I work at the Alignment Research Center (ARC). I write a blog on stuff I’m interested in (such as math, philosophy, puzzles, statistics, and elections): https://ericneyman.wordpress.com/
Eric Neyman
Karma: 6,068
Do we know that he didn’t just change his mind?
I think voters might not trust that the politicians will in fact vote in accordance with their constituents’ beliefs. It’s hard to credibly commit to that.
Campaign update: With 56% of the vote in, AI safety champion Scott Wiener and Connie Chan (both Democrats) have advanced to the November election. Wiener got 41.3% of the vote, compared to Chan’s 28.6%. This is roughly what I expected.
Wiener is favored to win in November, but it’s not a done deal, for a few reasons:
Voters for the third-place finisher, Saikat Chakrabarti, might go disproportionately to Chan over Wiener: I think Chakrabarti is more ideologically similar to Chan overall, though it’s not totally clear.
Republicans really don’t like Wiener, and might prefer Chan over Wiener.
Nancy Pelosi, who endorsed Chan, will probably use what power she has to try to further boost Chan.
Wiener is about 70% to win per prediction markets, and that sounds about right to me. (Maybe it’s a tad low; my intuition is more like 75%.)
Getting elected involves compromising your values.
Usually, when people say “compromising values”, it carries a connotation of low integrity. That’s not really my intention here. Instead, I mean it in a more neutral way: if you’re running for office, it’s pretty likely that your values will be out of step with your constituents’ values in one way or another. Maybe you have a wider moral circle of concern, or have a stronger sense of justice, or whatever, leading to you holding views that are unpopular among your constituents.
This makes you less likely to win. And so, you have three choices:
Stick to your values, tanking the decreased likelihood of winning
Change your values, or at least commit to acting on your constituents’ values rather than your own
Deceptive alignment: Run on your constituents’ values, but once you’re in power, act on your own values
Options 1 and 2 trade off against each other, and different politicians have reputations for being at different points on the spectrum. For example, Bernie Sanders is thought of as pretty close to 1, while Gavin Newsom is thought of as pretty close to 2.
The most clean-cut example of #3 that I know of is Jimmy Carter, who deliberately tried to appeal to segregationists in his campaign for governor, but then famously declared that “the time for racial discrimination is over” in his inaugural address. I find this interesting to think about, because I think that racial discrimination is bad, but at the same time I think it’s a bad policy to be deceptively aligned to your constituents.
Anyway, I sometimes think about this in the context of electing AI safety champions to Congress. I think the value proposition is great, but the more they take route 2 over route 1, the lower the value of electing them. (Meanwhile, I think that AI safety champions should steer away from route 3, though in my opinion it’s fine to place a different emphasis when campaigning than when policymaking.)
Previously, I’d written about my support for Alex Bores and Scott Wiener, who are running for Congress. I wanted to highlight another person who’s running for office, namely Will Dreher, who’s running for the Washington State House. As far as I know (I haven’t checked thoroughly), he holds the distinction of being the first serious candidate for elected office to put AI as the top issue on his platform.
Dreher’s platform talks about a number of risks from AI, ranging from risks that are already widely discussed in politics (such as deepfakes and job displacement) to catastrophic and existential risks. Based on my conversations with him, it seems to me like he’s taking catastrophic and existential risks from AI pretty seriously and is interested in pushing for legislation that mitigates those risks.
Some quotes from his platform that I like:
“AI companies’ billionaire (soon to be trillionaire) CEOs, meanwhile, alone are deciding pressing public questions like what guardrails will apply to AI’s military applications or whether to release potentially dangerous new models that could pose catastrophic risks, like cyberattacks, loss of control, or bioengineered pandemics. Yet these companies have shown no willingness to slow down—indeed, they have pledged to fight state attempts to regulate them.”
“I will… pass legislation addressing the catastrophic risks that certain AI applications pose, including right-to-warn protections for AI employees who perceive risks from their companies’ work, regulatory audits and controls that apply before even internal deployment of new models, and required transparency by AI companies of risks flagged internally during safety testing.”
