site banner
naraburns  nihil supernum  1yr ago (text post)   37969 thread views

Culture War Roundup for the week of March 27, 2023

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.

11
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Was a bit surprised to see this hadn't been posted yet, but yesterday Yudkowsky wrote an op-ed in TIME magazine where he describes the kind of regime that he believes would be necessary to throttle AI progress:

https://archive.is/A1u57

Some choice excerpts:

Many researchers working on these systems think that we’re plunging toward a catastrophe, with more of them daring to say it in private than in public; but they think that they can’t unilaterally stop the forward plunge, that others will go on even if they personally quit their jobs. And so they all think they might as well keep going. This is a stupid state of affairs, and an undignified way for Earth to die, and the rest of humanity ought to step in at this point and help the industry solve its collective action problem.

The moratorium on new large training runs needs to be indefinite and worldwide. There can be no exceptions, including for governments or militaries. If the policy starts with the U.S., then China needs to see that the U.S. is not seeking an advantage but rather trying to prevent a horrifically dangerous technology which can have no true owner and which will kill everyone in the U.S. and in China and on Earth. If I had infinite freedom to write laws, I might carve out a single exception for AIs being trained solely to solve problems in biology and biotechnology, not trained on text from the internet, and not to the level where they start talking or planning; but if that was remotely complicating the issue I would immediately jettison that proposal and say to just shut it all down.

Shut down all the large GPU clusters (the large computer farms where the most powerful AIs are refined). Shut down all the large training runs. Put a ceiling on how much computing power anyone is allowed to use in training an AI system, and move it downward over the coming years to compensate for more efficient training algorithms. No exceptions for anyone, including governments and militaries. Make immediate multinational agreements to prevent the prohibited activities from moving elsewhere. Track all GPUs sold. If intelligence says that a country outside the agreement is building a GPU cluster, be less scared of a shooting conflict between nations than of the moratorium being violated; be willing to destroy a rogue datacenter by airstrike.

if its presence in the CW thread needs justifying, well, it's published in a major magazine and the kinds of policy proposals set forth would certainly ignite heated political debate were they ever to be seriously considered.

"Yudkowsky airstrike threshold" has already become a minor meme on rat and AI twitter.

Yudkowsky knew this was coming for 15 years... and yet he founded an NGO.

I don't know if he was just completely delusional or he was once full of it. But I remember in 2012 thinking "Well obviously this guy doesn't believe it if he's founding an NGO and taking a 50k salary... if he believed it he'd be smuggling fentanyl, building up a warchest and recruiting hitmen, assassins and death squads."

Ie. If he actually believed it he be ruthlessly grabbing as much power as he could without regards for life, decency, or his immortal soul... Hell this is what his author insert characters actually did in HPMOR, the children studied war, and the adults waged it... they didn't sit around raising funds or publishing white papers.

I mean now it looks like he's closer to believing it... but i still don't think he stack the bodies himself.

If I had infinite freedom to write laws

Well, at least we're safe from that. There are a myriad ways it could end up horribly for all of us, but at least it wouldn't be ending horribly for all of us this particular way.

Also, I am not sure what this is - some weird kind of LARP fiction or he actually thinks this is something that can be taken as any kind of practical advice? If it's the latter, I have very hard time accepting the fact that a functioning adult familiar with world affairs and human history can believe this is any kind of practical advice.

I've lost pretty much all respect for Yudkowsky over the years as he's progressed from writing some fun power-fantasy-for-rationalists fiction to being basically a cult leader. People seem to credit him for inventing rationality and AI safety, and to both of those I can only say "huh?". He has arguably named a few known fallacies better than people who came before him, which isn't nothing, but it's sure not "inventing rationality". And in his execrable April Fool's post he actually, truly, seriously claimed to have come up with the idea for AI safety all on his own with no inputs, as if it wasn't a well-trodden sci-fi trope dating from before he was born! Good lord.

I'm embarrassed to admit, at this point, that I donated a reasonable amount of money to MIRI in the past. Why do we spend so much of our time giving resources and attention to a "rationalist" who doesn't even practice rationalism's most basic virtues - intellectual humility and making testable predictions? And now he's threatening to be a spokesman for the AI safety crowd in the mainstream press! If that happens, there's pretty much no upside. Normies may not understand instrumental goals, orthogonality, or mesaoptimizers, but they sure do know how to ignore the frothy-mouthed madman yelling about the world ending from the street corner.

I'm perfectly willing to listen to an argument that AI safety is an important field that we are not treating seriously enough. I'm willing to listen to the argument of the people who signed the recent AI-pause letter, though I don't agree with them. But EY is at best just wasting our time with delusionally over-confident claims. I really hope rationality can outgrow (and start ignoring) him. (...am I being part of the problem by spending three paragraphs talking about him? Sigh.)

People seem to credit him for inventing rationality and AI safety, and to both of those I can only say "huh?".

This seems pure pedantry. Obviously, the concept of rationality and ideas around optimizing for truth or utility have existed for a very long time. And there is plenty of science fiction which featured dangerous AI.

But I really can't imagine anyone even vaguely familiar with Yudkowsky or AI alignment or the rat-sphere would struggle with the claim that Yudkowsky "invented rationality and AI safety."

Capital-R Rationality is the modern groups that formed around the Bay-area subculture which Yudkowsky largely founded. It is MIRI, CFAR, Effective Altruism. It's Yudkowsky, Scott Alexander, Julia Galef. It's LessWrong, SSC, and even this very forum.

And AI Safety? Yes, the concept existed. But the entire modern edifice of AI alignment all arose from Yudkowsky's initial writings on LessWrong. He was responsible for drawing attention to it, he lead MIRI for a while, he set down a great deal of theory.

I'm really struggling to believe someone smart enough to post on TheMotte would so easily be confused by this.

BTW, if you want to read a good example of pre-Yudkowsky rationality, I recommend The Demon-Haunted World. Carl Sagan did a lot to help me learn how to think clearly, in my formative years.

I want to be clear that this is coming from somebody who once liked his writings. I didn't worship him. I didn't learn much from him. But he has always had a fun and unique writing style.

But believe me, there's no confusion here. Capital-R Rationality may be something that crystalized around LessWrong and the Sequences, but the concepts of rationality are hardly new; we're building on a legacy of humans struggling to explain the Universe that has been built over thousands of years. Yudkowsky wrote some entertaining essays, some of which are insightful (and some of which are silly, particularly when he veers into fields of science he doesn't know well). You could credit him with collecting and indexing a few good ideas. But he's very bad at practicing what he preaches - Scott, for instance, is far better at actually making and testing predictions than Yudkowsky. I suppose cult leaders don't usually lower themselves to the level of scrubbing the temple floor.

As for AI Safety, no. No, no, no. There's absolutely no defense for his egotistical claim in the April Fool's post. Futurists have been discussing AI safety since at least Asimov's Three Laws. What do you think AI researchers did before him, shrug and go "hmm, I wonder if making this neural net behave is something I should study sometime"? Maybe I can trace one particular flavour of the "edifice" to his writings - superintelligence-goes-FOOM-breaks-out-of-black-box-and-builds-nanotech-in-a-bio-lab - but AI safety as a whole would still exist and look pretty much the same without him. Arguably, it would be healthier, with the many people with different intelligent perspectives not being drowned out by his singular view and stubborn insistence that he knows the unknowable future.

Thank you for clarifying your points, but I think this ultimately falls into a disagreement about what the invention of a concept is. It seems you don't really disagree that Yudkowsky invented Rationality as a subculture, you just don't find it particularly impressive - which is fine, but I doubt most readers are confused reading about the difference between the subculture and the idea of rationality.

On AI Safety, I am very skeptical of your claim that current discussion around AI safety would look the same without Yudkowsky. I'm sure in a counterfactual world that someone like Nick Bostrom or whomever would still come up with many of the ideas, but this is true for most things. Yudkowsky definitely had an outsized influence, and I think if the next AI researcher survey put out a question about major influences, he would definitely perform very well.

People seem to credit him for inventing rationality and AI safety

The latter sounds particularly ridiculous considering Terminator 2 was a huge hit in 1991 when Yudkowsky was barely a teenager and AI safety has been a staple of scifi literature since the 60s.

Longer than that -- R.U.R., the 1920s story that gave us the word "robot", featured a robot revolt.

I see we're back to trying to outlaw mathematics. I encourage everyone to read this article by Stephen Wolfram describing how LLMs work before panicking. I cannot understand the degree to which LLMs have apparently broken some people's brains.

I'm not sure why you find that article reassuring. Wait until you hear about the shitty hardware that human brains run on, only 30 Watts! Yud isn't even saying that the current LLMs are all that dangerous, he's saying that we're pouring 10B/y and all the top talent into overcoming any limitations to making them as smart or smarter than humans. What would make you scared?

I do not think the takeaway from the article is about the hardware that LLMs are being run on. It's about the way LLM's function. The LLM doesn't understand the content of the query or its response the way you or I do. It just understands them as probabilistic sequences of tokens and its job is to predict the tokens that should come next. An interaction I recount in another comment showcases this issue. I point to the article because it is not clear to me that what LLMs do (token prediction) is the kind of thing that can be extrapolated to the dangers people like Yudkowksy are worried about with respect to unfriendly AI.

What would make you scared?

If we had an AI that actually understood the meaning of what it was being asked.

The LLM doesn't understand the content of the query or its response the way you or I do. It just understands them as probabilistic sequences of tokens and its job is to predict the tokens that should come next.

