The Nevada Department of Labor has announced that it will use artificial intelligence to speed up the unemployment benefits appeals process by analyzing transcripts of appeal hearings and making a recommendation decision.
The department said it has been “buried” under a pile of claims since the start of the pandemic and is desperately looking for ways to get back on track.
It has also confirmed that it will not be training a new generative AI model for the system, but will instead be using Google’s Vertex AI Studio, which will reportedly cut the review process from hours to just five minutes – despite new research suggesting that AI models in general worse than humans in ‘every way’ when summarizing documents and often results in extra work for employees.
Clearing the backlog
Experts caution against this approach, not only because Large Language Models cannot understand text or reason logically and within context as humans can, but also because it may not save the department much time.
“If someone goes through something thoroughly and correctly, it doesn’t save them that much time,” said Morgan Shah, director of community engagement for Nevada Legal Services. “At what point do you create an environment where people are encouraged to take shortcuts?”
Attorneys at Nevada Legal Services are concerned about the model’s lack of accuracy. They worry about AI “hallucinations,” an industry term used to describe when an AI model gives factually incorrect or misleading answers.
Furthermore, every decision the AI makes is double-checked and reviewed by a human referee before it is made. However, if the human referee makes a decision based on the AI hallucination, a court of law may not be able to overturn the decision.
The infamous IBM quote comes to mind: ‘a computer can never be held responsible, so it should never make a management decision’. Research has shown that many of us still very wary of AIparticularly in high-risk products (such as medical diagnostics and automated vehicles). The success of this experiment could impact a wide range of government departments in the future.
Via Gadgets