The biggest artificial intelligence firms train their products on oceans of public data, using massive amounts of energy — and a new poll shared exclusively with POLITICO shows Americans would like regulation on both. The Artificial Intelligence Policy Institute, a pro-regulation AI think tank, found that 60 percent of respondents said AI companies should not be able to train freely on public data. Nearly three quarters of people polled said companies should be “required to compensate the creators of that data.” And 78 percent said there should be regulations on the use of public data to train AI models, with Democrats only slightly more enthusiastic than Republicans. The online poll of 1,039 people was conducted on April 12 and 13. It’s the first time AIPI, a prolific pollster since its launch last August, has taken the public’s temperature on AI companies’ data training practices — as lawmakers agree that the boom in generative AI has made data training an urgent policy concern. Respondents also liked the idea of a special tax on electricity for AI companies, with about six in ten saying they would be in favor of the government collecting revenue “to support and upgrade the electrical grid infrastructure.” Just over half of respondents said they were in favor of the Biden administration’s proposed funding for the AI Safety Institute, albeit with a partisan split: 65 percent of Democrats supported it, compared to 47 percent of Republicans. The demographic breakdowns in this round of polling track overall with those from previous rounds from AIPI. Older voters are more likely to fear AI and call for its regulation than younger voters, and bipartisan consensus largely reigns, although it shows some faint cracks. In addition to polling on regulation, AIPI polled respondents on their personal attitudes about AI, including whether they worry AI will “soon” be able to do their work. The pollsters posed the question as a two-part experiment, asking it both before and after playing respondents a “gravelly blues song” generated by AI music firm Udio. At first, people didn’t seem to break a sweat about robots taking their place: Only 31 percent said they were either “very” or “somewhat” worried AI might be able to do their job, and 44 percent responded with a rather wishy-washy “not very worried.” After hearing the music, however, people’s attitudes seemed to flip, with 45 percent saying they were either “very” or “somewhat” worried — and a closer look at how their responses changed reveals the ways in which exposure to generative AI tends to leave someone with a strong impression, for better or worse. That’s because the number of respondents who outright fear AI — those who responded they were “very worried” — remained almost exactly static, decreasing by one percentage point. But “somewhat” worried jumped from 16 to 31 percent, and “not at all” worried also jumped from 19 to 30 percent. Those respondents therefore likely switched allegiance from the nondescript “not very worried” category, indicating that it’s difficult to walk away from an experience with modern generative AI tools without your opinion being sharpened. Which direction one goes might depend on education: College-educated respondents showed a big before-and-after jump, going from 21 percent “not at all worried” to 39 percent. The “somewhat worried” among non-college respondents increased from 13 to 30 percent. (For what it’s worth, this tracks with what your college-educated author described in a March edition of DFD: “The awful AI music that might take over the world.”) Read the full polling toplines and crosstabs.
| 