This to me seems like a pretty shallow explanation of understanding and the same criticism can be applied also to humans. According to some people like Scott Alexander, human brain is "just" a multi-layer prediction machine. It seems that the feeling of understanding itself is nothing extra special, some people on drugs like LSD feel as if they cracked the code and now understand the whole universe and their place in it. In practice understanding can be viewed as ability to give correct output given an input. We do not have access to many other methods, that is why we use tests to see if students understand things they learned.

Additionally I do not think that saying that the LLM doesn't understand the content of the query or its response the way you or I do. is that much reassuring. Quite to the contrary - LLMs give correct answer to very large set of problems and yet obviously they came to that place using completely different approach compared to humans. This makes them more alien, more inscrutable and thus more dangerous in my eyes.

If we had an AI that actually understood the meaning of what it was being asked

How would we actually detect this, though? I don't think we know how to detect qualia or consciousness yet.

I confess I do not have a great idea of how to answer this. I'm not exactly sure qualia or consciousness are requirements for understanding either.

Fair enough. "Understand" is a pretty underdefined term in this context, and I was thinking of our internal feeling and experience of knowing that, say, "chair" refers to a certain concept, but that's not the only thing "understand" can refer to. Certainly an AI could be said to "understand" the meaning of something if it behaves in a way that is equivalent to someone who understands the meaning, even if it doesn't experience qualia or have consciousness, and thus it has no internal experience of its own. For a chatbot like ChatGPT, that would be producing text that looks like it was typed by someone who understood the meaning of the prompt. It's not clear to me how we can tell that some text produced by ChatGPT crosses some threshold where we can say that it shows some level of "understanding" the meaning. It certainly makes errors often where it clearly doesn't understand the prompts, but humans often don't understand the meaning of what is asked of them either, and so a perfect record would be too high a bar. I was just curious where you would place the bar.

I think you're being misled by a very specific failure mode of LLMs trained on tokenized input. The spelling and number of words is explicitly scrubbed from their input. Asking for word counts is like asking a blind person who reads and writes Braille about the shape of letters.

Good enough next-token prediction is, in principle, powerful enough to do anything you could ask someone to do using only a computer. I'm not claiming that this is a plausible route to super-powerful AI. But the "just" in "It just understands them as probabilistic sequences of tokens" seems totally unwarranted to me.

To be clear, I think LLMs can do a lot of really impressive things. I've used Github Copilot in my job and it was able to autocomplete some mostly correct code (variable/property names needed fixing) just from my writing a comment. It was pretty cool! But the leap from Copilot or GPT-4 or whatever to "We need international regulation on GPU production and monitoring for GPUs and to air strike countries that look like they have too many GPUs" is absurd.

Good enough next-token prediction is, in principle, powerful enough to do anything you could ask someone to do using only a computer.

I guess with the caveats "good enough" and "in principle" I am not sure I disagree but I am also not sure any LLM will be "good enough."

I still don't understand why you think the capabilities of current LLMs are an important factor in how scared we should be about AGI in the medium term. I also don't understand what threshold of capabilities you want to use where we could wait until we see it to coordinate a slowdown. The better these things get, the more demand there will be for their further development.

What would make you scared?

"If I had infinite freedom to write laws" <---- this would

Fortunately, it's not likely. As for AI, I don't think there's anything that made me scared yet. Maybe there will be, but I'd have to see it first.

I agree that totalitarianism is really, really scary and plausible.

But I'm saying that wrt AI, if you wait until you see something really scary, it'll probably be too late.

I don't have anything to rely on but my previous experience. My and humankind's previous experience tell me: a) totalitarians killed literally millions of people, all in service of the goals that they considered noble and existentially important and b) so far, every time we were told something would surely kill everybody it didn't, and not for the lack of such predictions. Of course, it could very well happen that this time it's different and it will kill everybody. But I see no way to establish that, and the bar for resorting to measures that are already known to kill millions is very, very high, and merely proclaiming "yes, those guys all were wrong and false prohpets, but it's different with me, I am the real prophet" - it's just not enough. I understand that it could be very frustrating for a real prophet, but it doesn't mean I would abandon my reason.

This sounds totally reasonable. We certainly could be in a world where there are threats that are too hard to distinguish from fake ones to respond to them without screwing ourselves in other ways.

I guess I won't try to further convince you here, other than to say "every time we were told something would surely kill everybody it didn't" is certainly a valid reason to discount doomsday prophets in general, but not a good reason to dismiss the possibility of doomsday.

Any concept which can be expressed in words is now, in principle, understandable by a computer. I am confused by your confusion.

Maybe you and I have different definitions of "understand" but this seems clearly false to me.

ETA:

Perhaps better to say, I don't see what this statement has to do with LLMs.

Let me break it down:

  1. LLMs understand words and sentences (or at least the relationships between them).

  2. If a concept can be expressed in words, then what that concept means is encoded by how the words that express that concept relate to the words that express any other concept.

  3. Therefore, as an LLMs capability to relate words and sentences to each other approaches or exceeds human level, the LLMs understanding of concepts approaches or exceeds human level.

