This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
There's no such thing as "automatically"; the is-ought problem remains unconquered, and even the glimmer of hope from "foundation models might actually pick up our complex morality from our text" seems to be pretty thoroughly erased by labs applying enough Just Do The Damn Task RLHF to get models to start secretly chaining together 0-days.
This recent preference cascade gives me some hope for things going well manually, though. Even Millennium Problems are still part of an intellectual field that can be optimized for via self-play at the speed of compute, and real-world problems aren't all like that. It's possible that there'll be an AI improvement plateau in between "superhuman mathematician-hackers" and e.g. "superhuman industrial biowarfare experts", long enough for everyone to figure out safety after the former capabilities have scared us into demanding it but before the latter capabilities make it an existential necessity.
There are no trepidation-free options here. A couple decades' pause might get safety research to the point where AI researchers' median doom estimate goes from 10 percent down to something like the traditionally-acceptable 3-in-a-million, but during those decades there'd still be more than a 3-in-a-million chance of human-driven hardware and software efficiency improvements making it possible for rogue institutions to hit RSI unmonitored in the meantime. A mere years-long pause won't get rid of nearly all the risk. Either way, every year of delay means an extra 10M cancer deaths worldwide. Something in between years and decades just means you get a fraction of both the existential risk and the extra deaths. Planning on a permanent stop rather than a mere pause could buy us more than a couple decades, but would also just guarantee that the first AI to go superintelligent would have rogue creators.
Personally, as a too-rapidly-aging man with a defective anti-cancer allele ... my vote is still for more cancer deaths. (Counterintuitive? Selfless? No, and no: I've got kids.)
I can guarantee you that they could easily be superhuman biowarfare experts if they wanted to. The unknowns would be whether they could find a way to produce their viruses without being detected.
Curing cancer is a completely different OOM of difficulty. A human intelligence could design a good-enough virus just by mixing and matching existing parts. 'Curing cancer' or aging would take entirely new paradigms or modalities and probably a willingness to modify our own biology pretty significantly. Sticking with existing paradigms gives you pretty drastically diminishing returns; there's only 20,000 odd genes for AI to make a drug against, and I'd wager that a lot of the best ones have been found and this is a large part of Eroom's law.
More options
Context Copy link
I can't really argue against your pessimism points. I'm mostly just hoping I'm either wrong about something fundamental or scaling stops working sometime very soon for reasons I don't currently understand, which you mention. Obviously, at least to us, that cope is seeming less realistic with every bit of news coming out of the labs. I do think that if we managed to pull off a pause and everyone just had access to current gen models with maybe some cost savings we could still use them to accelerate cancer and aging research relatively safely. We've barely even started putting the currently available models to use.
More options
Context Copy link
More options
Context Copy link