I'm interested how the immortality questions would change if you asked whether the world would be better if everyone had the option to "to live forever in perfect health and youth". It is notable due to how much support it's getting compared to the other transhumanist proposals on the list, and I think the "would you" framing is the cause.
For instance, I think people would also say they'd like an option to "change their physical appearance to whatever they want" or make themselves much smarter whenever they want. Its just they don't think the world would be better if everyone could do that.
This would also explain why the numbers don't budge much if you add the overpopulation stipulation. If people are interpreting the question as only affecting their own actions, why would overpopulation matter?
cool survey! im surprised x-risk isnt even a thought amongst the top worries on ai. it really is a bubble
id be interested to see this on a larger sample size, or a different population (EU, urban v rural etc) maybe ill run this myself...
maybe thirty something percent support just about anything and about half that number are opposed
In late 2024, I was on a long walk with some friends along the coast of the San Francisco Bay when the question arose of just how much of a bubble we live in. It’s well known that the Bay Area is a bubble, and that normal people don’t spend that much time thinking about things like AGI. But there was still some disagreement on just how strong that bubble is. I made a spicy claim: even at NeurIPS, the biggest gathering of AI researchers in the world, half the people wouldn’t know what AGI is.
As good Bayesians, we agreed to settle the matter empirically: I would go to NeurIPS, walk around the conference hall, and stop random people to ask them what AGI stands for.
Surprisingly, most of the people I approached agreed to answer my question. [1] I ended up asking 38 people, and only 63% of them could tell me what AGI stands for. Some of the people who answered correctly were a little perplexed why I was even asking such a basic question, and if it was a trick question. The people who didn’t know were equally confused. Many simply furrowed their brows in confusion. Some made a valiant attempt—I heard a few artificial generative intelligences and even an Amazon general intelligence.[2]
Judging from the response I got on X (the everything app), this was a very surprising outcome. I ended up running this experiment again at NeurIPS 2025 with an even bigger sample size (n=115).[3]
After this first experience with surveying people, it became clear to me that the next step was to venture further outside the bubble, and survey the general US population. It turns out that this is already somewhat of a solved problem. A lot of people care about what the average American thinks. The market, in its infinite wisdom, has provided a solution—you can just pay pollsters to run random questions.
It’s impossible to actually sample from the distribution of all Americans. So you find some other approximate distribution, such as the distribution of all people who answer polls on the internet in exchange for amazon gift cards. You ask them a thousand demographic questions, like “how old are you” and “how much money do you make”. Then, since we know from the US census what these distributions are supposed to look like, you can correct for the distribution shift using importance sampling.[4]
With this caveat in mind, I embarked on a journey to ask normal Americans a bunch of weird questions using one of these polling services.[5]
The first question I ran in early 2025 was about how Americans feel about living forever (or at least, a very long time). I’m personally a big fan of not dying, so I was very curious to see how my fellow Americans felt about this.
The exact wording is “If you had the option to live forever in perfect health and youth, would you choose to? (Assume you could still change your mind at any time if you ever got bored of it.)”, and the possible responses are “Yes”, “No”, and “Not sure”.
Before you read further, take a guess at the result.
.
.
.
.
(don’t peek)
.
.
.
.
.
I anecdotally had the sense that this was a deeply unpopular opinion; certainly many of the people I talked about these results with thought it would be deeply unpopular. So I was surprised and relieved to find that actually 66% of respondents said Yes, with 14% saying No, and 20% saying “Not sure”. As a follow up, it turns out roughly a third of Americans think developing the technology to enable life extension should be a top priority.[6]
To really get a sense of why people felt the way they did, I also put free response boxes for people to express why they’d want to (or not want to) live forever. The results are enlightening; here are some of my favorites:
Of course, not everyone is as optimistic that living forever would be good. Here are some of the things people are worried about:
Since it seemed like overpopulation and inequality were the main things people were worried about, I also asked a version of the question where I stipulated that these things were solved. Surprisingly, this barely shifts people’s opinions, and we get almost exactly the same response! My guess is this is a sign that the real objection is more about the vibes than any specific issue. It’s also a sobering reminder of the limitations of this methodology.
