Artificial intelligence is moving into government faster than almost anyone expected, reshaping how public agencies serve millions of people. The promise is enormous: faster services, fairer decisions. So is the risk of leaning on tools that don’t actually work.
“There’s a revolution happening in how AI is changing the work of government,” says Justin Milner, who leads Arnold Ventures’ Evidence & Evaluation team. “The question is whether it actually makes things better.”
For Arnold Ventures, the answer to that question is the same as it is for any policy claim. You test it. Stanford’s Regulation, Evaluation, and Governance Lab (RegLab) and Colorado Department of Labor and Employment did just that in one of the most consequential corners of government: the unemployment claim.
When the pandemic hit, 46 million Americans turned to unemployment insurance, and the system buckled at the worst possible moment. As RegLab director Dan Ho notes, the share of on-time claims decisions fell from above 90% to just above 50%.
Deciding claims is meticulous work, and the staff responsible often handle heavy caseloads. RegLab and Colorado collaborated on an AI tool to improve the efficiency and quality of unemployment insurance claims decisions, without removing the human element.
“Evaluation really is the critical thing to be able to tell the difference between a solution that can really be transformational versus AI snake oil.”Dr. Daniel Ho, Director of RegLab, Stanford University
They ran a rigorous evaluation, measured whether the tool actually helped, and shared the results publicly so other states could learn from them. “Evaluation really is the critical thing,” Ho says, to identify a genuinely transformational solution from those that fail to deliver on their promise.
That work drew national recognition. The RegLab and Colorado tool won the Rising AI Pilot category in the first AI in Action Awards from the Center for Civic Futures, a nonpartisan nonprofit focused on responsible AI in government.
Arnold Ventures supported this collaboration, alongside Stanford Impact Labs and Stanford HAI, because it shows how government can adopt new, proven technology at scale. A tool that speeds up unemployment decisions in Colorado could help other states, too, and the published evidence gives them reason to trust it before adopting it.
“It’s important for the country,” Milner says, “because this could be a model for other states.”
It comes back to a discipline Arnold Ventures stands behind. Test what you build and let the proof determine what scales. When public systems build on evidence, they serve people better, especially the next time a crisis hits.
For more information about Arnold Ventures’ funding opportunities, please visit our grants page.
