Less than four years after ChatGPT’s debut, we are nowhere near understanding the role AI will play in our lives. Will it be as transformative as the internet? Will it cause mass unemployment? Will the next generation forfeit its capacity to think to AI models? Will we all soon be using AI agents to book our hotels and flights?  

How to scythe through the bramble? Cory Doctorow, a science fiction novelist and one of our keenest observers of technology, is out with The Reverse Centaur’s Guide to Life After AI: How to Think About Artificial Intelligence—Before It’s Too Late, a companionable guide to the subject. Doctorow does offer some answers to the questions above. No, there will be no mass unemployment, he says. No, human cognition and creativity will remain critical. No, AI agents are unlikely to become as widespread as the hype would have us believe, in part because companies are too paranoid to give third parties free rein to index their pricing information. As for whether AI will be transformative, Doctorow believes it is useful and will continue to make our lives easier; if you consider that transformative, then yes. 

Notoriously prolific, Doctorow published his AI book only nine months after his previous one, Enshittification: Why Everything Suddenly Got Worse and What to Do About It. More original and revelatory, Enshittification presents a fully developed theoretical framework for understanding how Big Tech platforms came to dominate our economy and society so thoroughly in just two decades. The Reverse Centaur’s Guide is like Herman Melville following Moby-Dick with the 50-page novella Bartleby, the Scrivener, about a bafflingly impudent clerk: a worthwhile read, but not a seminal work. 

Doctorow grounds his perspective on AI in the shaky economics underpinning the industry’s boom. Seven technology giants, all heavily invested in AI—Nvidia, Microsoft, Google, Apple, Meta, Tesla, and Amazon—account for 35 percent of the stock market, meaning that a great deal is riding on AI’s success. Because none of the AI models have produced profits on their own, the corporations behind them have resorted to circular financing and accounting gimmicks. Consider Microsoft’s partnership with OpenAI, the company behind ChatGPT, in which Microsoft holds a 26 percent stake. Microsoft gives OpenAI $10 billion worth of tokens to access its data centers, which OpenAI records as investment revenue at face value. OpenAI then redeems those tokens with Microsoft to power ChatGPT. It is not hard to guess what happens next: Microsoft claims to have generated $10 billion in AI-related revenue for its cloud-computing division. 

Such legerdemain is ubiquitous in AI. A report released in May by the Open Markets Institute, where I work, described how tech corporations such as Meta and Oracle use special-purpose vehicles to finance the construction of expensive and unpopular AI data centers, keeping the investments—and accompanying debt—off their balance sheets. 

Another dubious aspect of AI is its “dogshit unit economics,” a term coined by the tech critic Ed Zitron. Unlike the internet and other technologies that were costly when first introduced but became cheaper over time, AI continues to grow more expensive. Doctorow explains that to reduce “hallucinations,” or errors, AI companies break a prompt into multiple pieces and repeatedly query a model about each one, incurring additional costs every time. As a result, each successive generation of AI models becomes more expensive to train and operate than the last. 

Why, then, is Big Tech going so hard on AI? Because it allows investors to view these firms as growth stocks rather than mature companies. Google, which earned $132 billion in profits last year, is a mature company with two established monopolies in search and digital advertising. Being seen as a growth stock, however, allows it to use its shares to acquire other companies and compensate employees. And it is not just Google. “Keeping the growth story alive isn’t about one company or one sector. The entire U.S. economy hangs in the balance,” Doctorow ominously forecasts. 

Investors, not the rest of us, are the audience for this growth story and its inflated numbers. Doctorow breaks down how a reputable bank like Morgan Stanley can claim that AI will be worth $16 trillion. Multiply the number of radiologists in the United States, 32,000, by their average salary, $360,000, and then multiply that figure by 80 percent—the amount their bosses might be willing to pay AI systems to do the same work—and you get $9.2 billion. Repeat that exercise across every industry, and you can see where the $16 trillion figure came from—and how absurd its underlying assumptions are. “AI companies are selling the replacement of workers with chatbots, but chatbots just can’t do workers’ jobs,” Doctorow writes. “They can help some workers, some of the time, but no one wants that—at least, no one in a position to buy hundreds of billions of dollars’ worth of AI products wants that.” 

Doctorow agrees with other AI skeptics that a financial reckoning is coming. What he adds to the “whither AI” conversation is his deep knowledge of technology, his experience writing science fiction, and his work with the Electronic Frontier Foundation advocating for the rights of technology users and workers. In exploring whether AI-generated art is any good, Doctorow notes that he makes millions of minute artistic decisions while composing a novel, most of them unconsciously. Novels, paintings, and other works of art emerge from these choices, all made in service of conveying “that big, numinous, irreducibly complex feeling.” What, then, of AI image generators or chatbots used for writing? “The generative AI doesn’t know anything about the big, numinous, irreducibly complex feeling in the artist’s head,” he writes. “All it can do is add vaporous filler to the meaning that is contained in a human user’s prompts.”  

When Doctorow addresses the often-raised questions of whether AI is, or will become, conscious and whether it might eventually destroy humanity, he is similarly authoritative and, to me, settles the matter in two pithy sentences. “A conscious being isn’t a word-guessing app that knows more words and has more computing power to guess with,” he writes. “Throwing GPUs and training data at AI isn’t going to make a superintelligence.” 

Doctorow introduces terminology from automation theory, in which centaurs are humans assisted by machines, while reverse centaurs are humans conscripted to serve as assistants to machines. He sees the distinction between the two as central to the AI debate. A driver using a system that warns them when their car drifts into another lane is a centaur. Workers who drive Amazon vans are reverse centaurs. These workers, classified as contractors so that Amazon need not pay them minimum wage or provide benefits, must use nine apps that score their performance according to metrics such as braking and swerving. As reverse centaurs, the drivers have jobs only because the van can neither drive itself nor carry a package to a customer’s door. 

Doctorow wants us to be centaurs. In his own life, he has found many handy uses for AI, including first using open-source models to download and transcribe dozens of hours of podcasts, and then turning the transcripts into a searchable database. He does us a solid by deflating the rhetoric surrounding AI and placing the technology in a more realistic light. As the closest thing we have to a tech prophet, he also reminds us that “the most important fact about a technology isn’t what it does, it’s who it does it for, and who it does it to.”  

Amen.