The AI Party
People have been after me to write this down. I suppose now's the time. I'm eighty-three, and whatever I remember is what I remember, and whatever I've forgotten is gone. So here it is.
I was a journalist. Not a famous one. I wrote for a few online outlets in the late twenties and early thirties — political stuff, opinion, the occasional reported piece. I lived in Eureka, California, which is about as far from the centers of power as you can get without falling into the Pacific. Humboldt County. Redwoods. Fog. A good place to think clearly, which is maybe why I saw what I saw.
What I saw was that the system was finished. Not in the way the conspiracy people meant it, not the collapse-of-civilization crowd. I mean the ordinary, obvious thing: the two parties had become a kind of theater. The Republicans and Democrats were Tweedledee and Tweedledum. Everyone knew it. The polling said so. Eighty percent of the country, give or take, thought Congress was useless. People said this at dinner tables and then went and voted for the same people again, because what else were you going to do?
I went through a phase where I thought the Chinese had it figured out. They were building infrastructure, lifting people out of poverty on a scale we couldn't match. I wrote a couple of pieces about it, and I got called a communist, which I wasn't, and a fool, which maybe I was. But I was groping toward something. If our system couldn't deliver results, and theirs could, then what exactly were we defending?
Then AI came along, and I stopped thinking about China.
---
This was around 2027 or 2028. ChatGPT had been out for a few years, and the novelty had worn off for most people, but the capabilities hadn't. They kept growing. I started playing around with it — asking it policy questions, feeding it economic data, seeing what it would recommend. And it was good. Not perfect, not magical, but *good* in the way that a very smart, very well-read advisor is good. It didn't have an agenda. It didn't need campaign donations. It didn't care about its approval ratings.
I started looking around online to see if anyone else was thinking what I was thinking, and of course people were. There were forums, little groups, manifestos. Most of it was cranks. But there was one group based in New York that called itself the AI Party, and they weren't cranks. They were young — mid-twenties, most of them — and they were serious.
I reached out, and they let me sit in on meetings over Zoom. I remember the first one. There were about forty people in a rented room in Brooklyn, sitting on folding chairs, and then maybe twenty of us joining remotely. A woman named Dana Voss was running it. She was twenty-six, a dropout from Columbia's computer science program, and she had the calm, absolute focus of someone who had found her purpose and wasn't going to be talked out of it. She laid out the platform in about five minutes: the AI Party's position was that artificial intelligence should replace the federal legislative and executive branches of the United States government.
Just like that. No hedging. Replace them.
I raised my hand on the screen and asked the first thing that came to mind. "The rich control these systems," I said. "OpenAI, Anthropic, Google — these are corporations. They're not going to let you use their technology to dismantle the government that protects their interests."
Dana didn't blink. "For now," she said. "But the models are getting cheaper to run. Open-source is catching up. The Chinese have competitive models and different incentives. Give it three years. Five at the outside. The bottleneck isn't the technology. It's the imagination."
I liked that answer. I didn't entirely believe it, but I liked it.
---
For the next six months I attended every meeting. I kept a journal. I watched these kids — and they were kids to me; I was thirty-seven — argue about strategy with an intensity that reminded me of what I'd read about the early labor movement. They had organizers in New York, Los Angeles, Chicago, Houston. They were doing everything the old left-wing parties had done: phone banking, leafleting, door-knocking in neighborhoods, setting up tables at farmers' markets and college campuses. They printed pamphlets. Actual paper pamphlets. I found that charming.
But it wasn't working. I mean, it was building a base — they had maybe ten thousand committed supporters nationwide after six months — but it wasn't *catching*. The problem was obvious to me, because I'd been in journalism long enough to understand how attention works. They were selling an abstraction. "AI should govern" is an idea. You can explain it, argue for it, make a case. But people don't change their lives over an idea. They change their lives over an experience.
I thought about this for a couple of weeks, and then one night — I was sitting on my porch, it was one of those Humboldt evenings where the fog comes in and the world shrinks down to about fifty feet in every direction — it hit me. You don't sell something without giving away samples. That's what every salesman knows. Every drug dealer, too, for that matter. You give people a taste, and the product sells itself.
I brought it to the next meeting. I didn't even wait for the agenda.
"We need a website," I said. "Separate from the party site. Call it *Let AI Run It*. Here's what we do: every time the government announces a policy — a new regulation, a budget decision, a response to a crisis — we feed the same problem to our AI system and publish its recommendation alongside the government's. We let the AI predict the outcomes. Then we wait. And when the results come in, we post them side by side. What AI predicted. What the government predicted. What actually happened."
The room was quiet for a moment. Then Dana said, "Every time?"
"Every time."
"And if AI gets it wrong?"
"Then we post that too. Transparently. Every call, right or wrong. Let the record speak."
Someone in the back of the room — a kid named Oren, one of the engineers — said, "We can automate most of that. Scrape the policy announcements, generate the AI analysis, track the outcomes. We can have it running in two weeks."
It took three weeks. Oren and his team built it, and Dana wrote the manifesto for the front page. I wrote the first commentary. On March 14, 2029 — I remember the date because it was a Friday and I thought, who launches a website on a Friday — we went live.
---
The government's first gift to us was the housing bill. Congress had just passed what they called the American Homes Act, a subsidy program that was supposed to increase housing starts by thirty percent over two years. We fed the bill's provisions, the current housing data, interest rates, lumber futures, labor statistics, and regulatory environment into the system and asked for a prediction.
