A New York lawmaker used AI to come up with a housing bill. Is the legislation any good?

In what could be a first in U.S. politics, a New York lawmaker revealed this week that he used AI to both “think up” and write a new piece of legislation.

The bill itself, submitted in May, is short and mostly unremarkable—it would give renters the right to request a new copy of their lease from their landlords. What stands out most is the bill’s “AI disclosure,” explaining that it was “researched and written by artificial intelligence, with the accuracy and language reviewed and refined by humans.” 

New York state assemblyman Clyde Vanel, a Queens Democrat and intellectual property lawyer who submitted the bill, tells Fast Company that the idea to use AI began with a sense of curiosity after seeing “technology replacing a lot of the work that I used to do as an attorney.” As a lawmaker, he wondered, “is it possible that the system can replace me?”

The results, he says, blew him away. Vanel’s staff used Auto-GPT, an experimental open source tool that uses the GPT-4 engine to autonomously perform successive tasks, including searching the web, to achieve a set of user-defined “goals.” 

Tyler Fritzhand, Vanel’s legislative director, tells Fast Company that they didn’t specifically set out to produce a housing bill. Instead, they simply assigned Auto-GPT a role (“You are a New York state lawmaker, who is trying to solve problems”) and a few goals: research New York law, find a gap in New York law, write a bill to close that gap, and write a memorandum justifying how the bill closes that gap. “We just let it run for a long time,” he says. The program “kind of talks to itself. It will show you a thought process, and when it has to do something next, you can say yes or no. What we did is type in a command that allows it to just keep saying yes.”

The process took around “two to three hours of total runtime” and cost “ballpark, maybe $1 or $2” in API access fees, says Fritzhand, adding that it required “a little bit of coding knowledge” to operate.

The program ultimately produced a few bills addressing different issues.  Some tried to close loopholes in New York’s gun laws, “which it didn’t do a great job of,” Fritzhand says. He isn’t actually sure how the AI came up with the tenant bill. “We looked online to see if there are any housing rights organizations, tenants rights organizations that were talking about the issue. Nobody was talking about it. It just thought that on its own,” he says. But Vanel and the staff agreed: “This is a pretty good idea. Let’s run with this.”

Before submitting the final copy, Vanel’s office made just a few modifications. Auto-GPT’s version proposed that any landlord who didn’t give their tenant an extra copy of their lease would be guilty of a Class E felony—a crime punishable by one to five years in jail. “That was a bit much, so we took that out,” Fritzhand says. The staff also added a line limiting tenants’ requests to twice a year—“we don’t want tenants to abuse this right and ask for their lease every single day from their landlord.”

But the final bill has perplexed New York housing advocates, who say it seems to be tackling something that seemingly nobody asked for. 

On a textual level, it’s “a perfectly fine piece of drafting,” says Ellen Davidson, a veteran tenant attorney at New York’s nonprofit Legal Aid Society. “But if you’re relying on the AI to think about what’s missing in landlord-tenant law in New York state, I don’t think they did a great job.”

Instead of using AI to come up with ideas, lawmakers should just talk to their constituents, Davidson says. “I’m not totally convinced that for the tenants in his district, that this was the most important issue they had, or even one-hundredth most important problem.”

Vanel, for his part, says he doesn’t understand the criticism. “The use of AI doesn’t diminish the seriousness of the issue,” he says. “It’s a good idea for people to have access to their leases if they lost it, whether or not somebody complained about it. The bill drafting process takes many layers, and this is just the first step.”

Vanel says he’s planning to introduce more AI-generated bills in the next legislative session, which he hopes will send his colleagues a message:  “Don’t be afraid to use technologies. It’s a human-machine interaction, and you have to work with it to make it better.”