That does not sound like voters being willing to talk about existential risk from AI. It's voters being willing to talk about AI risk in the sense of people abusing AI in mundane ways, which is not the same thing.
I'm surprised I can use phrases like "could lose control over" at all. Then I talk about a candidate who's doing something closer to an AI safety agenda, including "accountability for mass damages", instead of x-risk irrelevant priorities like banning chatbot medical advice.
I'm not pushing my luck... like I'm not asking people to agree to something like "I'm more worried about human extinction than unemployment". But I'm discussing AI safety agenda, and getting there by saying I'm worried about "not controlling" AI. I think that's un-mundane.
Most of the anti-AI sentiment has to do with its use by large companies, governments, employers, etc. All of these are outside of the personal control of anyone you're talking to, and that's what "loss of control" over AI means to normies. AI actually taking over the world is such a weird idea that most people wouldn't even think of that unless they are talking about a movie.
I'm reminded of the UFO enthusiasts who saw that a lot of astronomers are concerned about "the UFO problem", and concluded that astronomers think UFOs are spaceships. What it actually meant was that astronomers rejected the spaceship idea so strongly that to an astronomer, "the UFO problem" means "the problem that people keep believing in UFOs".
Anecdotally when I talk to normal people about AI the main questions are "Will AI take my job?" and "Will AI take over the world / become Skynet / etc?".
Although job concerns are taken more seriously than loss of control.
I've thought of a snappy reply to the Skynet thing which is like -- Skynet was sci-fi because we don't give computer programs we don't understand control of our nukes systems. It wasn't sci-fi because there could be advanced programs we don't understand. Well now we're doing that.
It's not meant to be deep it's kind of pointing out, hey, you already agree whatever this tech is, we don't completely understand it -- that's a much lower bar than "tries to break containment" and so forth -- and we are doing many of the sorts of things you should not do with an ill-understood tech.
When I say "surprisingly open" I do not mean "I explain the paperclips thing then we align on p(doom)". I mean things like: "we could lose control over it", "the labs developing it don't even know how it works," "they're going way too fast with no regard for safety," "they're in an arms race and I don't know where it's going." Or more simply "I am alarmed, do you share in my alarm?" This is "surprising" because my 2025 prior is "it's a word predictor" and "it's a bubble."
I also claim this is exactly where you want the conversation. Most of the heavy lifting is done. People will just demotivate if the conversation is about how everyone will die soon, and I don't think "your vote has been estimated to reduce the chance of human extinction as much as 4 minutes of technical AI safety research" is going to work.
This would bite us if we had a credible "stop" movement, because a full stop will now cause recession. That would force a split between jobs-harm and extinction-harm. I will point out that political alliance building epistemics is mostly about UFO enthusiasts and astronomers figuring out they both want better telescopes, not aligning on p(aliens).
I'm actually not very surprised about this. AI has never polled that well in any Western country, and is doing steadily worse.
Eventually we will get a group of voters in the "I use AI all the time at work, and while they laid of a bunch of junior people and old stick-in-the-muds, I'm doing just fine telling the AI what to do!" and who actually need to be persuaded that they're only fine until they're next laid off — and that that won't be very long. But for now the pro-AI faction in is mostly billionaires and their astroturf.
I'm encouraged but not very surprised. I think there's a faulty assumption in the AI risk community that public attitudes toward AI will never change. Ironically, that's the same mistake the public is making about AI.
But AI will change, dramatically, to have more agency and competence. And public opinion will change in response.
We'll see A country of alien idiots in a datacenter, and that will be alarming in a very intuitive way: a species, not a tool. Let's hope there's still time for public opinion to slow down AGI when it changes.
Do others in your group share the same opinion? By the example, I feel you may be above average at engaging people.
Explicitly using x-risk language, no, just me that I know of. Definite yes, to high appetite to talk about AI. Definite yes, to job loss, so that I think "mass job loss" is within public consciousness now.
There's a fallacy/bias at play here. At some psychological or mythological level, people basically conflate "recessions" with "the apocalypse." So "mass unemployment" and "lose control and kill everyone" are pretty near in the Overton window, so if we're up to mass unemployment, we can start talking about "safety testing" even though that's not directly relevant to unemployment.
Not mentioned in my post, and should have in retrospect, but Bernie Sanders has taken interest in AI-caused human extinction, explicitly, using numbers like "10% or 20% chance" in interviews. He is very respected with the left and with some smaller number of right-populists so it's important to see if he's successful in normalizing the topic.
Oh lastly -- I came in with more canvass experience but numbers and conversation quality are pretty consistent between me, and the other AI safety motivated volunteers. The fundamentals are doing most of the work.
TL;DR: Voters are now surprisingly open to talking about existential risk from AI. This seems to have changed in the last 6 months. When campaigning for AI safety-friendly politicians (e.g., Alex Bores), we should talk more about AI in general, and about AI risk in particular. This is currently actionable for the CA-11 and NY-12 Democratic primaries. I include concrete advice to turn basic conversations during political canvassing into persuasive conversations centered on AI risk.
Public opinion around AI has rapidly soured in the 12 months. According to a March 19-23 Quinnipiac poll,
Anecdotally, I've noticed more willingness among non-AI-focused media to discuss widespread harm from AI. Most visibly, gradual disempowerment is a hot topic (NYT), and right-wing pundits like Steve Bannon have supported Anthropic's red-line against lethal autonomous weapons. Memorably, my cousin, a county commissioner in a rural area, has told me about farmers showing up at city council meetings, sending emails, and posting on Nextdoor in opposition to nearby data center construction.
Turning this sentiment into constructive x-risk-reducing policy, as opposed to, for instance, ill-advised bans on AI in mental health, will likely be incredibly important. So, electing competent politicians friendly to x-risk seems like a very leveraged intervention.
Since 2020, I've done ~160 hours of persuasion-focused canvassing for various campaigns. I live in NY-12, and so have spent around 5 shifts over the past month canvassing for Alex Bores. I'm pretty good at getting someone stopping to read my campaign flyers to chat a bit more, and have about 4-8 conversations per hour that I consider persuasive. By persuasive, I mean I'd predict they're at least 15% more likely, in my opinion, to vote for Bores, and that they take either take campaign literature or give me their name and zip code. This is a very high number - I'm usually bottlenecked on my attention, not waiting for voters to talk to me. Here's what an average conversation looks like:
I've never experienced this sort of issue traction in my canvassing experience. About 90% of my voters have the full conversation with me, talking about my issue, in my language, on my terms. This is surprising! Usually canvassing is about connecting what the voter cares about to the candidate.
Therefore, I think people that care about existential safety should put more effort, on the margin, into talking to voters about existential risk. At the very least, political engagement with catastrophic risks seems easier than expected with college-educated voters, and can be a tractable way to put AI safety-friendly politicians in power.
Appendix: How to talk to voters
I'm part of a group of rationalist/EA/AI safety folks volunteering for the Bores campaign; you can join us by filling out this form!
Tips for talking to voters about AI: