Pullman

WSU builds a search tool so you can ask the law a plain question

Ever tried to read a Washington state law?

4 min readResearch
WSU builds a search tool so you can ask the law a plain question
David Makin

We have watched this program put up 73 points on a Saturday and then, on a Tuesday, drop a tool that might do more for the state of Washington than any single win ever could. The WSU Complex Social Interactions Lab just built an open-source AI model that lets you ask the Revised Code of Washington a plain question and get the right statute back. That is not a press release. That is a genuinely useful thing.

Let's be honest: nobody reads the RCW for fun. It is dense, formal, and written to withstand a court challenge, not to answer a city planner who needs to know if her short-term rental ordinance conflicts with state law. The old method meant either paying a lawyer for hours of cross-referencing or guessing keywords until Google gave up. The new model, built on Google's EmbeddingGemma 2 architecture, translates your normal question into math that the code understands. In benchmarks, it found the exact governing statute in its top ten results 87.9% of the time. That is not perfect, but it is fast enough to matter on a phone in the field. For comparison, the football team has been chasing a league win in Texas, Cougs broke a streak. Now they chase a league win in Texas, and that kind of focus on results is the same energy the CSI Lab brought here.

David Makin, the lab's principal investigator, said making the law more accessible is a pathway toward a more just legal system. He is right, but the practical upside is narrower and more immediate. City and county officials can check whether a proposed rule conflicts with state law without billing the taxpayer for a legal opinion every time. Law enforcement officers can look up a statute based on a scenario, not a citation number. State agencies can automate compliance checks. And because the team published both a full-precision model and a compressed version that runs on a standard laptop or phone, nobody has to pay a subscription fee or sign a corporate data-sharing agreement. Shlok Tomar, the lead AI engineer, said they wanted to keep this out of proprietary corporate walls. That is a very Coug move, build something good and leave it on the porch for everybody.

There is a catch, and the researchers say it themselves. The tool is designed to aid human legal judgment, not replace it. An 87.9% accuracy rate means roughly one in ten top results is wrong, and in a legal context, wrong is expensive. The model does not interpret; it retrieves. It finds the statute that matches the intent of your question, but it does not tell you what the statute means or how a court has applied it. That is still a job for someone who passed the bar. The risk is that speed makes people skip the human step. The team has been clear about the limitation, and we should hold them to it. Meanwhile, Kirby Moore finally cracks a smile after WSU's 73-7 shellacking, Kirby Moore finally cracks a smile after WSU's 73-7 shellacking, which is a reminder that even when the score looks dominant, there is always another layer to pay attention to. The same applies here.

The specific thing to watch is who builds the first public-facing app on top of this model. The CSI Lab put the files on Hugging Face. Any civic technologist, any county IT office, any student with a laptop and a weekend can spin up a legal search tool for their community. If that happens in a dozen small towns across the state by this time next year, this will matter more than any single play call. If it sits on a server and nobody uses it, then it is just a cool project. The lab did its job. Now it is on the rest of us to see if we can use it. Go Cougs.

From Research

PULLMAN, Wash. — Navigating the complexity of statutory legal code often requires hours of tedious cross-referencing, but a new artificial intelligence tool developed at Washington State University could make searching state laws as simple as asking a plain question.

Read the original at Research

Briefing

Source
Research
Published
November 18, 2026