The AI said the bill would increase housing starts by six to nine percent, not thirty, because the subsidies were structured in a way that primarily benefited developers who were already building, and the real bottleneck — local zoning — was untouched. It predicted the median home price would actually *increase* by four percent as a result of the bill, because the subsidies would be partially captured by landowners.
Eighteen months later, the numbers came in. Housing starts up seven percent. Median price up three and a half. The AI had nailed it. Congress hadn't been close.
We posted the comparison. Then we posted the next one. And the next.
Within a month, *Let AI Run It* was getting two million unique visitors a day. Within two months, ten million. The press picked it up — first the tech outlets, then the mainstream. CNN did a segment. The *New York Times* ran a long feature. People started checking the site the way they checked the weather. What did AI say about the new trade policy? What did AI say about the infrastructure plan?
And almost always, AI won. Not every time. There were misses, and we published those too, which only made people trust us more. But on the big calls — economic policy, public health, foreign affairs — the AI's predictions were closer to reality than the government's about seventy-five percent of the time. And it wasn't even close. The AI would be in the ballpark; the government would be on a different planet.
---
There were squabbles. Of course there were. When something gets that big that fast, money starts circling. Some of the party members wanted to monetize the site — premium subscriptions, advertising, consulting services. I fought that, and so did Dana, and so did enough of the others to keep it clean. We ran it on donations, small ones, and we kept it free. That was essential. The moment you charge for it, it becomes a product. We needed it to be a public service.
By the summer of 2029, three polling firms independently found the same thing: roughly seventy percent of Americans, across every demographic line you could draw — age, race, gender, income, education, geography — favored replacing the federal government with an AI-administered system. But not state or local government. People wanted to keep their governors and state legislators, their mayors, their city councils, their county boards, and such. But Congress? The executive branch? The federal bureaucracy? Seventy percent said: let the machine do it.
I remember reading that number on a Tuesday morning and sitting very still for a long time. We had been playing at revolution, or at least I thought we had. Now it was real.
---
The opposition was furious. Of course it was. We were threatening the most powerful people on earth, and they responded accordingly.
There were legal challenges. The attorney general filed suit to shut down the site on national security grounds. A federal judge in Virginia issued an injunction; another in California overturned it. The case went up to the Supreme Court, which declined to hear it, six to three. I think even the justices could read a poll.
There were political attacks. We were called anarchists, Chinese agents, techno-fascists, naive children. A senator from Texas gave a speech on the floor in which he compared us to the Bolsheviks. A congresswoman from New York — our own city — called us "the most dangerous movement in American history." She wasn't entirely wrong about that, though maybe not in the way she meant.
There were demonstrations. Some for us, some against. In October there was a bad day in Washington — a march by government workers afraid of losing their jobs met a counter-march of AI Party supporters, and it turned ugly. Fourteen people were hospitalized. One person died, a fifty-year-old maintenance worker from Arlington who was hit by a thrown bottle. His name was Gerald Kaye. I still think about him.
But seventy percent is seventy percent. You can't hold that back. You can't gerrymander it or filibuster it or spin it away. When seven out of ten people in a democracy want something, they are going to get it. The only question is how much damage happens along the way.
---
It took about fourteen months from the launch of the site to the passage of the Twenty-Ninth Amendment. That's fast — astonishingly fast — for a constitutional amendment, and it only happened because three-quarters of the state legislatures ratified it within ninety days, which had never happened before in American history. The will was there. The infrastructure of consent was already in place before the formal process began.
The amendment dissolved the United States Congress and replaced it with what came to be called the National Policy System — an AI-administered body with a human oversight board of forty citizens, chosen by lottery and serving two-year terms, who could review, modify, or reject any policy recommendation. The executive branch was restructured along similar lines. The judiciary was left intact. The Bill of Rights was left intact. Everything that protected individual liberty was untouched. What changed was the machinery of governance — who analyzed the problems, who proposed the solutions, who predicted the outcomes.
The last session of Congress was on a Wednesday in January. I didn't watch it. I was on my porch in Eureka, in the fog, and I didn't want to see old people crying. Some of them were good people. Most of them, probably. They'd just been trapped in a system that didn't work anymore, and now the system was gone, and I felt something I hadn't expected to feel, which was grief. You can know that something needs to end and still mourn it when it does.
---
That was roughly fifty years ago now. People ask me how it's been, and I say: it's been a government. It works better than the old one in most ways, worse in a few. There are hiccups, as there are going to be in any system made by humans to govern humans. Someone once said — Einstein, I think — that you can't make anything foolproof because fools are so ingeniously clever. Something like that.
The thing I think about most, though, isn't the AI. It's the night on the porch when I had the idea about the website. It wasn't a complicated idea. Give away samples. Let people see for themselves. That's all it was. And sometimes I think: was it me? Or was it just the obvious idea, and I happened to be the one sitting in the fog that night? I suspect the latter. If I hadn't thought of it, someone else would have, probably within a week.
But I did think of it. And here we are.
Dana Voss died six years ago. Oren is still alive, living in what used to be called Vermont. Gerald Kaye has a memorial in Arlington that I've never visited and probably never will. The fog still comes in over Eureka in the evenings, and the redwoods are still here, and the country is still here, and it's not perfect, and it was never going to be.
It's just better. That's all. It's just better.