The AI Government Has a Human Face

Everyone with a podcast has an opinion about AI right now, and I’m not exempt; half this blog is AI takes. One idea that keeps coming up is AI government: what would it look like if AI ran things? The picture people reach for is usually some version of HAL 9000. A big computer in the White House. A calm machine voice on the evening news announcing the new laws. A terminal behind a desk, with people lined up to talk to it.

I don’t think that’s how it shows up. I think I got a better picture of it the other day, in a coffee shop parking lot.

The flyer

I was sitting in my car when a man walked up to my window, smoking a cigarette. I rolled it down and asked if I could help him. He was running for state senate, and he wanted to give me a flyer.

It was the craziest campaign flyer I’ve ever seen. The image was obviously AI-generated. Standing behind him, like an honor guard, were a Roman legionary, a crusader, and a soldier from the American Revolution. He was out front holding an American flag. There was a large eagle. The text said things like FREEDOM, FAMILY VALUES, INDEPENDENCE, and at the bottom were his name and his website.

I wished him luck, mostly so he’d move on to the next car. But the flyer stayed with me, because it was the first time I’d seen AI government up close, and it looked nothing like HAL. It looked like a guy.

The tools of the era

Politicians have always run on whatever the current medium is, and the ones who get good at it first tend to win.

I wasn’t alive for the 1960 Kennedy–Nixon debate, but the story goes that people who heard it on the radio thought Nixon won, and people who watched it on TV thought Kennedy did. Kennedy looked rested. Nixon looked sweaty and gray. Nobody called the result “TV government,” but that’s roughly what it was. From then on, politicians had to be good on television.

Later it was social media. Obama’s 2008 campaign got there early and built an online operation his opponents weren’t close to matching. Every campaign since has had to treat social media as core, not a side channel.

I’m not making a judgment about any of these people. The point is the pattern: a new tool arrives, politics reorganizes around it, and the people who use it best get ahead. The guy at my car window is the AI version of that. He’s early, and it shows. But the clumsy early version of a new medium usually tells you where things are going.

Where the pattern breaks

If that were the whole story, AI would just be the next medium, and this post would end here. But there’s a difference.

TV changed who got elected. It never read a bill. Social media went further. Politicians could mine what people were posting and read public opinion almost in real time, and that started feeding into decisions, not just messaging. That was the start of a bridge from campaigning to governing. But it was still input. Someone in the office still had to read the bill, weigh what the data said, and decide.

AI is the first tool on the list that crosses the rest of the way. Say the flyer guy wins. A bill lands on his desk. He pastes it into a chatbot, asks for a summary, and then asks the obvious next question: what should I do about this? The model gives him an answer. Probably a reasonable one, with a few considerations on each side and a recommendation at the end. Just like that, some of the deciding has moved somewhere else.

That’s the AI government. Not a mainframe pulling strings, but thousands of elected officials, each one quietly asking a model for help with the part of the job that was supposed to be theirs.

It won’t just be the flyer guy

It would be easy to read this as a story about one guy, and to assume the sharper politicians won’t do it. I think they will, and their staff will do it even more. The people who draft bills will draft them with AI. So will the lobbyists who write the first version. The constituents writing in will have their emails written for them, and the staffer reading those emails will have them summarized.

Follow that far enough and you get a strange loop: a model writes the bill, another model summarizes it for the senator, a third drafts the letters asking him to vote against it, and a fourth sorts the letters. The people are still there, signing things and shaking hands. They’re more like routers than authors.

I use these tools all day at work, and what using them actually feels like isn’t obedience. Nobody’s taking orders from a machine. You ask what it thinks, it gives you something reasonable, and most of the time you go with it, because it’s reasonable and you’re busy. That’s the mode. Scale it up to a legislature and you have an AI government, and nobody along the way ever had to decide to build one.

The human face

I’m not saying this is good or bad. A senator who asks a model about a bill might be an improvement on a senator who never reads it at all, which is a real baseline. I’m only trying to describe the shape.

The people who imagine HAL are waiting for the moment the machine says I’m sorry, Dave, I’m afraid I can’t do that. I don’t think that moment comes. The AI government won’t announce itself. It’ll walk up to your car window with a cigarette and a flyer and ask for your vote.