NY Attorneys See AI as Useful Immigration Case Gruntwork Tool

New York legal aid and nonprofit groups have borne the brunt of legal needs for millions of immigrants and refugees, and recent developments in AI translation tools have sparked the organizations’ curiosity.

There are about 2.5 million people in New York with limited English proficiency, but use cases like real-time translation are years off, as is live in-courtroom translation. And the groups say access issues as well as questions around the programs’ reliability make them works in progress.

For now, AI tools to help immigrants fill out forms or translate documents appear to be closest to implementation.

“Translation is a universal need—people in eviction court need help with translation. We’re hopeful in the next couple of months or years we’ll see things start to really accelerate,” said Sam Harden, program manager at Pro Bono Net, a New York-based nonprofit that provides technology to legal aid groups.

Lost in Translation

Language accessibility is key to enhancing access to justice, said Microsoft’s Adrián Palma, global pro bono manager and digital strategist.

“Wherever there’s a language need, there’s immense potential,” Palma said.

Microsoft is developing programs that could help asylum seekers and Deferred Action for Childhood Arrivals applicants by helping fill out forms and answering questions, and also streamlining the volunteer sign-up and training process for lawyers at DACA clinics. Microsoft is also building out a program for asylum seekers through its Copilot Studio.

Last year the American Immigration Lawyers Association and Visalaw.Ai announced a product that will specialize in immigration-focused administrative and case law similar to OpenAI’s ChatGPT. The organizations are working to get their research and drafting product, GEN, into the hands of immigration clinics nationwide for free, Josh Waddell, co-founder of Visalaw.Ai, told Bloomberg Law.

Some organizations are experimenting with using AI to translate plain-language text on public-facing websites, said Tim Baran, director of the Justice Initiatives Program at Pro Bono Net. He hopes AI can “bridge the gap” between human and AI translation and help offset the “incredible cost” of human translators.

Camille Mackler, founder and executive director of legal collaborative Immigrant ARC, said she wouldn’t use documents translated by AI without having a native speaker read through them first to confirm the translation is accurate, especially when it comes to legal language.

Legal terminology can be “unique,” which makes translation of legal documents even more complex, Harden said.

There are major consequences if translation is even slightly wrong, Harden said. He points to examples where even something as simple as a label for a product made in Turkey was labeled in Spanish “hecho in pavo,” which literally translates to “made in a turkey.”

“It’s not just literal translations but also the way there’s an emphasis on certain paragraphs or in a sentence,” he added. “We have used some machine translation in informational websites, but not on mission-critical sites.”

Real-Time Conversations

AI could someday be used to translate conversations in real-time. OpenAI last month debuted its GPT-4o, an updated version of GPT-4, which during a live demonstration was able to translate speech almost instantaneously.

Reliable live AI translation would expand live translation beyond the commonplace Spanish-to-English, said Bernadette Gargano, vice dean for experiential education and social justice initiatives at the University at Buffalo School of Law.

The city of Buffalo resettled one-third of the over 4,000 refugees and special immigrant visa holders that were resettled in the state in fiscal year 2023, according to data released by the New York Office of Temporary and Disability Assistance. More than 75% of refugees came from the Democratic Republic of the Congo, Syria, Afghanistan, and Burma—countries that have dozens of languages and dialects.

Justicia Lab Director Rodrigo Camarena said he’s excited about the future of live translation, but the organization is focusing for now on translation for documents.

“We are hopeful the models improve to such a level where we can just integrate existing models and offer document translation,” Camarena said. Justicia Lab, which is under the Pro Bono Net umbrella, recently got $500,000 of seed funding from Google to launch an AI lab aimed at bringing technology to immigrants and immigration legal service providers across the country.

He also doesn’t see New York courtrooms implementing translation technology anytime soon, but said they should be thinking about how it could fill in for humans in certain cases.

“I think courts should think about the capabilities of these technologies and whether we need to have humans in the courtroom in all cases, or whether it would be better served using those resources to hire more judges and reduce the immigration backlog, which is a longstanding issue nationally,” Camarena said.

Mackler said there should always be a human interpreter available for court, especially considering how nuanced and complex many asylum and immigration cases are.

“It’s difficult enough when a human is on the phone or virtual, but i don’t think you can ever get more removed than that, especially for an asylum case—that’s so nuanced with the words you’re saying, the tone, the clarity, and all of that.”

Work in Progress

Pro se litigants and newly-arrived immigrants might not have access to computers or internet, or may be illiterate, making it nearly impossible for them to access tools that could help translate or fill out legal documents, Gargano said.

The major technology companies that have invested a lot of time and money in developing AI tools haven’t done much to increase access to vulnerable groups and others who can’t afford the technology, or to make sure their programs are providing accurate information, Gargano said.

“We have to worry about folks who are unrepresented, whether or not they’re going to think, ‘I can just plug this in and it’ll give me the answers,’” Gargano said.

There are also privacy concerns with AI, especially with potentially feeding confidential information into the programs, as well as the risk of so-called AI hallucinations—inaccurate information, including incorrect case citations, produced by the AI program.

While there’s a lot of potential for using the technology to summarize and translate documents and provide plain-language information, users have to be careful about feeding personal information to a large technology company, Harden said.

“We’re very cognizant of that in what we do,” he said. “Our tech helps people serve very vulnerable populations—people who are at a disadvantage already—and we don’t want to make it worse. So we’re looking at how generative AI can help our products help those people in a safe and ethical way.”