Artificial intelligence (AI) software could be used to put words into politicians' mouths, a US Senate committee has heard.
Democrat Connecticut Senator Richard Blumenthal opened the Senate Judiciary Committee's subcommittee hearings into the rising tech phenomenon on Tuesday by asking online AI 'chatbot' ChatGPT how it thought he should address the room.
The software proceeded to deliver a speech it had composed itself, mimicking his voice. The senator admitted he was impressed — but also afraid it could be used to impersonate and misquote him.
"What if I had asked it, and what if it had provided, an endorsement of Ukraine surrendering or Vladimir Putin's leadership?" Blumenthal asked.
Sam Altman, CEO of ChatGPT's developer OpenAI, testified at the hearing.
"As this technology advances, we understand that people are anxious about how it could change the way we live. We are too," Altman said.
Committee chairman Dick Durbin, the senior Democrat senator for Illinois, said the government should create a new agency to police AI technology.
"We're dealing with innovation that doesn’t necessarily have a boundary. We may create a great US agency — and I hope that we do — that may have jurisdiction over US corporations and US activity that doesn’t have a thing to do with what’s gonna bombard us from outside," Durbin said.
Republican South Carolina Senator Lindsey Graham, the ranking member of the committee, supported the idea of a regulator that would issue licenses for AI systems.
Altman later told journalists that he supported both proposals — ominously comparing the them to the International Atomic Energy Agency (IAEA), the UN agency tasked with enforcing the Treaty on the Non-Proliferation of Nuclear Weapons (NPT).
"It's difficult. It sounds sort of like a naïve thing," the OpenAI boss said. "But we’ve done it for other industries. IAEA did it. And I think this is a technology we should treat with that level of seriousness."
Both Republican Missouri Senator Josh Hawley, who has previously argued for tighter regulation of social media tech firms, and failed 2020 Democrat presidential primary candidate Amy Klobuchar both raised fears that AI could be used to influence voters and spread misinformation during election campaigns.
"It's one of my areas of greatest concerns — the more general capability of these models to manipulate, to persuade, to provide sort of one-on-one disinformation," Altman responded.
New York University Professor Gary Marcus, a psychologist and cognitive scientist who has called for a six-month pause in AI development to assess the possible risks, will also testify before the committee.