After the results for this experiment came in, I decided to test a bunch of other random weird beliefs. If you want to guess at these before seeing the results (or you’re curious what the exact wording is, because that can substantially change the result), click here to see all the questions I ran before scrolling down further. If you’re willing to spend a lot of time looking at a giant wall of questions before continuing with the rest of this post, it’s really a great way to test your calibration.
.
.
.
.
.
.
.
First, despite being very pro living forever, Americans are much more skeptical of cryonics—even if they could be revived a few decades after their death to live forever thereafter, only 27% are in favor of being preserved, and 46% are opposed (the rest are unsure). Space colonization also has pretty lukewarm support, coming in at 37% in favor and 16% opposed, and cognitive enhancement for all is only a little bit more popular (42% in favor, 19% opposed). Also, for some reason, people are really opposed to a hypothetical cheap, painless, and safe arbitrary modification of physical appearance (only 23% in favor, with 37% opposed!).[7] In retrospect, the backlash against Ozempic is a sign, but I was still quite surprised. Terraforming other planets so that humans can live on them is also pretty unpopular, coming in at 37% in favor and 16% opposed. Thankfully, for most of these questions, a huge chunk of people are still undecided.
One of the most surprising results to me was that only 51% of Americans are in favor of literal post-scarcity (complete freedom to work on anything you want, as much as you want, and still enjoy a high quality of life), with 25% opposing. I was so shocked by this result not being 80%+ in favor that I reran a variant of this question with different wording. My original question asked whether the world would be better or worse if everyone had the freedom to work on whatever they want, as long as they want, and still enjoy a high quality of life, and anything we don’t want to do is done for us by robots. I thought maybe that set off some “AI taking jobs bad” instincts; for the new question I took pains to clarify that the stuff is literally conjured out of nowhere with magic and is not taken from anyone else, and got an even worse result (38% support, 34% oppose). This is even more crazy, so I ran a third version on the hypothesis that people don’t like magic, or that not having to work sounded too crazy. This version asked whether it would be good if everyone made 10x more (inflation-adjusted) than they do currently. This polled only somewhat better, with 39% in favor and 19% opposing. I’m still pretty confused what conclusion to draw from this; this is probably worth digging more into.
Tying back to the original question that started this quest, I had to know: how much are Americans feeling the AGI? I could of course ask if Americans know what AGI stands for, but some early results from asking random people on the streets spatially and temporally further away from NeurIPS suggested that the number would round down to 0%. So the more interesting question is; given a description of superhuman AI, do Americans think it’s possible?
It turns out that when I first ran this poll in mid 2025, only 25% of people thought AGI would ever be possible. That’s only half a year ago in normal people time, but an unfathomably long time in AI-land, enough for empires to rise and fall, models to be deployed and obsoleted, and even a single entire ML conference review cycle to run its course. Since then, more Americans have started feeling the AGI; a recent rerun of this question came out 10 percentage points at 35%. I’ll see you all again in another half year for the followup.
Should we build the AGI though? It turns out that people are extremely opposed to the idea of building superintelligent AI. Only 6% think it would be a good idea, and 75% think it’s a bad idea. I’m curious to see how this one changes over time too.[8]
AI existential risk also doesn’t seem to have become politically polarized yet. Two-thirds of Americans don’t associate AI x-risk with any particular political party, and the remaining third is split exactly in half on whether preventing AI x-risk feels like a Democratic or Republican issue. If we plot this data, we obtain this unusual shape that science has yet to find a name for:
As for specific risks from AI, Americans are most worried about misinformation and deepfakes (70%), followed by fraud and cybercrime (66%), and privacy and surveillance (59%). Surprisingly, people are roughly as worried about losing control of AI (57%) as they are job loss and lower wages (56%)! I would have thought that job loss would feel very near at hand, whereas loss of control would be a very weird abstract idea to people. There’s a huge drop off from there to the next biggest worries: military use (37%), mental health (35%), environmental impact (38%), bias (36%), and inequality (36%). My guess is this is because misinformation and deepfakes feel very visceral—fake news is a widespread idea, and you don’t have to be an AI connoisseur to notice that large sections of the internet are now filled with AI generated slop.
A few people also filled out the “other” box for specific risks they’re worried about. My favorite response was a shockingly accurate description of how hopeless it would be to fight back against superhuman AI:
Me too, buddy. Me too.
What about going further afield of AI? The beautiful thing is that you can just ask whatever you want.[9]
First, more broadly, Americans are very pessimistic about the future. Only 14% think that society is currently trending in a positive direction.
