Machines that draft laws: they’re heeeere

Presented by

With help from Derek Robertson

The recent flurry of excitement around AI has raised the question — where does it end? Could machines write our journalism? Our poetry? Our laws?

Well, actually, machines are kind of already in the business of writing laws. And there’s even an ongoing patent fight about them.

For years — longer than you may think — legislative offices have used software to help draft bill amendments and also to understand what new versions of a law actually do. More legislative bodies adopt such software every year. Even the U.S. House of Representatives uses it. (More on that later.)

To see how software is literally shaping our democracy, I called one of the leading companies in this space — XCential, makers of a product called LegisPro. CEO Grant Vergottini and President Mark Stodder offered to sit down with me to explain what, precisely, their software does. I brought along my POLITICO colleague Alec Snyder to ask the tough questions. (Alec, who analyzes bills from start to finish, has racked up many more hours puzzling over the arcana of U.S. Code than I have.)

Vergottini, originally an electrical engineer, got interested in the idea that you could build templates for designing laws, much like you can build tools for designing microprocessors. It simply required viewing laws as consistent structures of information. But the stakes are a little different. Drafting legislation “requires an extreme amount of precision,” said Vergottini, “because we are changing the law.”

Achieving that level of precision is not simple. To start, XCential models all the existing laws by a legislative body as a “big hierarchical structure” so that “every single piece of the law can be referred to in a precise way,” Vergottini said. (These are not small projects: As part of a 2018 contract, XCential converted the entire U.S. legal code into an XML data structure.)

From there — in automation terms — every new bill can be seen as a set of change instructions for that underlying information.

In Congress, that giant dataset allowed XCential’s team to develop natural language processing (NLP) tools that U.S. congressional offices now use to decipher how a bill would impact a current law, or how different versions of a bill compare. Writing laws in a machine-readable format also “allows government to be much more transparent,” Stodder added. For example, the Commonwealth Secretariat (which works closely with the U.N.) uses LegisPro to import laws from around the world into its search engine to track climate change legislation.

Like many applications of machine learning — including the ones in the news these days — law-writing tools are a lot more about pattern recognition than “thinking.” The XCential execs are careful to say they don’t see their software as a form of AI: “Lawyers get really nervous when we talk about anything bordering on intellectual processing of the law,” said Vergottini.

While the tools may be new, the idea is not. Oregon has relied on a customized IBM mainframe “for decades” to draft legislation, said Dexter Johnson, the Legislative Counsel for the state. But that mainframe is “kind of old and doesn’t integrate well with other software,” Johnson said. So the Oregon legislature gave XCential a $6.5 million contract to build a customized version of its LegisPro software in time for Oregon’s 2025 legislative session.

If it sounds like a good business — well, it is, and there’s currently a squabble over who gets the credit (and profits). XCential is currently locked in an intellectual property battle against D.C. lobbying giant Akin Gump Strauss Hauer & Feld, the powerhouse white-shoe law firm.

Akin Gump sued XCential in 2022, saying that XCential stole an idea developed by one of its lawyers, Louis Agnello. The idea, according to Akin’s lawsuit, was to build software that “would take proposed redline changes to an existing law and generate a draft bill in the format suitable for submission to Congress.” An Akin Gump spokesperson told me that Agnello’s idea would “revolutionize the process of generating amendments to existing federal legislation.”

At Akin Gump’s request, starting in October 2018, XCential began developing software to do just that. But, according to a petition Akin Gump later filed with the US Patent and Trademark Office, it took the company almost a year and Akin Gump chose to walk away, citing a “loss of faith” in XCential’s capability and a price tag for the software that was higher than they were expecting. Then in September 2019, XCential filed a patent for the tools it had developed. Akin’s partners say the idea belongs to them.

At the heart of the IP battle is a rather straightforward question: What constitutes a truly new idea? David Hricik, a law professor at Mercer University, was pretty skeptical about Akin Gump’s claim to having one in a blog post about the dispute. Drafting bills to meet Congress’ required “replace x with y” format was not a “revolutionary” idea but a widely known one, Hricik wrote, adding that “under the trade secret laws I’ve looked at, knowing a problem exists isn’t a trade secret. It’s the solution that might be.”

Akin Gump’s spokesperson said the software company “breached a binding non-disclosure agreement, and brazenly filed a patent application based on Agnello’s idea,” and said the firm is “standing up for what is right.” Vergottini disputes Akin Gump’s claims, saying the software he filed a patent for was specific to LegisPro’s inner workings. “We did not patent the world,” he told me, only “the parts that are novel in our implementation.”

If nothing else, XCential’s multi-decade foray into building legislation drafting software is a reminder that — despite the seeming breakneck releases of revolutionary software lately — getting to the “future” can be an incremental, iterative process. And when not slowed down by IP battles, that cautious pace is not necessarily a bad thing when the technology is meant for a space where getting it even slightly wrong can wreak real-world harm.

europe rolls up its sleeves on ai

Our European counterparts broke down for Pro subscribers today what they accurately called a “chunky” 30-page guide to upcoming discussions of the EU’s AI Act, providing a few key takeaways.

The issues up for discussion as the Act nears its fruition include its overall scope, what qualifies as “AI,” which kinds of systems are then prohibited as such, and which systems qualify as “high-risk” and are therefore subject to tighter regulation. Which all might sound… like pretty broad, high-level discussions for a piece of legislation that’s supposedly nearing its final draft text.

Europe’s Morning Tech authors note that it’s “a lot of heavy stuff for MEPs to go through,” and “one wouldn’t be surprised if some of it is not dealt with on Wednesday.”

There are a few relevant specifics for Americans keeping an eye on what the first big continent-wide attempt to legislate the technology might look like: Namely, language that would “outlaw biometric categorization, predictive policing, and systems creating or expanding face databases by indiscriminate scraping from social networks or CCTV footage.” — Derek Robertson

bing's bumpy demo

Last week, Microsoft showed off its shiny new AI-assisted update to its search engine Bing, hoping to get a leg up during a rare moment of vulnerability to Google’s search supremacy.

They might want to do a little more beta testing. In a blog post published yesterday, researcher Dmitri Brereton ran down a laundry list of factual errors Microsoft’s tool made, including:

  • Describing the Philadelphia Eagles as the winner of last weekend’s Super Bowl (before it happened)
  • Providing misleading descriptions of Mexico City nightlife
  • Botching the numbers in a financial statement for clothing company Gap Inc.
  • Removing Croatia from the European Union

Among other things. Microsoft’s director of communications told The Verge that the company is “expecting that the system may make mistakes during this preview period, and the feedback is critical to help identify where things aren’t working well so we can learn and help the models get better,” which makes sense for such a nascent technology.

For now, the bumpy demo is mostly a reminder that even the most seemingly all-knowing generative models are still accountable to their all-too-human designers. — Derek Robertson

tweet of the day

the future in 5 links

Stay in touch with the whole team: Ben Schreckinger ([email protected]); Derek Robertson ([email protected]); Mohar Chatterjee ([email protected]); Steve Heuser ([email protected]); and Benton Ives ([email protected]). Follow us @DigitalFuture on Twitter.

If you’ve had this newsletter forwarded to you, you can sign up and read our mission statement at the links provided.