Sustaining symbiosis between a misaligned AGI and humans would seem extremely hard. If superintelligent and capable of manufacturing or manipulation, the AGI will eventually come up with better and better ways to accomplish it’s goals without humans. Temporarily avoiding catastrophic misalignment doesn’t seem sufficient to entrench symbiosis with a misaligned system in that case. I am generally pro-symbiosis as a political goal, but am not optimistic about it long-term for AGI without a lot more detail on strategy.
Also I don’t think intentionally deploying not fully aligned AGI really mitigates incentives to conceal behavior. You can know something is not aligned, but not know how much. You also can under or overestimate a system’s capabilities. There will still be instrumental power seeking incentives. It’s basically the same issues that apply with trusting humans, but more power. I don’t think it matters if there is a minor degree to which the incentives are less intense, when capabilities can be disproportionate and there aren’t counterbalancing incentives.
Overall, given the focus on the potential for catastrophic narrow alignment, we need something like broad intent alignment, which may be the same thing you are aiming for with symbiosis. I like some of the analogies to Academia and Academic freedom here, however academics are still human (and thus partially aligned) and I’m not sure people have a good grasp of what norms are and aren’t working well.
Thanks Linch!
Re 1, I think we are on the same page now. I’ll consider his Iraq war views as basically not strong evidence either way.
Re 2, I don’t think the second part is confused, but agree it is not relevant to Trump, just strategy selection for EAs.
Re 3, this is something I’d put like ~80% credence in, and I think it is more important than most points. The 20% comes from increased volatility/unpredictability.
Re 4, I believe C, and put a very low probability on B. I think it was rational for Trump, given his info state, to believe that he lost at least one state due to illegal voting. I think the vast majority of spammy claims and cases Republicans pressed were not credible at all, and I don’t know how to feel about these overall in terms of norm decay, vs. attempting to get the legal system to check a lot of potential claims quickly when you don’t yet have good evidence (I oppose anyone that was knowingly making false claims/cases). I do think their worries about mail-in ballots and vulnerability to illegal voting are justified, and that there is a lot gov could do to increase justified confidence in the elections. The states close to or below 1% margins went D with mail-in votes: its not surprising mail-in votes were more D heavy, but it’s easy to see why they thought they got cheated. I’m pretty sure things like the Electronic Registration Information Center don’t work super well given that I received mail in ballots and political calls for places I deregistered and no longer lived.
On the part about firing, see this NBER paper. It is necessary to firing people for political reasons to increase competency, and the left purges the bureaucracy more thoroughly that the right historically, granting the left cost advantages for programs they want. The problem is choosing good programs to do.
Re: Resisting expert pressure areas I think Trump made good decisions despite expert pressure:
- High confidence: Energy policy (also in Europe) + strategy for getting allies and NATO to pay more, negotiating the Abraham accords.
- Medium confidence: his version of the Afghan pullout strategy (keeping Bagram), striking Soleimani, using tariffs to renegotiate trade deals (though bad execution in some areas)
- Mixed: COVID: bad cuts (justified citing CFHS on the U.S. being the most prepared), bad to initially downplay, good on travel relative to experts at the time, good to do Operation Warp Speed and push for earlier scaling, inconsistent on masks, good on re-opening earlier and schools.
On Harris’ record: It’s fair she didn’t have much influence on CA policy and wasn’t in a good position to influence much in Congress either. The bills she’s proposed would have cost more than $20 trillion by now, but those didn’t pass and may have just been to send signals.
I agree you have no reason to take anything I say on expert interviews at face value. I think your set of views is reasonable to have given your network.
Re 5: Due to greater economic policy rationality, explicit false beliefs on climate change that are typical of many Republicans are less costly in practice than Democrat implicit false beliefs on climate trade-offs. Texas is building more clean energy capacity than basically everywhere else in the U.S. combined. Environmental reviews, lawsuits, and over regulation of nuclear power are all issues that largely come from the left and make it hard to do any construction that would reduce emissions. Because climate is a virtue signaling topic for the left, typical proposals sacrifice more value than they could hope to save due to uneconomical spending proposals and bans (e.g. on pipelines with allies, fracking, etc.) To be fair, Harris has shifted to be pro-fracking now I think, but she did propose $10 Trillion in climate spending before. We could debate the merits of the Paris agreement pull out and I agree the U.S. should be more energy efficient per capita, but fundamentally it doesn’t make sense to handicap the U.S. economy more than the Chinese economy and have allies free-ride on U.S. defense spending at the same time.
Agree that NIMBYism and car culture pose big problems and conservatives can be worse on both, though as Dems control the cities and the policies that drive cost growth in them, I think they are more to blame in the worst cases. As an example, the environmental review to even look at digging another metro tunnel under the bay was set to cost a billion dollars. In the bay most of the NIMBY arguments complain about gentrification, stopping greedy developers, and protecting the environment. For national policy, Trump’s head of HUD claimed to be anti-NIMBY and aimed to condition HUD funding on local zoning reform. That said, Walz is YIMBY too, the Biden admin does seem to be trying harder to increase housing supply now, and some of the permitting form looks potentially promising provided lots of the things they add on don’t become veto points. Overall, I do think conservatives will be more NIMBY in the suburbs, but will open more areas to development and lower crime in a manner that facilitates relatively more urban density.
In terms of reducing wokeness there’s both policy and attitudes. A Trump admin can continue repealing policies that incentivize and force people and companies to be more woke if they want to succeed or to defend rights that have little to do with discrimination. At the same time, people being mad about Trump will increase woke reactions, so that’s fair and I am not sure how things net out on polarization. Causing the far right to go nuts doesn’t sound great either when they have all the guns, but either way I don’t want to be held hostage by extremist reactions.
On the praise for dictators thing, I misinterpreted your comma. Disregard.
Re: JD’s statements around the assassination: I directionally agree, though if we consistently apply the standard that people who publicly jumps to conclusions about responsibility in response to violent events shouldn’t be in office, then I’m not sure that many presidents/VPs reach the bar.
On immigration: my understanding was that the border proposal was unacceptable because it explicitly tolerates allowing just under 5,000 people in per day via illegal border crossings rather than via border control points. If this specific claim is not true, that would substantially change my view of how bad his opposition to the border compromise is.
On AI I share the same hopes as you. I don’t want ideological capture by e/accs or EAs though, because both are too myopic. I want them counter balancing each other, and I want tech acceleration mostly focused on things other than AI and narrow/harder to abuse applications of AI. I think we need substantial growth to deal with the debt burden, and generate enough value to have more positive-sum politics and foreign policy. At the same time, I think it is hard to directly attack most of the EO as stated. One issue is largely on how the involvement of the government to assure that AI increases equity will lead to a lot of negative-sum behavior and censorship that has nothing to do with safety. Thiel sometimes articulates the more extreme version of the longer-term concern in terms of authoritarianism, but that seems further off.
Overall, I agree the biggest threats and opportunities aren’t necessarily right or left wing. I feel now like I have a few points on foreign policy and immigration policy that could cause me to make large updates if I find more decisive counter-evidence to my current position. I think it may take me longer to sort through cruxes/points of info that would make me decisively more fearful of dictatorship risk.