I think you're right to mention polarization as a risk, and I think this should be a top priority. The problem is that if one side gets really excited about fighting AI progress, the other side will fight them on it, just out of habit and pragmatism.
I think the left comes out as a bit more anti-AI, but it would be fantastic if you're right that conservatives are actually more concerned about job loss. That would help keep things balanced, which I think is pretty crucial. We don't want to wind up in a global warming situation where the science and logic speaks pretty clearly, and yet very little is done because of polarization, probably fanned by pragmatic self-Interested politicians.
I don't know what to think about the rest of this. Playing politics is really distasteful, but it does sort of seem like we're going to want to make some compromises in the interest of surviving. So maybe we should be saying: "The Epstein class is building AI, saying it's not going to take our jobs and definitely won't get out of control."
I'm not going to call "polarization" a myth, but it definitely smooshes together two concepts: "we fight too much, never concede, and there's gridlock," with, "they're not doing what I want because they're too polarized."
I really don't think that second thing happens. I would invert it to "they're not doing what I want, and I'm too out of touch to know why." Usually the pattern is: the right does not trust institutions, and the left sees it as illegitimate that they do not trust institutions, so they call them polarized. They saw global warming as a scheme to extract taxes and they see science as corruptible, like any institution is corruptible.
So, the main pattern to watch out for: if the evidence chain is liberals cite the elites then yell at conservatives, it doesn't work.
These suggest you can fully get to work just with one side, being mindful that either side often unilaterally instigates polarization. Red/blue working groups would be good.
You don't think that fighting too much results in cooperating (to do things) too little?
The only way you can get to work with one side in the US is the rare occasion that house, senate, and presidency are controlled by one party, or you go straight to executive orders.
Which I guess is what I'm hoping for anyway. But I want anti-AI candidates from both parties because the issue isn't polarized.
The effect of polarization is apparent in the finding that belief in global warming fell among conservatives even as the evidence kept mounting. Having erroneous beliefs will certainly keep you from supporting actions and enemnity will make you oppose them.
I agree with you that the actual actions to take wrt global warming are highly debatable but the reality of the situation really wasn't. It was debated to death anyway.
See AI scares and changing public beliefs for a link to the falling GW beliefs data
I think you need to be more precise with the word "polarization" because it sneaks in several unrelated concepts:
For example I might claim this flow took place. Obviously, very oversimplified, but it at least illustrates how these concepts play off each other:
In this telling, I think the terrifying thing is issue organizing elevating to ideological organizing, and it's really bad both sides let contempt slip in so much.
Point (2) is contestable. Maybe conservative contempt for liberals is a first mover. But I really don't think so, because I think an object-level issue is first mover. I'm also biased in that I'm liberal and I prefer a theory that's exposing weaknesses in liberals' thinking so we can do something different than we've been doing. In other words I'm setting liberal contempt, and conservative rejection of global warming evidence, in mistake theory not conflict theory and I think that's the way to go.
Liberals seem to let elites fearmonger them. This turns into a good, hard question for me: being fearmongered is ineffective politics as it looks like we're ceding power to the fearmongerers. Normies - including liberal - see Sam and Dario as the AI safety people, who are also the elites who benefit from people caring about AI safety, and they are hyper-skeptical. As an actually fearful person, how to not follow this pattern?
(Stanning Dario is probably not the way to go, that looks more like an elite/polarized pattern. I think rationalists agree on that, and it's not really taken off anywhere else to my knowledge.)
Is any of this on topic: well, as you apply polarization to this problem, I'm just saying the concept doesn't cut reality at the joints. It leads to poor empathy for "their" point of view, it implies that issue-based / ideological / affective polarization happen at the same time instead of at different times. It also can make it too easy to be not self-critical, as it's treated more like a force of nature and less as a response to concrete things people do.
other "leaf" points
Trifectas aren't rare, they're about half the time. Without trifecta, farm bill, defense bill, debt ceiling manage to pass, plus a $2 trillion pandemic relief bill. Biden got a 2024 foreign aid (defense) bill through.
I don't know how much social change has taken place by pressing all parties equitably. Usually you press one party. One exception I know is that the 2003 Iraq invasion was pretty bipartisan, maybe not even an irrelevant example in the context of AI scares.
Status: brain-dump of Giridharadas The Persuaders, Ch. 5. This chapter is an extended interview with Anat Shenker-Osorio (ASO, my acronym), who has a public presence and published books in her own right. This is more theory; I don't have a direct AI application, but if you're wading into politics more, this is recommended background reading.
