A poll has shown voters in California to support SB-1047, a bill that would place some basic requirements on the most powerful AI models.
The Artificial Intelligence Policy Institute (AIPI) found that 59% of voters support the bill, whilst only 22% oppose it.
If passed, SB-1047 would only apply to models above a certain threshold, requiring the companies training them to carry out safety tests and making them liable for catastrophic harms.
Figures in AI, such as Yann LeCun, who goes against the majority of his colleagues with his view that the chance of catastrophic outcomes from AI is effectively 0%, have opposed the bill, claiming that ‘regulating basic technology will put an end to innovation’.
Others have argued that, whilst the bill is a step in the right direction, it doesn’t do nearly enough to protect us from the existential risk posed by training frontier AI. The snail-like pace of progress on the alignment problem, compared to the rapidly diminishing set of tasks just about still out of the reach of AI, means such safety measures as those proposed in the bill are unlikely to be effective against smarter-than-human models.
When asked their opinion on specific elements of the legislation, the majority of Democrats, Independents, and Republicans all overwhelmingly came out in favour of the following measures:
- Require safety measures for advanced AI models to prevent hazardous capabilities and risks to public safety. (74%)
- Mandate cybersecurity protections, testing, safety protocols, and risk management for high-risk AI models. (73%)
- Require that models have emergency shutdown capabilities such that model developers can disable instances of their model in their position in case of an emergency. (73%)
- Hold developers of large models liable for catastrophic harms caused by their models. (70%)
- Require that AI companies test large models to determine whether they’re capable of catastrophic harm. (67%)
Daniel Colson, Executive Director of AIPI, said that the poll made it ‘abundantly clear’ that voters are in favour of strong regulation to mitigate the worst risks of AI.
“Voters from across the political spectrum—and professional backgrounds—overwhelmingly back public policy that ensures AI safety and prevents catastrophic harm. Policymakers must take note of this decisive public mandate for AI regulation.”
