To Serve Man: What a 1962 Twilight Zone Episode Tells Us About the AI Industry in 2026
The Public AI Brief · Issue No. 27
I have been away from this newsletter for a few months. Not because there has been nothing to say about artificial intelligence in the public sector. Quite the opposite. I have been watching the news accumulate, thinking about what kind of publication this should be, and trying to find a frame worth writing about.
This week, I found one.
A colleague of mine, an economist at Penn, raised a question that has been bothering me for a while. I’m parphrasing, but her question was, if these AI tools are so valuable, why are the companies building them burning through cash at a rate that would sink any normal business? They are not making money. In most cases they are not close to making money. And yet they keep building, keep offering, keep giving things away. What exactly is the business model here?
It seems like some economists are just now catching up to questions science fiction was asking over sixty years ago. And that question sent me back to a show I have thought about many times since I was a kid.
The Episode
In March of 1962, a CBS anthology series called The Twilight Zone aired an episode titled “To Serve Man.” If you have not seen it, the setup is this: an alien species called the Kanamits arrives on Earth. They are enormous, impassive, and apparently benevolent. They end world hunger. They share technology that eliminates the need for weapons. They offer free trips to their home planet. Humanity, understandably, is overwhelmed with gratitude.
The Kanamits leave behind a book. Cryptographers work furiously to decode it. They manage the title first: To Serve Man. Reassured, people begin boarding spacecraft for the Kanamit home world in large numbers.
Then, as the episode’s narrator is himself about to board, a colleague comes running across the tarmac. She has decoded the rest of the book.
"Mr. Chambers, don't get on that ship! The book — To Serve Man — it's a cookbook!"
The question I have: “Are we being fattened up to be devoured?”
Rod Serling’s Twilight Zone was, at its best, a machine for making people uncomfortable about the present by disguising it as the future. The show ran from 1959 to 1964 and managed to smuggle serious commentary about McCarthyism, nuclear anxiety, racism, conformity, and the dehumanizing effects of technology past network censors and corporate sponsors who might have killed it outright had it been less clever about the packaging. Serling fought those fights constantly. The science fiction frame was not just an aesthetic choice. It was a survival strategy.
We don’t have anything quite like it today. We have prestige television with bigger budgets and longer episode counts, but very little that does what the Twilight Zone did at its peak: take a mainstream audience, meet them where they are, and then turn the lights on in a room they did not realize they were sitting in.
I have been thinking about that room a lot lately.
The Cookbook
Here is the thing about the Kanamits that makes the episode work as horror rather than just plot twist: they were not lying. The book really was about serving man. It said exactly what it said on the cover. The problem was not deception in the conventional sense. The problem was that humanity accepted surface-level evidence of benevolence and stopped asking harder questions. What kind of serving? Serving whom? Serving toward what end?
My economist colleague’s question deserves the same scrutiny applied to the current AI moment.
The major AI companies are not profitable on their core products. They are offering tools at prices that do not reflect actual costs, in some cases offering them free, and absorbing staggering losses in the process. This is not a secret. It is discussed openly in financial reporting and in industry media, occasionally with a kind of admiring wonder at the boldness of the bet. The assumption embedded in most of that coverage is that the losses are temporary, that scale will eventually produce unit economics that make the math work.
What gets less attention is that the acclimation to higher costs is already underway, and it is happening on multiple fronts simultaneously. As the saying goes, “if it’s free, you’re the product.” AI companies have made this explicit. Don’t want your data used to train their frontier models? Pay for the privilege of privacy. Want real functionality rather than the stripped-down version? That will run you upwards of $200 a month. That’s more than a CrossFit gym membership, more than most streaming subscriptions combined, and a price point that has somehow become normalized in the span of a few years.
But the costs don’t stop at the subsciption price. Even if you never use these tools, you are absorbing them. Residential electricity prices jumped 7.1 percent in 2025, more than double the inflation rate, and topped 20 percent in some states. In the PJM grid region, which covers 13 states from Illinois to the Atlantic Coast, data centers drove an estimated $9.3 billion increase in capacity market costs, costs that are then socialized across every household and business on the grid. Wholesale electricity prices near data center clusters have risen as much as 267 percent since 2020. I live in Maryland. Just this month, my utility bill went up $100 a month. I didn’t get a vote on that.
The mechanisms are different but the logic is the same. None of this is happening by accident. These are policy choices, about who bears the cost of infrastructure, about how public resources get allocated, about which industries get subsidized and which populations absorb the tab. I tell my students that public administration is fundamentally about who gets what, when, and how. Right now, the answer to that question is being written into utility rate structures, federal procurement decisions, and agency budget lines, mostly without the public realizing it is being asked.
My colleague’s answer is that the industry losses are not a problem to be solved later. They are the strategy right now. The goal is integration. The goal is dependency. The goal is to get these tools woven into workflows, institutions, and human habits deeply enough that the cost of leaving becomes prohibitive. At that point, the pricing conversation changes entirely, and it changes in favor of the people who own the infrastructure.
This is not a novel observation about technology markets. It is, more or less, the history of enterprise software. But the speed and depth of the current integration push is different in ways that matter for government specifically.
This Week’s News Feed
I don’t need to speculate about the direction of travel. This week’s news is the evidence.
The General Services Administration launched a platform called USAi last year, billed explicitly as a way to “accelerate AI adoption across the government.” It was free. This week came the announcement that GSA will now require agencies to pay for it. That arc, from free adoption accelerant to paid dependency, took less than a year. The cookbook was right there in the title. It was called an adoption tool. Nobody asked: adoption toward what, and at whose eventual eventual expense?
The CIA, according to reporting this week, recently used AI to generate an intelligence report for the first time. Deputy Director Michael Ellis announced that agency staff will eventually manage teams of AI agents. The Secret Service is embedding AI experts across the entire organization. The VA is requesting a 10.9% budget increase driven primarily by what its own documents call “AI Infrastructure.” New Orleans has trained an AI agent on three years of 311 call data and is preparing to hand it the phones. New York State is scaling AI training to over 100,000 employees.
Each of these stories is being written as progress. And in many ways they are progress. I am not making the argument that these tools have no value. I use them constantly, I teach with them, and I have built programs around helping public servants use them well.
But research from KPMG and The Register this week found that organizational leaders are continuing to increase AI spending even when they cannot demonstrate return on investment. That is not a story about confident investment in a proven technology. That is a story about momentum that has decoupled from evidence. And when spending decouples from evidence in government, it is worth asking who benefits from that momentum and who is building dependency on whom.
What the Twilight Zone Understood
Rod Serling was not a pessimist about technology. He was a skeptic about power, and about the human tendency to accept gifts without examining the terms. The Twilight Zone was full of wishes granted in ways that destroyed the wisher, of machines that served their stated purpose perfectly while missing the point entirely, of futures that arrived exactly as advertised and turned out to be unbearable.
The show trusted its audience to sit with that discomfort. It did not resolve the tension. It presented the evidence and let people draw their own conclusions. That is a rarer thing than it sounds.
I’m going to try to do something in that spirit going forward. Just the lens, applied honestly to what is actually happening.
This week, what is actually happening is that governments at every level are accelerating their integration into AI systems at a moment when the long-term pricing and dependency implications of that integration remain almost entirely unexamined. The tools are real. The value is real in many cases. The Kanamits were not wrong that they intended to serve man. They were just more literal about it than anyone thought to ask.
It is worth reading further before we board the ship.