I think that Washington is one of the more important states for AI regulation, as a number of AI companies have big offices there. So I’m excited to have an AI safety champion in the Washington legislature. A number of donors who are concerned about catastrophic/existential risk and want to see greater regulation of frontier AI companies have donated to him, and I think further donations are worthwhile. You can donate up to $2.4k here if you’d like (donate via this link, rather than directly through his website).
Mechanistic estimation for expectations of random products
Thanks! I meant to make the narrower point that my probability that the race will be decided by a small number of votes has gone up. I’ve expanded on footnote 8 to clarify how I’ve updated / what has changed.
Yeah. I guess another piece of it is that it’s straightforwardly unpleasant to have attack ads being run against you, regardless of how much they affect your chances of winning.
Update on the Alex Bores campaign
Some examples (epistemic status: not very thought-through; I’m more confident that there are uses than of any specific uses):
If there’s a massive increase in cyberattacks, prediction markets could help predict the scale.
AI tools might result in a bioengineered pandemic; here, the value is similar to what it would have been for Covid. But also you might have useful markets like “Will there be a consensus that AI helped engineer this virus”, if there’s no consensus.
If we lose control of some (non-superhuman) AIs, in a way that turns out to be hard to shut down, it may be useful to predict what kinds of things those AIs will try to do.
There’s been increased discourse on whether prediction markets are net-positive for the world.
My take: so far, they haven’t been clearly net-positive. However, when I think back to Covid, I sure wish that prediction markets had been as mature of a technology as they are today; prediction markets in February 2020 on how many cases there would be in April 2020 would have probably made the world marginally more sane.
Prediction markets are probably most useful in a crisis, where decisions need to be made quickly based on uncertain information. I find it plausible that we’ll have such a crisis within the next decade, particularly in the context of AI. And I think that the benefits that prediction markets are likely to provide in such a crisis will likely outweigh the negatives incurred thus far through things like increased sports gambling.
I think that, while many LessWrong readers do believe that one party is way better than the other, such that the inter-party quality variation is far larger than the intra-party quality variation, this is not true of all readers.
And I think it’s a reasonable move to write a post that says “Assuming that these are your values/beliefs, you should do X” without taking a position on whether those values/beliefs are correct: it can be valuable and action-guiding for such people!
This consideration is meant to be included in the evaluation of Donna and Randy. As in, I am supposing that they are of similar quality after taking into account the dynamic you mention.
I don’t understand how what you’re saying is in tension with what I’m saying. My post makes no object-level claims about the relative goodness of Democrats and Republicans. I’m merely positing a hypothetical in which you think Donna and Randy would be equally good as president, despite being nominated by two different parties, one of which you prefer to the other.
Is it fair to assume that Obama-McCaine and Obama-Romney were the background thoughts that lead to this post?
Nope. I was thinking about this in the context of imagining hypothetical nominees in the 2028 presidential election (I probably won’t say who specifically I was imagining).
Suppose that you generally prefer Democrats to Republicans, but Republicans nominate Randy, who’s an above-average Republican, for president, while Democrats nominate Donna, a below-average Democrat, for president, such that you’re actually roughly neutral between them.
Even though you think Randy and Donna would be about equally good as president, I claim that you should vote for Randy. That’s because, if Randy becomes president, he’s “locked in” as their party’s nominee in the next presidential election, which is great from your perspective. You’d much rather the next presidential election be contested between Randy and a generic Democrat, than between Donna and a generic Republican.
This difference can be important enough that you might sometimes want to vote for Randy even if you actually prefer Donna as president by a small-to-medium amount.
(This of course works symmetrically if you switch the two parties in my example.)
Huh, I guess I’m not familiar with the connotations? I l’m used to seeing it used literally.
LW react suggestion: “big if true”
Ah I see. I think the analogous thing would be if Harris but not Trump could appear on PBS. Which I think would be quite bad. But maybe not so bad that it would tempt me into calling the US “not a democracy”.
Thanks for writing this, Jeff!
If anyone is interested in receiving political donation recommendations, feel free to fill out this form!