This is a very high-level explanation, but I think it captures the core intuition.

I must ask, what is the most powerful LLM you have personally interacted with? Default ChatGPT feels like talking to a 6-year-old after using Bing and GPT-4.

I guess (1) is where I disagree. I think there's a substantial difference between understanding the meaning of a word and understanding that word's statistical relation to other words. LLM's understand the latter, but not the former.

I must ask, what is the most powerful LLM you have personally interacted with? Default ChatGPT feels like talking to a 6-year-old after using Bing and GPT-4.

I've used Bing and it's still pretty easy to ask it questions that demonstrate it doesn't understand the semantic content of what it is being asked. Here's an interaction I had just now:

M: Write a fifteen word story

B: Sure! Here’s a fifteen-word story for you: “The last man on Earth sat alone in a room. There was a knock.”

M: Write a fifteen word story

B: Sure! Here’s another fifteen-word story for you: “The baby’s cry was the only thing that could stop the apocalypse.”

M: Write a fifteen word story

B: Sure! Here’s another fifteen-word story for you: “The last thing I saw was my alarm clock flashing 12:07 before she pushed her long rotting nails through my chest.”

I asked it the same question three times and never once did it give me a story that actually had fifteen words in it. It clearly does not understand what "fifteen" means and how the presence of the word in the my sentence should constrain its answer.

Hmmm, you're right. GPT-4 does in fact have trouble writing 15-word stories.

Me: Write a story with exactly 15 words

GPT-4: Moonlit whispers, midnight rendezvous. Time halted, love bloomed. Hands entwined, destiny united. Eternal adventure commenced together.

Me: How many words does this story have?

GPT-4: This story has exactly 15 words.

Me: Count them

GPT-4: Here is the story with each word numbered:

1.Moonlit

2.whispers,

3.midnight

4.rendezvous.

5.Time

6.halted,

7.love

8.bloomed.

9.Hands

10.entwined,

11.destiny

12.united.

13.Eternal

14.adventure

15.commenced

The story indeed has 15 words.

Me: You forgot to number "together"

GPT-4: My apologies, I indeed missed the word "together." Here is the corrected numbering:

1.Moonlit

2.whispers,

3.midnight

4.rendezvous.

5.Time

6.halted,

7.love

8.bloomed.

9.Hands

10.entwined,

11.destiny

12.united.

13.Eternal

14.adventure

15.commenced

16.together

The story actually has 16 words. Thank you for pointing that out.

Me: Fix it to make it 15 words

GPT-4: Here's the revised story with exactly 15 words:

Moonlit whispers, midnight rendezvous. Time halted, love bloomed. Hands entwined, destiny united. Eternal adventure commenced.

On the other hand, I was able to get it to write a 68 word story on the first try.

Me: Write a story with exactly 68 words

GPT-4: Beneath the murky sky, a lonely robot wandered the abandoned city. For years, it searched for signs of life, collecting trinkets from the ruins. One day, the robot stumbled upon a withered book, its pages filled with tales of love and laughter. Entranced, it began to read. As the robot savored each word, it understood the beauty of human emotion and yearned to be more than just metal.

Trying this experiment with multiple 2-digit numbers X, it looks like GPT-4 outputs a story with X ± 2 words. I have updated slightly against GPT-4's capabilities (I expected it to get the exact answer most of the time), but I still hold to my original thesis. This is a quantitative error, not a qualitative error. The presence of the number in the input sentence does in fact constrain the output. Asking for a story with 68 words gives an output with more words than asking for a story with 49 words, but less words than asking for a story with 96 words. The model does have some concept of what these numbers are.

All that is trivially shown false by how LLMs persistently fail at the most basic mathematical problems as soon as solving those would require understanding instead of just stringing words together. They are very efficient bullshit generators but to claim they "understand" anything is a massive exaggeration.

I think Yud is messing up badly and making AI concerned people look crazy. I"m more pro-rationalist than most people I talk to and there's so many leaps in his theory that it's hard to get normies to make. His own little ingroup venerates him and he's done a good job of persuading them but once this becomes a real technology where money and improvements to people's lives are at stake he's gonna get shredded in the public arena by people actually good at this.

As a semi humorous aside the obvious correct tactic for AI doomers who don't think we're on the brink of FOOM is to try to do an AI false flag attack, get GPT-6 to blow up a building or something and try to launch a Global War on Terror style response.

On the contrary. Yudkowsky got me thinking about the UN coming to search for Weapons of Mass Instruction. I don’t think we have anything resembling the political will for bringing force to an AI inspection, not when it can be pattern-matched to that mobilization.

AI doomers are smart enough to know the predictably horrible outcome that awaits them if they piss off both the AI and the people building the AI too much.

I was writing this whole screed comparing restrictions on AI and firearms, where regulators by definition have to hit law-abiding agents harder than criminals. But I guess that’s the thrust of his argument—we need more power to deter major players, and we don’t have the precommitment to deter minor ones. If they even can be deterred.

