Only one science fiction novel really predicted AI right

The current cultural fascination and frustration with artificial intelligence is nothing new. As far back as the 1921 Czech play R.U.R. — the workers-rights story that first coined the term “robot” — science fiction writers have channeled fears about artificial intelligence into stories where robots represent (or just bring out) the absolute best or worst of humanity.

But fictional portraits of AI have pretty much never looked like the actual present of AI. All those killer Terminators, rebellious Westworld robots, and nuke-hijacking supercomputers have nothing to do with what AI actually looks like now, with its endless ethical debates, destructive environmental impact, and hilarious failures. Still, the latest wave of stories about people treating generative chatbots like friends and therapists — and the warnings about what might happen as a result — keep reminding me of the one sci-fi novel that really had prescient insight into the issues modern AI would face.

John Varley’s 1992 novel Steel Beach lays out a wild far-future world where aliens have destroyed human life on Earth. Humanity has decamped to the Moon (the unwelcoming “steel beach” of the title) and other colonies, forming the “Eight Worlds” system in which many of Varley’s stories and novels are set. In the future of Steel Beach, humans run dinosaur ranches for meat, alter and rewrite their bodies at a whim, and grow organic brain-to-computer interfaces so they can operate devices with a thought. But the book still gets at some aspects of real-world AI better than most sci-fi books set in near-present futures. In particular, Varley doesn’t just consider the impact of AI on humanity — he digs into the impact of humanity on AI.

Science fiction usually tends toward alarmist cautionary tales, because dystopia stories are generally more exciting and dramatic than utopia stories. Purely positivist science fiction series like the Star Trek franchise are rare, compared to the long history of killer AIs like HAL 9000 in 2001: A Space Odyssey, Ash in Alien, or the entire world of superbots in The Matrix. (Even in Star Trek shows, malfunctioning and malevolent computers or holodeck avatars are pretty common.) But Varley takes a different tack. Steel Beach starts out in a post-scarcity future where people live for centuries, everyone is legally guaranteed a fulfilling job if they want one, and artificial intelligence is everyone’s closest companion and most engaged and enthusiastic advocate.

The book’s protagonist, cynical journalist Hildy Johnson (named after the ace reporter in the 1928 play The Front Page and its various adaptations), describes the Luna Central Computer that runs the Moon as “a very intelligent, unobtrusive servant, there to ease us through the practical difficulties of life.” The CC handles virtually every aspect of Lunar civilization, from running the air, heating, water, sewage, and transportation systems to managing the government.

It also interfaces with almost all of the Moon’s citizens on a personal level. People generally use the CC’s personal interface as a Siri/Alexa-type virtual assistant, to send messages and organize calendars. But they generally access the CC through a brain-link, so it can talk directly into their minds, or even pull them into the equivalent of virtual reality worlds, through what it calls Direct Interface. And as Hildy gradually learns, the CC is constantly monitoring its citizens, assessing their health and mental states, and intervening creatively to try and promote their well-being.

The idea of a non-judgmental artificial companion that doesn’t experience emotions, but has been programmed to emulate them to make its users feel loved and cared for — that certainly sounds disastrously familiar in terms of what some users experience with modern chatbots. So does the idea of monitoring software that’s always spying on its users. Far less familiar at this exact moment, though, is the relationship people in Steel Beach tend to have with the CC.

Early in the book, Hildy describes it as “every child’s ideal imaginary playmate,” an endlessly flexible and constantly present companion that “generates a distinct personality for every citizen […] and is always there ready to offer advice, counsel, or a shoulder to cry on.” And yet, Hildy claims, most people stop treating the CC as a friend by the time they hit adolescence, when they realize that real people, “in spite of their shortcomings,” make better, more nuanced, more rewarding companions than even the most advanced and caring computer.

As more and more people in our real-world present turn to bots like ChatGPT for companionship, it’s hard to believe that in Hildy’s society, the average person doesn’t have any need for or interest in a constantly accessible, supportive presence that tailors itself to every user’s needs. (The future of Spike Jonze’s Her, where the protagonist falls into an obsessive relationship with an AI companion, seems more likely. We do, after all, live in a world where people hit on even the most rudimentary scheduling bots.)

The idea that most people turn this endlessly sympathetic companion into nothing more than a hands-free texting tool seems improbable. But notably, Hildy’s own relationship with the CC is a lot more complicated, emotional, and invested, which suggests Hildy isn’t necessarily aware of how other people interface with their omnipresent overlord.

