Washington is fumbling through a slate of potential artificial intelligence regulations — some focused on global competition, some on AI-generated deepfakes and some arguing that the government should get its arms around how it’s using AI before it tells anyone else how to do it. All the while, the tech continues to rapidly evolve with little oversight. The most sweeping plan on the table for now is Senate Majority Leader Chuck Schumer’s proposed “roadmap” that sets out bipartisan legislative priorities, but that’s more of a plan to have a plan, and less an actual piece of legislation. In a statement Thursday, Schumer called the bill focusing on government procurement and use of AI, co-sponsored by Sens. Gary Peters (D-Mich.) and Thom Tillis (R-N.C.), "essential for the federal government to deploy AI so it protects people’s civil rights, prevents bias, and ensures people’s privacy." What lawmakers put together will depend in part on what they think will make their constituents happy — and a new poll offers a window onto their thinking. A survey from the pro-regulation Artificial Intelligence Policy Institute, shared exclusively with DFD, shows that voters largely feel positive about the roadmap’s regulatory priorities but might still need a bit more education about what they would actually do. AIPI pollsters asked voters how they feel about the roadmap’s tenets including research funding, safety testing and labeling false information. Overall, more than half supported the AI roadmap with only 11 percent saying they oppose it. The survey also picked voters’ brains about a few other AI-related policy developments, and recent accusations of lax safety practices at OpenAI. When it comes to the principles laid out in the Senate AI roadmap, 45 percent of respondents supported “federal funding to advance AI research and keep the U.S. ahead on AI,” with 19 percent opposing. Just over half supported “federal funding to safety test advanced AI systems to reduce risks,” with 16 percent opposing. And nearly three-quarters of people surveyed said they support “creating privacy protections when AI is used” and 8 percent opposed. Still, voters still do not feel confident in their opinions, or knowledge, about AI. Large chunks of voters responded “not sure” to AIPI’s questions about the roadmap. About a third were unsure in response to the question about research funding; just more than a third were unsure if they supported “planning to responsibly integrate AI into critical infrastructure”; and 32 percent were unsure about “establishing copyright protections in the age of AI.” More than a third of voters said that after hearing about what’s in the roadmap, they’re still unsure whether they support it or not. Unsure responses cut relatively evenly across all demographics — except for political party. A consistently smaller share of Democrats than Republicans and independents responded “not sure” to questions, which could indicate as in other recent polling from AIPI their relative techno-optimism and engagement with the topic. Most voters are worried about the use of AI in this year’s elections, but have mixed feelings on what to do about it. Asked “how concerned” they are about “misleading AI-generated content” being used in political ads this year, nearly three quarters of people polled said they were either “very” or “somewhat” concerned. But asked about the FCC’s proposed rule to require disclosure of AI use in election-related ads, support dropped to 64 percent, with another quarter of respondents saying they’re not sure whether it’s a good idea. Asked about whether the AI disclosure rule should be extended to all ads, 62 percent of respondents supported the idea; another 30 percent were unsure. There’s an enthusiasm gap between Democrats and Republicans on the Senate AI roadmap. Perhaps unsurprisingly for a legislative package that, while bipartisan, was introduced by the Democratic Senate Majority Leader amid a Democratic presidential administration, Republicans consistently lagged their counterparts in enthusiasm for the roadmap’s proposals. Seventy-three percent of Democrats supported “training workers for the AI economy,” while 41 percent of Republicans did. When it came to federal funding for safety testing of advanced AI systems, nearly three-quarters of Democrats supported it compared to 38 percent of Republicans. One place where they found common ground: “Monitoring to prevent dangerous AI from falling into the hands of our adversaries,” which enjoyed the support of 84 percent of Democrats and 72 percent of Republicans. No, really: The national security rationale for cracking down on AI companies is a potent one. After a former employee recently accused OpenAI of safety lapses that could make the company vulnerable to Chinese or North Korean espionage, 76 percent of respondents said they were “very” or “somewhat” concerned about the accusation. Eighty percent said they supported “a law that mandated that all AI labs constructing the most powerful AI models must use state of the art security to prevent the models from being stolen by hostile actors,” and 69 percent said they would support a law that would make it illegal for AI companies to “retaliate” against leakers concerned about AI safety. The poll interviewed nearly 1,500 respondents online over the first few days of June, with a 5-point margin of error. Read the full toplines and crosstabs.
| 