Regardless, I’m sort of coming around to the idea that we should stop digging? If progress really is coming from megacorporations, and not from rival governments, hobbyists, or niche nonprofits, then maybe it’s a legitimate brake rather than a cope. That’d probably just push efforts into quiet national labs without further Yudkowskian air superiority. We’re not in any position to send WMD inspections, nor to take drastic action when stonewalled.

I've been over the doom porn esque feel of AI talk for years now it seems and it just keeps cranking higher & higher. We're basically on track with doom porn global warming catastrophe. It's just a race to the nonexistent finish line to see who can be doomier and gloomier on what subject.

Personally I'm ready to be told AI will cure blindness, and cancer, and give us 2 hour work days, and on and on.

I can't take any of these people even remotely seriously. AI will be fine. Our lives will be better. Just like having a fridge, or shoes. But just like food poisoning or dress shoes inspired plantar fasciitis - some shit will happen. Shut up and stop trying to retard the future (not to anyone here - just a general sort of want from me).

What do you think society looks like, in a thousand years, if "strong AI" is developed within a hundred? Do humans still control the 'general direction' of society? Their own personal fates? What role does AI play / not play?

Wasn't it covered in detail by Iain M Banks?

Those are really good and important questions and not the rapid takeoff nano-bot apocalypse Yudkowksi and AI doomers want to foreground.

There are lots of horrendously bad non-nanobot outcomes, but AI notdestroyallvalueintheuniverseism has even less ring than AI notkilleveryoneism.

If the concern is that AI will drive us to (near) extinction like humans have whales, I would observe that culling humans seems to lack the economic incentives of whaling -- although it's unclear from the whales' perspective what those incentives are, I suppose. Matrix-like human farming arrangements seem like fiction tropes, but not practical future realities.

Compared to our ancestors a few centuries ago, we consider ourselves wiser in many ways, and we appreciate the conserving wildlife is worth dollars we might spend elsewhere in many cases (although admittedly our actions fare quite poorly). I see the particular case of AGI capable of consistently outsmarting and overpowering humans but also incapable of having a rational discussion about why human culture is worth preserving over paperclips seems unlikely. Worth considering, certainly, but the singular focus on it seems driven by a God-less eschatology in which humans are presumed to be not worth saving.

I'm not sure what that "rational discussion about why human culture is worth preserving" would look like if one agent didn't already values those things more than almost any competing concern. How much of your time do you dedicate to preserving, idk, virus culture? Random rock pile culture? Could ghonorrea or tuberculosis convince you that its presence (and use of precious resources such as human bodies) was a net positive, if it were sufficiently eloquent?

How much of your time do you dedicate to preserving, idk, virus culture? Random rock pile culture? Could ghonorrea or tuberculosis convince you that its presence (and use of precious resources such as human bodies) was a net positive, if it were sufficiently eloquent?

Probably? We still keep around Smallpox samples in labs purely for the potential research its genetic diversity brings. People are surprisingly passionate about random rock piles, doubly so for interesting cases [1] [2] [3] [4]. There are a few Star Trek episodes that ponder the idea of sentient parasites, for which I'd admit the answers are less clear.

Okay, but smallpox is a good example - we keep it in an entirely disempowered state, as we do for almost all wildlife. But again, if we didn't already think something was worth keeping around in the state they want to be kept in (e.g. serial killers), what could they possibly say to change our minds?

Sure, 'total human death' isn't guaranteed. but ... what will happen to humans? Will we fully separate? Will we be left alone ... given our own portion of the earth, fenced off, to develop and grow and fight wars in? Can we go outside? Will we have to run out from under the feet of the AI when it decides to plant down a gigafactory wherever it decides? But, separation doesn't seem likely, given how AI is being shoved into every piece of human civilization it can be, even as primitive as it is today ... but if it's integrated, then what? Will it remain subservient to explicit human decisionmaking despite being gigantically more capable than humans? Is that even meaningful? Maybe humans are "aligned to dogs", but will we all be domesticated as dogs are? Why won't the reins of control over corporations, human lives, politics be handed over to it, to the immense benefit of whoever does so? And what will it do with those reins? Why would it give humans the unprecedented and strange level of 'personal freedom / control over course of life' we have today (and is that even worth much)? What prevents it, as it stumbles through its own development of technology and politics and 'action' generally, from 'changing its mind' about what the course of human civilization should look like - just as we have, looking at how much human 'values' and political and economic organization has changed over the past few centuries?

How is the AI going to do any of that. There’s so much leaps of logic in AI that it’s impossible to even argue against.

Sure, AI being capable of all that should be argued for first. But plenty of people believe "AI will become as smart as, or smarter than, humans eventually", but don't take that all the way to "... and will determine the future of human civilization on this planet" as a result.

