Last week, the United States Senate held a hearing on artificial intelligence. Experts were invited. Senators asked questions. Democracy functioned.

Some of the questions were very good.

Some of them were not.

This is a celebration of both.

The Hearing Begins

The room filled with the kind of bipartisan energy Washington usually reserves for naming post offices. Both sides agreed: artificial intelligence was happening, and the Senate should probably do something about it. What that something was remained, as of press time, unclear.

The witnesses — a carefully assembled panel of tech executives, academic researchers, and one person from a nonprofit who would be ignored — sat at the long table with the patient expressions of people who have explained things to rooms of confused authority before.

The chairman gaveled the hearing into order and welcomed everyone to what he called "this moment in our country's history with the machines."

A strong start.

The Questions (A Selected Highlights Reel)

Senator A — a man who has served in the United States Senate since the invention of the fax machine — leaned forward with the focused energy of someone about to say something important.

"Now, when ChatGPT is writing someone's emails for them," he began, "where does the writing actually happen? Is it on the computer, or is it — where does it go?"

The witness explained, gently, that it happens on servers. Senator A nodded slowly, wrote something on his notepad, and did not follow up.

Senator B asked whether AI could be programmed to "only learn good things." The witness said this was, in principle, the goal, and that it was also the hard part. Senator B said that didn't seem that complicated. The witness agreed that it did not seem complicated.

Senator C had concerns about AI taking jobs. The witness said this was a legitimate concern. Senator C said her office had already started using AI to draft constituent responses. The witness looked at the camera. The camera looked at the witness.

The Metaphors

Senate hearings about technology run, as a rule, on metaphors. Technology is hard to regulate because it is abstract; metaphors make it concrete. Some work better than others.

"Think of AI," said one senator, pausing for effect, "like a very fast calculator."

A researcher at MIT, sitting in the third row of the gallery, quietly put her head in her hands.

Another senator compared large language models to "a library that talks back." This one was actually not bad. The witness said it was a reasonable approximation. The senator looked pleased and wrote it down, probably to use at dinner parties.

The comparison that generated the most internal commentary from the assembled technologists was the senator who described AI training data as "basically like feeding the machine homework." This is technically one metaphor away from being accurate. The senator then asked if we could "just give it better homework," which is also technically one peer-reviewed paper away from being a real research question.

The Moment Everyone Will Clip

In every technology hearing, there is a moment. The Zuckerberg hearings had several. The Google hearings had the iPhone question. This hearing had its own.

A senior senator, gesturing with the gravitas of a man who has gestured gravely at many things, asked the lead witness: "If I asked ChatGPT a question right now — let's say I asked it to write a law — would it write a good law?"

The witness said it could write something that looked like a law.

"Better than what we write?" the senator asked.

The room went very quiet.

The witness said there were significant differences between legal text generation and actual legislative deliberation.

The senator said "I'm not so sure" and moved on.

The Part Where Regulation Was Discussed

After approximately three hours of context-setting, the hearing arrived at its stated purpose: what should be done about AI?

Options discussed included: requiring AI to label AI-generated content (the AI would do the labeling), creating a new federal agency (a bill to this effect was already in committee, where it was aging gracefully), requiring companies to explain how their AI makes decisions (the companies noted this was technically difficult; senators noted this was their problem), and, from one senator who had clearly just read a headline, banning AI from writing news articles.

This last proposal generated the most discussion, partly because several senators had already received AI-generated briefings that morning and partly because one of the witnesses gently pointed out that the senator's concern about AI-generated misinformation was being addressed, at this very hearing, by a room full of humans who had just spent 40 minutes debating whether AI "dreams."

The Takeaway

Here is what will happen next: a working group will be formed. The working group will produce a report. The report will recommend further study. A subcommittee will be established to review the report. The subcommittee will hold its own hearings. By the time any legislation is passed, the AI industry will have released four more model generations, each of which will make the previous legislation obsolete.

This is not cynicism. This is institutional momentum, and it applies to every major technology transition in American legislative history — from railroads to radio to the internet. The Senate is not uniquely incompetent about AI. It is consistently, reliably, structurally behind technology. This is arguably a feature, not a bug: moving fast and breaking regulatory frameworks has its own costs.

The senators who asked confused questions are not stupid. Some of them are very smart. They represent millions of people who are also trying to figure out what AI means for their lives, their jobs, their children's futures. The confusion in that hearing room is a fair representation of the confusion in the country.

The difference is that the country doesn't get to vote on what the AI does next.

The Senate, theoretically, does.

Blake Holt attended the hearing virtually, through a recording, with a glass of something. The senators' identities have been deliberately left unspecified because they are all, in various ways, all of them.