the programming. Writing code without the help of Talos, or even a lowly codemonkey or datajinn, was not something Julia had much experience with. In the same way that few contemporary writers could compose even a five-hundred-word essay without the help of AI as research assistant, fact-checker, dictionary, thesaurus, grammarian, and, in extreme cases, amanuensis, very few contemporary programmers could create a functioning nontrivial application without the help of codedaemons, bug-genies, patchsprites, scriptpixies, and a whole fairyland of similar artificial intelligences.

Homo sapiens had always externalized their minds into the world, oozing books, drawings, plans, recordings, the same way honeybees made their minds visible in the form of wax comb and sweet honey, but the trend had never gone as far as now, when most of one’s knowledge consisted of knowing where to look things up and how to give an AI the best prompts, and more of one’s mind existed outside the skull, infused into fiscjinns and memoelves and egolets, spread among artificial assistants and helpers and aide-mémoire, imprinted in cogitrons and electrons and logons, than remained inside the squishy gray matter inside the skull.

Source.

Jordan Schneider: Let’s start with the idea of choosing a techno-thriller as a genre to explore something every white-collar worker is grappling with today.

Ken Liu: Genre labels are largely irrelevant to me. All of my fiction — whatever the marketing genre — is fundamentally technological. Whether it’s the Dandelion Dynasty, my short fiction, or the Julia Z series, they’re all stories about what it means for humans to express parts of themselves through technology.

If there’s something unique about humans compared to other species, it’s fundamentally our technological nature. This is important. A lot of things described as “sci-fi” aren’t really sci-fi at all — they have very little to do with science. They’re technological stories. “Techfi” is far more interesting to me. Technology and science are completely different disciplines, and the vast majority of so-called sci-fi is really techfi, because it’s really about what it means for humans to express themselves via their creations.

We are the only species who express who we are through the things we make. We imagine things that did not exist in the universe, then actually bring them into being — concretely substantiating our mental constructs in the world. And these technological manifestations, this stuff we ooze out, in turn changes who we are. We converse with, interact with, and co-evolve with our own creations. No other species does this.

One of the great philosophical debates in our tradition is whether humans are more human without technology or with it. This debate goes back to Plato, to Zhuangzi, to all the great philosophers. What is language? The entire skeptical interrogation of language itself is really this debate about human nature.

In the contemporary world, we often default to the position that technology is somehow external to who we are — something we should be wary of. To me, this is nonsensical. Human technology is a manifestation of human nature. It’s in fact the most human thing we make. You cannot understand human nature without understanding human technology — it’s literally a tangible substantiation of what is inside our minds. To understand what human nature is, we have to interrogate human technology and truly understand how we co-evolve with our own creations. That’s what the Julia Z series is really about.

Irene Zhang: You use the metaphor of the jinn to describe what the marketing world might call AI agents. That obviously comes from Arabic mythology and Islam. Why did you choose that metaphor, and how do you think about the metaphors we use to understand AI?

Ken Liu: The immediate answer is that I was interested in the word “cotton gin” — it’s short for “cotton engine,” which is just the way we play with language. Why not take that “gin” and turn it into a different kind of “jinn”?

If you look at how technology is expressed through language, it’s very mythological. Think about how we name our technology. Why did the U.S. decide to name its space programs after Greek and Roman gods? There’s a mythological component to the way technology is manifested because technology is not independent of who we are — technology is how we dream.

The reason technology is so expressive of human nature is that it’s a manifestation of our deepest desires and dreams. We’ve always used mythology to express and understand technology. Look at how technology companies talk about and market their creations — there’s always a mythological component. If I didn’t name them jinns, that would be weird. It has to be a mythological name, because that’s how these companies think.

Eli Whitney's original patent for the cotton gin and the Al-Jinn (The Jinn), the 72nd chapter of the Qu'ran. Sources: Encyclopedia Virginia and Wikimedia Commons

Jordan Schneider: Is this time different? You made the argument to the negative, but in that passage, you’re saying that externalizing your brain to the extent your characters do — or we’re doing today — is something unique in human history.

Ken Liu: Externalizing the brain into our creations is not unique. Every child who has learned to read has experienced that moment of communing with mental patterns from creatures long ago. When you read Plato’s dialogues or Zhuangzi’s stories today, you’re communing with minds from thousands of years ago. That’s very strange if you actually think about it — you’re engaging in something no other creature can do. We’re communing with mental constructs from the past.

Consider what happens when you do arithmetic — long division, an integral, working out a tensor. You’re using pen and paper to externalize your brain. Your cognitive function is literally externalized on the paper. It’s a very strange thing. Your brain is out there, you’re interacting with it using your body, then getting it back. No other creature does this.

Is AI significantly different from that? I don’t think so. The best way to understand large language models is to go back and read the structuralists from the 1980s. Roland Barthes said that in a deeply literary society, burdened or blessed with millennia of writing and millions upon millions of authors, we are surrounded by words — by their minds. A modern writer, a “scriptor,” is not an author who creates out of nothing, but someone basically babbling in the presence of a complete corpus of past writings. You are just playing with words, reference upon reference, allusion to more reference. You’re acting as a channel, a conduit to this playful field of past writing as you babble more writings.

Barthes wrote this as a way of talking about the death of the author, but reading it now in the age of large language models, you realize that’s exactly what he was describing. The large language model is a substantiation of that imagined dictionary of all writings. It’s language coming to life. It’s you interrogating the entire corpus of what humans have written — this “pluribus,” this multi-mind, that you’re engaged with.

That’s my argument for how AI is not really different from how we’ve always dealt with technology.

Now, there are some interesting differences. For the first time in history, we’re confronted with the idea that intelligence and consciousness are not the same thing.

If you examine older sci-fi literature, there’s a huge fundamental assumption that something intelligent will necessarily be conscious — that the more intelligent something is, the more it necessarily comes with intention, will, desire, and the sense of being something, of some mind behind the intelligent acts.

What we’re seeing now is that there’s no doubt these models are intelligent. A lot of the popular discourse — “it’s just a very powerful autocomplete” — is very silly. That description is technically true, but it means nothing. You might as well say humans are nothing more than compilations of statistical likelihoods. Yes, that’s technically true, but so what?

The real issue is this — if something can write essays, pass the bar exam, and get a perfect score on the SATs, to say it’s not intelligent is a nonsensical declaration. It’s clearly intelligent, but it’s not conscious. I don’t think many of us would argue that LLMs are conscious.

That is very strange. The fact that we can have intelligence completely divorced from consciousness, from will, from intention, from subjectivity — that is weird. We’re still coming to terms with it. We’re trying to understand why we value subjectivity so much, yet don’t seem to think intelligence by itself is all that valuable anymore. Many of us now seem to be leaning in that direction.

That’s honestly why a show like Pluribus on Apple TV is so interesting — it’s mythologically engaged with this particular question: what matters more, subjectivity or intelligence?

The Age of Slop

Jordan Schneider: One theme you pick

[truncated for AI cost control]