GPT-4 sure seems like it's capable of a lot of things! What, exactly, separates 'GPT-4 but scaled up 1000x' from what people can do? Even outside that, what, in principle, prevents a computer from being intelligent? And we're sure trying very hard to create intelligent machines. Even if you'd claim something not-mechanistic, like a spirit or something is important in human experience or intelligence - clearly the 'matter', the incredibly complex biology of the brain, is plays a big role in human intelligence, so why can't machines with that be intelligent too? And then technology has revolutionized every other facet of human life, and the smartest people are putting their effort into creating intelligence on a computer ... even if the current 'make transformers scale' thing fails somehow, they'll be hard at work for something that does.

I’m not saying there won’t be AGI sometime. Or very smart AI. But then it’s the taking over the world in an unexplained way that’s not explained.

The AI is rounding up humans and turning us into batteries within a few years. What’s never explained is how.

If you accept the possibility of the existence of a hyper-intelligent AI then there is no utility in trying to guess how such an AI might take over the world. By mere virtue of the AI being hyper-intelligent and you being, well not, the AI is sure to out think you.

Also I disagree that such explanations don't exist in the first place. Off the top of my head I think of nanobot swarm, but I could think of ten more in an hour.

More comments

It’s not about humans being “worth saving”. It’s about humans being a threat to the AI (because they can theoretically build another AI that competes with it), and the AI having an instrumental goal to remove that threat.

If the AI is so smart couldn't it easily disempower humanity without even going through the effort of killing us all?

That might end up being the same thing. It doesn't need to actively kill humans, deciding that it needs some finite resource like fossil fuels more than humans do would kill billions without any intervention more aggressive than blocking access to production sites. It seems plausible that an all-powerful AI might not mind people living in pre-modern farming communities, but that would also mean huge reductions in QoL and the number of humans around.

To defend Yudkowsky a bit, if his solution were predicated on international treaty when there are high economic and geopolitical incentives to defect, you would absolutely have to enforce it with an extreme amount violence or threat of violence. Airstrikes would probably not be good enough to stop nations from trying- although I look forward to the next Top Gun movie which will feature an airstrike on a data center instead of a nuclear research facility. Maverick will come out of retirement to deliver the killing blow to the grey goo after beating the AI-controlled bogeys in a dogfight. This will do much more to garner public support for Yudkowsky's proposal than his blog.

But there's too much pearl-clutching over the "advocating for violence" criticism. AI is going to be involved in violence, in many ways. Responses like:

There’s no such thing as lawful or legitimate aggression in int’l relations, because there’s no sovereign monopoly on its use which is the definition of legitimated force. It’s anarchy and it’s all illegitimate where aggressive.

Are just completely tedious. People are going to die because of AI, there's no question. It's only a question of when, how and why.

Are just completely tedious. People are going to die because of AI, there's no question. It's only a question of when, how and why.

Already happened.

https://thebulletin.org/2021/05/was-a-flying-killer-robot-used-in-libya-quite-possibly/

Last year in Libya, a Turkish-made autonomous weapon—the STM Kargu-2 drone—may have “hunted down and remotely engaged” retreating soldiers loyal to the Libyan General Khalifa Haftar, according to a recent report by the UN Panel of Experts on Libya. Over the course of the year, the UN-recognized Government of National Accord pushed the general’s forces back from the capital Tripoli, signaling that it had gained the upper hand in the Libyan conflict, but the Kargu-2 signifies something perhaps even more globally significant: a new chapter in autonomous weapons, one in which they are used to fight and kill human beings based on artificial intelligence.

The Kargu is a “loitering” drone that can use machine learning-based object classification to select and engage targets, with swarming capabilities in development to allow 20 drones to work together. The UN report calls the Kargu-2 a lethal autonomous weapon. It’s maker, STM, touts the weapon’s “anti-personnel” capabilities in a grim video showing a Kargu model in a steep dive toward a target in the middle of a group of manikins. (If anyone was killed in an autonomous attack, it would likely represent an historic first known case of artificial intelligence-based autonomous weapons being used to kill. The UN report heavily implies they were, noting that lethal autonomous weapons systems contributed to significant casualties of the manned Pantsir S-1 surface-to-air missile system, but is not explicit on the matter.)

BTW, Lybia was also place where air power was first used in war, back in 1911.

https://en.wikipedia.org/wiki/Italian_invasion_of_Libya

On 15 October 1911, nine Italian "Specialist Battalion" aeroplanes and 11 pilots landed in Libya. On 24 October, the Italian pilot Captain Riccardo Moizo carried out a reconnaissance flight in Tripolitania, reportedly the first ever strategic reconnaissance by aeroplane.[7] On 1 November, another Italian pilot, Giulio Gavotti, dropped four 1.5 kg bombs on Ain Zara, pulling the pins with his teeth. This was the first aerial bombing in history.[8]

That’s a piece of software not AGI. Not proper AI either.

Shit, how’d we let Turkey get there first?

It's not possible without a world state with global hegemony. If China claims their fancy new datacenter is REALLY just for the tracking of political dissidents and their social credit scores and is totally not running an AI, what is an anti-AI US going to do about it?