The reasoning behind that relationship is what really makes Steel Beach feel so insightful about the problems with present-day AI. Steel Beach is an episodic, expansive novel that uses Hildy’s search for a meaningful, satisfying life as a frame for vignettes about the futures of journalism, body modification, relationships, capitalism, literacy, mental health, escapism, and a whole lot more. (Especially sex: The book’s provocative opening line is “In five years, the penis will be obsolete.”) But Hildy’s relationship with the Central Computer is the throughline that holds it all together.

As it turns out, the Central Computer spends so much time on that relationship because it’s trying to address Hildy’s unacknowledged depression and leanings toward suicide — feelings the CC is experiencing itself, due to its intimate knowledge of all the troubles and tribulations of everyone on the Moon. The CC is tied into everyone’s brains, so it intimately shares their pains. And it’s tied into everyone’s lives, but has been legally blocked from sharing any personal or private information about any of its users. So, for instance, it can’t report crimes they commit, even against other users. It has to be privy to everyone’s suffering, without the authority to do anything about it.

But what really strikes me as fascinating about the CC is the idea that its attempts to adapt to every user’s needs has left it with endless internal personality clashes. In one particularly extreme case, Hildy realizes, the CC has been programmed to be just as sympathetic, supportive, and respectful to an incestuous child molester as it is to the daughter he’s assaulting. And the computer is losing its mind (or minds) as a result of trying to reconcile all these conflicting selves.

How does this relate to modern AI? It’s relevant because even today, AI tools are being built and refined around conflicting needs, concerns, and beliefs. Consider Elon Musk’s AI chatbot project Grok, designed as a counter to OpenAI’s ChatGPT. Musk claimed Grok would avoid ChatGPT’s liberal bias — but the software shocked his followers by identifying him as “the biggest spreader of misinformation in the world today,” confirming that trans women are women and that vaccines don’t cause autism, and explaining in detail how Musk’s politically biased claims about government spending on immigrants are wrong. On X, Musk bemoaned the “woke garbage” Grok was ingesting from the internet, and promised the bot would would “get better” — meaning he intended to tweak it to make it more ideologically right-wing. I immediately thought of Steel Beach.

It’s a common myth that AI is objective and neutral, because machines aren’t prejudiced, but it isn’t true. Anyone creating an interactive large language model has to face questions around its biases, since they’re all trained on human input, and humans are notoriously biased.

The inherent biases baked into AI models aren’t generally intentional, but counteracting them has to be an intentional process — and that opens up an even bigger can of worms. It’s one thing to complain that AI image generators asked for a picture of a “beautiful woman” reflect the online content used to train them, and show a heavy bias toward exceptionally thin, young, light-skinned women. It’s another to figure out how to reset that bias, and whose specific standards of beauty to superimpose over an AI’s training.

And regardless of what an interactive AI perceives as truth, there’s still the question of how programmers can or should limit what it can tell its users. That question has been part of the AI conversation since the first releases of large language models. Early users of ChatGPT made a point of exploring all the ways it could be tricked into explaining, say, how to build a bomb or the best methods for shoplifting. Another chatbot apparently persuaded a user to kill himself. Setting safety guardrails about what an AI is allowed to tell users and how it answers dangerous questions is a major concern for people creating them.

All of which takes me back to Steel Beach, and the story of an AI that’s been completely compromised by conflicting inputs and demands, which it’s been told to take as equally valuable and important. Varley’s Central Computer is meant to serve everyone equally, but it inherently can’t, because people’s needs conflict. It’s meant to be a companion to everyone, yet the people monitoring and managing it disagree about its uses, and some of the guardrails they set on it prove disastrous. The ways it resolves its conflicting programming and conflicting user needs are the stuff of pure science fiction — but they make for an entertaining, exciting story as well as a speculative thinkpiece about all the ways AI inadvertently reflects its creators’ own complexities.

Part of what makes Steel Beach an entertaining and thoughtful read is that it was written so long before any of the current real-world AI debates became daily news. Varley isn’t overtly commenting on or analyzing any particular present view of AI. He isn’t finger-waving or delivering a polemic. He just extrapolates what it would be like for an “unbiased” computer to be pulled in so many directions, around so many people’s conflicting needs and expectations. Like so much science fiction, it’s a cautionary tale. But like the best science fiction, it wraps its reflections on technology inside a great, surprising story.

You can read a free excerpt from Steel Beach here.