Nevada has a new helper in processing its backlog of unemployment claims: Google AI. As reported by Gizmodo, the effort will have one of the company’s cloud-based AI models analyze appeals court records and recommend whether cases should be approved. A robot could decide whether or not you should receive the government money you applied for.
The Nevada Independent reported in June that an AI model trained on the state’s unemployment laws and policies would analyze transcripts from virtual appeals hearings, then issue a decision that state officials would review for errors and decide whether to honor.
It replaces the Nevada Department of Employment, Training and Rehabilitation’s (DETR) current process, which takes an average of three hours to do manually. DETR IT administrator Carl Stanfield told the Nevada Independent that Google’s AI (using the company’s Vertex cloud system) can make a decision in under five minutes. “The time savings is just phenomenal,” Stanfield said.
It’s easy to see why Nevada is eager to turn to this emerging technology: As of June, the state reportedly had a backlog of more than 10,000 appeals, about 1,500 of which were a backlog caused by the pandemic. And if the technology’s reviews are accurate, or if it helps human reviewers catch mistakes, it could save huge amounts of time.
But there could be psychological pressure for employees reviewing cases to sign off on the AI’s conclusions. “If a robot is handing down a recommendation and there’s pressure to clear the backlog just by checking a box, that’s a bit concerning,” Michelle Evermore, a former deputy director for Unemployment Insurance Modernization Policy at the Department of Labor, told Gizmodo.
Stanfield told Gizmodo that a governance committee would meet weekly while the state fine-tunes the model, and quarterly once it’s up and running to monitor for artifacts and bias. The risks for plaintiffs could be significant, since AI-powered systems could affect a plaintiff’s ability to appeal a false decision. “In cases involving factual issues, a district court cannot substitute its own decision for the appellate judge’s decision,” Elizabeth Carmona, a senior attorney at Nevada Legal Services, told Gizmodo. In other words, if a human reviewing the decision misses an AI error, the court may not have the legal authority to overturn it.
One Nevada politician put it a bit more bluntly: “Are we really going crazy?” Nevada Sen. Skip Daley (D-Reno) told the Nevada Independent this summer. “I question the whole notion of relying too heavily on algorithms and computers. I hope that we’ll be careful about it and think carefully before we say, ‘We have to be faster or better than other people.'”