He explicitly notes China would have to agree. That's not plausible in the near term (nor is the US doing it), and although it's difficult to predict how significant future AI developments would change China's willingness, it's probably not long-term either

If the policy starts with the U.S., then China needs to see that the U.S. is not seeking an advantage but rather trying to prevent a horrifically dangerous technology which can have no true owner and which will kill everyone in the U.S. and in China and on Earth

China

Since I probably won't get a better chance to bring this up: there's something I don't understand about the China fearmongering.

The basic argument for why AI is a good thing is that it will lead to the singularity. Post-scarcity future, incredible scientific development, transhumanism and the realization of human potential, and so on and so forth. Implicit in this argument seems to be the premise that, if we get a good outcome with a benevolent aligned AI, then the singularity will be a good thing for all humans. I've never seen it suggested that any individual or group in particular has anything to fear. It's implied that the benefits of the singularity will be relatively evenly distributed.

If that's the case, and AI is going to lead us to a utopia... why does it matter if China gets there first? The benevolent robot god is still going to lead humanity to the land of milk and honey either way, regardless of whether it happens to be built in the US or China, so why does it matter if China eclipses the US in AI development?

Or is the singularity not actually going to be evenly distributed? If China builds ASI first, are they just going to genocide the rest of the world outside of China, and then have fun until the heat death of the universe with the robot god? Why restrict ourselves to just analyzing things at the country level? Shouldn't we be equally worried about Sam Altman genociding everyone he doesn't like if OpenAI are the first ones to get to ASI? I don't know what the sequence of events is supposed to look like here.

The constant refrains of "we can't fall behind China" make AI sound a lot closer to a conventional weapon of war, rather than the pure unadulterated good that AI advocates want to present it as, which is all the more reason that it should be tightly regulated, in a proper international framework.

China implementing the land of milk and honey is fine. China implementing the Torment Nexus is bad. Both of these are true for any country.

A selection effect for the less careful or less altruistic raises the chance of getting a bad outcome.

If we get one of the intermediate outcomes, a powerful but narrow Tool AI with no take off, I want it to be ours and not theirs. Especially if its narrow expertise is military.

I also find the "WHAT IF CHINA/RUSSIA/ETC. GET THERE FIRST?" arguments extremely silly.

With a the exception of a tiny number of particularly unhinged sadistic psychopaths (the number is probably roughly epsilon), the vast vast majority of people are going to press the "create paradise" button not something so parochial as the "make my country hegemonic forever" or the "kill all the ethnic group I don't like" button. Even people who are, right now, merciless hardnosed Machiavellians would press the paradise button since they could do so without any trade-off for themselves or their in-group.

No, I'm pretty sure that if Putin didn't make Russia into anything more than a shithole for his cronies to feed on during a whole generation of his uninterrupted power, he also won't create a paradise with an AGI button. This is a genuine value difference. Some people don't value goodness, except for themselves and their loved ones.

And I think most machiavellians of Putin's caliber are not much better.

particularly unhinged sadistic psychopaths (the number is probably roughly epsilon)

«About 1.2% of U.S. adult men and 0.3% to 0.7% of U.S. adult women are considered to have clinically significant levels of psychopathic traits».

he also won't create a paradise with an AGI button.

You sure? It's not like he'd have anything to lose. Creating a paradise for everyone wouldn't detract from the slice of paradise available to Putin and his buddies. Would he really decline out of sheer spite?

«About 1.2% of U.S. adult men and 0.3% to 0.7% of U.S. adult women are considered to have clinically significant levels of psychopathic traits».

But it has to be a special kind of psychopath. Psychopathy only implies a lack of empathy and an antisocial personality. Such a person might deny paradise for others for personal gain, but they would have no reason to do so out of sheer spite, unless they were also particularly sadistic psychopaths.

Yes I'm sure. Technically it would incur some opportunity cost (that most would say a singleton should ignore, although I'd argue that an ambitious singleton may still find the cost intolerable). More importantly he doesn't seem to be able to contemplate such relationships. He's not just a Machiavellian psychopath but, first and foremost, a rat, suspicious and hateful of others. It's not only spite, it's a hard prior: sharing with people not in your family (or mafia) is expensive and gives them dangerous capabilities to harm you.

Real people are not rational economic agents – not only in the positive sense (having morals) or neutral sense (having cognitive biases and weird beliefs), but negatively too (willing to do net-negative things for no personal gain). He would not reason «hurr durr I'm evil I will crush those peasants even though it provides me no benefit» – he'd a priori distrust and reject the analysis that says that sharing paradise with peasants is cost-free. They could always revolt or something, and they'll be eating his resources, resources he could put to better use… It's just unpleasant. And this is a not so rare an attitude. Imagine the most stiff-necked, work-obsessed American Puritan from this forum, looking at junkies at the streets of SF. This is the merest shade of the contempt one can feel for non-productive people.

