AI agents raise new safety questions

GREEN BAY, Wis. (WBAY) - Policymakers from Washington to Wisconsin are weighing new guardrails around artificial intelligence. Part of that conversation is being driven by what AI can now do on its own.

One area getting attention is AI agents. They can do something different from generative AI that produces an answer, image or writing.

An AI agent can be given a goal, then work through multiple steps to try to accomplish it without a human approving each one. That greater autonomy is part of what’s driving questions about safeguards.

Researcher tracks AI’s societal impact

Paul Belanger is a UW-Green Bay teaching professor and director of its AI Think Tank. For more than two years, the group has studied AI beyond the technology itself, including what its development means for people and society.

“Rather than addressing, well, what’s the potential? What can it do for us? And now we’re seeing those behaviors really start to emerge. And I think that’s driving the conversation now, saying, all right, we were thinking about this three years ago, but now we need to do something about it,” Belanger said.

But Belanger says that doesn’t mean these systems have developed intent or are thinking for themselves. The concern is how much they can do while pursuing a goal, and whether the safeguards around them can keep up.

Lawmakers respond

Within the hour, Sen. Tammy Baldwin called for Congress to act on AI safety, including independent testing of advanced models.

We’ve reached out to Sen. Ron Johnson and Congressman Tony Wied for their position on AI guardrails and have not heard back yet.