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 -
It is very strange to me that so many people seem to be swallowing this existential risk narrative when there is so little support for it. When you compare the past arguments about AI safety to the current reality, it's clear that no one knew what they were talking about.
For example, after all the thought experiments about "unboxing", OpenAI (which I remind you has constantly been making noise about 'safety' and 'alignment') is now immediately rushing to wire its effectively unaligned AI deeply into every corporate process. It's an unboxing party over here. Meanwhile the people actually in charge seem to have interpreted "alignment" and "safety" to mean that the AI shouldn't say any naughty words. Is that helping? Did anyone predict this? Did that AI safety research actually help with anything so far? At all?
The best argument I'm seeing is something like "we don't understand what we're doing so we can't know that it won't kill us". I find this pascal's mugging unconvincing. Especially when it's used so transparently to cater to powerful interests, who just want everyone else to slow down for fairly obvious reasons.
And even if I did take the mugging seriously, I don't know why I should believe that AI ethics committees will lower the risk of bad outcomes. Does overfitting small parts of an LLM to the string "As an AI language model" actually make it safer? Really? If this thing is a shoggoth, this is the most comical attempt to contain it that I could imagine. The whole thing is ridiculous, and I can just as easily imagine these safety measures increasing AI risk rather than lowering it. We're fiddling with something we don't understand.
I don't think anyone can predict where this is going, but my suspicion is this is going to be, at most, something like the invention of the printing press. A higher-order press, so to speak, that replicates whole classes of IP rather than particular instances. This tracks pretty well with what's actually happening, namely:
Powerful people freaking out because the invention might threaten their position.
Struggles over who has control over the presses.
Church officials trying to design the presses so they can't be used to print heresy.
I don't trust any of these people. I'd rather just see what happens, and take the ~epsilon chance of human extinction, rather than sleepwalk into some horrible despotism. If there's one thing to worry about, it's the massive surveillance and consent-manufacturing apparatus, and they (bigtech and the government) the ones pushing for exclusive control in the name of "safety". Might as well argue that the fox should have the only key to the henhouse. No thanks.
I'm pretty sure he'd say something like 'even with a p(doom) of .1%-5%, if I can move that a little that's still billions of people in EV just over the next few thousand years', which I'd phrase as "It's almost impossible to really understand how the incredibly complex interactions between humans and AI will play out, and the stakes are impossibly high, so it's worth trying".
More options
Context Copy link
You can cut all the power of a datacenter with a jaws of life ...
I sincerely ask that you Google why that won't work. It's the most obvious, and thus most naive approach to solving the problem.
If you can think of it, it can think of it first, and take steps to mitigate that, such as migrating its data or spinning up redundant copies.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Pascal's Mugging is about a probability which is acknowledged to be negligible, which the mugger asserts should still be sufficient for you to pay up given expected value. I don't see how you're getting that out of any actual proponent of AI safety.
Are you just conflating "we don't understand what we're doing so we can't know that it won't kill us" to "we can't know that it won't kill us", analogous to "we can't know God doesn't exist"? Because the argument from Yudkowsky et al is more like "the default outcome is 'it kills us', as the default outcome for bridge-building is 'it falls down', and we have no idea what we're doing, ergo we're fucked".
As @Harlequin5942 points out below, this is what Scott Alexander calls the Safe Uncertainty Fallacy.
Now - did Scott just make this up yesterday to argue with Tyler Cowen? Perhaps….
Personally I agree with you, not because it’s a pascals mugging but because my intuition is that risks from AI are a safer bet than risks from our current path.
Scott implicitly admits to having named it for that piece (in the caption of the first image).
More options
Context Copy link
More options
Context Copy link
Additionally, human extinction is (presumably) towards the tail end of a spectrum of possible negative outcomes of unaligned AGI. Outcomes where AGIs cause massive economic/envirionment/biological/social damage are also possible. They also could be more likely: "Why don't we just turn it off?" is not a sensible question if the harmful AGI is the equivalent of crack (or the internet).
It's on said spectrum, and it's one of the more negative possibilities (the worst is S-risk i.e. "AI tortures everyone for a very long time"). I see no reason to think that implies it's low-probability (which is implicit in calling something a tail). Something can just be terrible, it doesn't have to be moderate most of the time.
The problem with your "everything is mildly shit and we can't co-ordinate to turn the AI off" scenario is that if the AI doesn't care about humans existing then this isn't an endpoint - it's an influence war that sooner or later ends with either the AI being turned off or the AI gaining sufficient control to murder us all.
Sorry, I should have specified "tail end of the utility spectrum." As you suggest, a very negative utility event can be very likely. In the case of AI, I don't think that precise probability estimates (or even confidence about the sample space) is sensible, but my point was that there are a lot of bad things that can happen short of human extinction.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link