Nevada deploys Google Vertex AI to review unemployment appeals
Nevada's Department of Employment, Training and Rehabilitation deployed Google's Vertex AI to analyze unemployment appeal transcripts and recommend decisions, reducing review time from three hours to five minutes. Experts have raised concerns about potential bias, hallucinations, and pressure on human reviewers to approve the AI's recommendations without adequate scrutiny.
What happened
Nevada's Department of Employment, Training and Rehabilitation began using Google's Vertex AI cloud system to analyze transcripts from unemployment appeals hearings and issue recommendations on whether cases should be approved. The model was trained on Nevada's unemployment law and policies and completes reviews within five minutes, compared to the three hours previously required by human reviewers.
Reported impact
Concerns have been raised about potential bias in the AI system and the risk that human reviewers may feel pressure to rubber-stamp the system's recommendations while clearing a backlog of over 10,000 appeals. If human reviewers miss the AI's mistakes, Nevada courts may lack legal standing to overturn the decisions, according to legal experts.
Index analysis
Michele Evermore, a former deputy director for unemployment modernization policy at the U.S. Department of Labor, stated: "If a robot's just handed you a recommendation and you just have to check a box and there's pressure to clear out a backlog, that's a little bit concerning." State Senator Skip Daly questioned the decision, asking: "Are we out of our ever-loving minds?"
Governance controls
A governance committee was established to meet weekly while the model is being fine-tuned and quarterly after deployment to monitor for hallucinations and bias. Human reviewers are required to review the AI's recommendations and decide whether to honor them before a decision is issued.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.