Common political conceptions
A lot of fury at Democrats, from their own side, comes to this argument:
Too many Democratic politicians play to the center, competing for moderate or swing votes who simply do not exist. Even worse, they come across as spineless and immoral by making compromises that don't make any sense. If they tack left, they can rally the base much better and win more, and they should definitely stand up to Republicans in Congress more.
I agree with 75% of this at object-level. Democrats really should move left, but it's because animating the base is more important to winning the center than they estimate, not because they should ignore the center.
A better model for base voters, swing voters, stances and messages
Swing voters are very much a thing
We don't need to distinguish swing, moderate, independent, Obama-Trump, or other labels that overlap a lot all that much.
The best way to think of a swing voter is they're the "good point!" people.
One pair of polls shows this:
Those are completely contradictory... showing that for such a voter, agreeability matters more than principle. Swing voters, basically, want politics to go away, and for the world to make a bit more sense.
Therefore, the way you win swing voters is by saying something simple, loudly, consistently, frequently. You get the megaphone to do that from the base.
Persuade swing voters by animating the base
ASO distinguishes "activists" from "the base."
When you do animate the base, they work 24/7 persuading swing voters. The base is big, loud on social media, enmeshed in swing-voter dense social networks. You want them doing much more for you than just turning out.
So this breaks the usual left v. center mental model: base+swing a synergy, not a tradeoff, and the synergy is a motivated base lifts swing voters, it doesn't turn them off.
The best message is one your opponent won't shut up about
ASO gets this from Republican pollster Frank Luntz. Republicans are very good at it. You want a message that:
Then your opponents do all your messaging work for you.
R: we're building a border wall
D: terrible wall, see, look how easily Mexicans just cut right through it
Swing: ohmygosh they really are trying to break in to our country
Democrats often find firm footing by making Republicans talk about The Economy as a deity you have to sacrifice children for.
D: Universal Pre-K, we're gonna pay for it by taxing the rich
R: You fool! the rich will just evade your tax by moving out of New York.
Swing: tax evasion is a crime tho?
Democrats have an opportunity on Israel's War. Today it goes like:
R: we're pro Israel
Some D: it's a genocide
Other D: look, I'm really sorry about the anti-semitism in our party. We also stand with Israel
Swing: hmm D are racist, R are not
It could go like this:
D: Israel has killed 20,000 Palestinian children, it's a genocide
R: OK firstly the 20,000 children number is unverified, secondly killing 20,000 children doesn't count as a genocide
Swing: I have a 6 year old...
Sell cake, not cake batter
Don't stay in the thick of intermediate, legal, procedural language. Try to avoid saying "health insurance" even, the goal is "see a doctor." Everyone can see a doctor.
Bad: We need to be mindful of balancing the need businesses have for immigrants with respect for the law, and far too many immigrants do not respect the law.
Good: build a wall. deport them. close the border. etc.
Bad: We’ll expand the Child Tax Credit and increase refundable benefits
Good: Everyone can afford groceries, gas, seeing a doctor
Bad: We’ll reform zoning and streamline permitting
Good: You shouldn't have to drive half hour just to get your kids to school
Really the only message Democrats need is we can have nice things. I don't think this is 100% true issue-by-issue, but it is mandatory as part of a general messaging spread, and I notice progressives seem to have embraced the two-pronged approach of "standing up to the rich" while "promising affordability."
What if AI safety gets polarized?
I'm a lot more worried about failing to socialize one side or the other at all to the issue. I'm also worried about misdiagnosing that as polarization: if the right gets its shit together on AI safety and the left does not, I'm not going to point fingers at the right for "polarizing" it.
I'm under the impression right now that:
To me, negative sentiment toward AI across the board is probably good, but the industry has a lot of opportunity to direct it to something useless, like banning a few data centers and doing not much else.
I also notice that HuggingFaceGate has lasted in the news way longer than OpenAI probably wanted. That seems important.
Lastly I'll point out Ronnie Cheung's phrase "your job is to destroy AI" stood out to me. It's more "cake" than "cake batter" and sounds nicer than stopping halfway, like the industry's preferred term "pace AI."
Also Ro Khanna coined the term the Epstein class. That's the level of contempt many Americans have to the CEO overlord class right now. Tech overlords is too weak.
seriously I'm just reproducing this from memory, I might have the #s flipped or polls worded wrong.