Aiming to address a lingering backlog of unemployment claims, Nevada has partnered with Google to implement an AI system for reviewing transcripts of unemployment appeals hearings.
Boosting Efficiency
The AI, built with Google's Vertex AI Studio, is set to significantly speed up the time it takes to issue determinations. Currently, this process can take several hours, but with the new system, the goal is to reduce it to just five minutes. The integration was listed in a Nevada budget document. Nevada DETR Director, Christopher Sewell, emphasized that human referees will still review AI-generated recommendations for accuracy and fairness, either finalizing them or making adjustments as needed.
Experts, including Morgan Shah from Nevada Legal Services and Michele Evermore, a former U.S. Department of Labor official, have expressed concerns about maintaining thorough human reviews to avoid potential biases introduced by automation. Google's Ashley Simms mentioned the company is working closely with Nevada to address these issues and adhere to regulations.
Data Privacy and Model Integrity
Under a $1.38 million contract approved by the Board of Examiners, Google's AI will operate without accessing personally identifiable information. Instead, it will be fine-tuned using DETR's database, which encompasses Nevada unemployment law and past appeals cases, enhancing the accuracy of its recommendations while safeguarding data privacy.
A governance committee will oversee the system's operation, meeting weekly initially and then quarterly to check for inaccuracies and biases. Generative AI models predict text based on learning from training data, which can sometimes lead to incorrect or misleading responses. Research indicates that available retrieval-augmented generation models can misfire a significant percentage of the time.
Addressing the Backlog
Nevada's unemployment system faced an overwhelming surge in claims during the pandemic, leading to a backlog of over 40,000 appeals by 2023, which has since been reduced to fewer than 5,000. Amy Perez, who has managed similar modernization projects, believes appropriately managed AI can help mitigate such delays.
Nevada is pioneering the use of AI to speed up decisions on unemployment benefits. The system will analyze data from appeal hearing transcripts and rulings, uploading it to the cloud for comparison to previous cases.