On a brighter note, I was able to a disprove a viral ragebait tiktok about how Americans would fail an English test meant for people learning English as a second language. I was proud to find that a good solid 85% of my fellow Americans got the problem from the tiktok right.
Because we love decision theory in this house, I wrote a question that explains Newcomb’s problem and asks whether to one-box or two-box. Americans are pretty split on this one; among the respondents who didn’t select “not sure” (honestly, kind of valid), only 46% would one-box. This is almost exactly the same as professional philosophers, who came out 44% in favor of one-boxing, according to a survey conducted by PhilPapers.
I was also curious whether people who are famous in SF are also famous among normal people. It turns out 36% of Americans know who Sam Altman is and can correctly say that he’s known for being an entrepreneur. Another 59% haven’t heard of him or don’t know what he’s known for. Honorable mentions to the remaining 5%, who think that Sam is a musician, actor, or congressperson. This same methodology finds that only 7% of Americans know who Geoffrey Hinton is, and 91% of Americans know who Elon Musk is.
I was in a discussion about whether lab grown meat would ever become widely adopted, so I asked a question about whether it would be a good thing if we could somehow create meat by growing it directly, without needing to raise and slaughter animals. It turns out 32% of Americans are in favor and 29% are opposed. When conditioning on the 55% of people who think meat production involves subjecting large numbers of animals to inhumane conditions, this tilts to 44% support and 21% oppose. I only ran the correlational study because it’s a lot easier, but I’d be interested to see whether there is a causal result on support for lab grown meat after you show people an educational video about factory farms.
Finally, for shits and giggles:
Unfortunately, only 32% got this one right. For comparison, 42% thought the answer was X Combinator, and only 6% went for W Combinator. Elon, if you’re reading this, I have a great business idea for you.[10]
What do we learn from all of this?
First, touching grass is great. At least, the kind of grass that grows on the beautiful rolling hills of cyberspace.[11]
What do I mean by this? Making contact with reality is important, and you don’t need to speculate about things when you can test them (carefully). Polling feels like a thing that only serious respectable people do, but you can actually just do things. There are a lot of limitations to this methodology, of course—all of us have divergences between our stated and revealed preferences; our own guesses as to how we’d behave in various hypotheticals can be an unreliable predictor of how we’d actually ask; wording can have huge impacts on how we respond; and respondents can be trolling us. But as long as we keep these limitations in mind, we can still draw useful conclusions, and learn new things about the world.
This one experiment singlehandedly reduced my social anxiety by a nontrivial amount. It turns out that most people are pretty friendly!
I also made sure to have another person with me whenever I was collecting data, and noticed some pretty big differences in enthusiasm of responses depending on who I was with.
I also ran a smaller p(doom) survey at ICLR 2025 (n=21) and found a mean of 18.7% and a median of 10%.
In general I’m pretty skeptical about controlling for things, but it’s way better than anecdata. For shits and giggles, I did do a few surveys of random people irl (e.g asking people who were walking through Central Park), but never anything at large scale.
Unfortunately, the specific pollster I used has a ToS that prohibits me from posting results from their platform in association with their name; presumably they don’t want anyone to leech off their reputability without paying stacks of cash for the enterprise tier or something. I tried to email them to pay more for the enterprise tier but I never got a response. So I’m not going to mention the name which pollster I used and you’re going to have to take my word that I didn’t just make these numbers up wholesale.
Exact wording was “Should developing the technology to greatly extend healthy youthful life be a top priority for humanity?”. 35% said Yes, 35% said No, 30% said Not sure.
I actually ran two variants of this question, one where I emphasized specifically that you could make yourself look like a celebrity (to help make the idea more concrete), and the other where I only mentioned some abstract characteristics like height, body type, and facial features, and got results within 1 percentage point of each other.
I also ran a variant of this question about recursively self improving AI. For that variant, 12% think it would be a good idea, and 69% think it’s a bad idea.
To be fair, the specific polling provider I ran these questions with has a review phase where they refuse to run certain questions. For example, they were really unhappy about my “would you live forever” question having a “you can die at any time” clause, so I had to replace it with a less direct “you can change your mind at any time”. But I’m sure you can find some other provider who would happily run these questions.
X (the everything combinator).
I do also touch normal grass.