This is not purely hypothetical. Just one example: Russia (pop. 147m including Crimea) has 17 million square kilometers of territory. Much of that, even livable parts, is very sparsely used or effectively uninhabited. Once, there was an initiative to stimulate economic activity and improve the real estate situation etc. with granting people large plots of land. By the time it was approved, it was downscaled to no more than 1 hectare (0.01 sq. km) per applicant on the Far East (area ≈7 million, pop. ≈8 million and falling), located at least 10 km away from towns over 50k and 20 km away from towns of 300+k, so effectively in wild boreal forest, in groups no more than 10 people (so, you are capped at 0.1 km^2 for your project), and with certain onerous rules of use (to begin with, you have to prove you're developing it over the first 5 years, lest it be taken away, and making it your actual property is not straightforward). Russian Far East… isn't very paradisiacal. It was not just affordable but very lucrative for them to offer Russians better terms. Still they didn't, so basically nobody bothered with the program. Lose-lose. (Hilariously, they've dedicated over 1.4m kilometers in the first round of the project – a hectare per Russian).

(At the same time, entire swathes of the region are leased for pittance to China.)

Speaking of paradise, Krylov:

…But Russia had no overseas colonies and did not even dream of them. Moreover, when enterprising Russians (usually against the wishes of the Russian government) tried to settle somewhere across the sea, the Russian government firmly suppressed their impulse. Alaska was sold to the Americans for a pittance. Specifically, for $7.2 million, the equivalent of $104 million today. That is, Alaska was sold for less than five dollars per square kilometer. The price of the transaction included all the movable and immovable property located in Alaska. The payment was not made in gold, but in non-cash dollars. Whether this money was received by the Russian government is not known - [it seems not] (http://www.opoccuu.com/alyaska.htm). And, of course, the board of the Russian-American Company knew nothing of the government's preparations to sell Alaska: the deal was being prepared in secret.

Again: five dollars per square kilometer is the red price of "territorial integrity."

It is clear that Russian settlements in California were, as they say now, " wrapped up", and the Russian-American Company - the same one which sponsored the Krusenstern - was liquidated.

However, one could at least say that Russia was desperately afraid of the mighty America, and that Alaska was "so empty and cold". But when the naive Miklukho-Maklai suggested to Alexander III that he create a "free Russian colony" (as opposed to Germany) in New Guinea, he was kicked out of the Czar's office with such force that he landed in Australia. In Sydney he settled more than fine, but returned to Russia and again came out with the same proposal at the highest level - with the same result.

This has always been the case. All proposals of Russian enthusiasts on the topic of development of any lands far removed from the motherland were met with an absolutely rigid "no". Russia, represented by its highest leadership, shoved away any overseas possessions. On the other hand, enormous efforts and funds were poured into Poland, Finland, and Georgia. These «extraordinarily valuable» acquisitions cost Russia very dearly, and in all senses. So much effort, life and resources were put into fiddling with the Poles and Chukonka people that it would have been enough for a full-fledged colonial expansion.

There can only be one reason for such a decisive rejection of the Paradise Islands. Fear. A conscious and clear fear of the Russian government that the paradise islands will have to be populated by Russians. Which may subsequently sever ties with the empire and create at least a small, but their own state. Which could, in turn, serve as an "embarrassing example" for the rest.

Do you think this psychology allows for your do-gooder decisionmaking?

I also recommend watching Expelled from Paradise on a similar topic.

This comment has made me think a bit harder about my assumptions. Perhaps such a spiteful disposition is more common than I had previously believed.

Where the hell does this distrust of breakaways come from? Has it really always been this way with Russians, or is this more an occasional tendency that sometimes flares up to imperial degrees?

Once, there was an initiative to stimulate economic activity and improve the real estate situation etc. with granting people large plots of land. By the time it was approved, it was downscaled to no more than 1 hectare (0.01 sq. km) per applicant on the Far East (area ≈7 million, pop. ≈8 million and falling), located at least 10 km away from towns over 50k and 20 km away from towns of 300+k, so effectively in wild boreal forest, in groups no more than 10 people (so, you are capped at 0.1 km^2 for your project), and with certain onerous rules of use (to begin with, you have to prove you're developing it over the first 5 years, lest it be taken away, and making it your actual property is not straightforward).

In comparison, Canada gave out 160 acres (65 ha) of land per household via its Dominion Lands Act, out of which you had to cultivate a quarter and build a permanent dwelling on it in three years. That's sixteen times more land if you use a replacement-level family of four as your benchmark. And you could double your homestead after you improved your original plot enough, giving you 320 acres (130 ha). They also had an exclusion zone around railways, but realized it made the program much less attractive and got rid of it.

4401,2 thousand ha of farmland lay fallow in Russia in 2020. That's 27 507 Canadian-style plots they could have given away, about half of them in praire-like conditions of Siberia. What's worse, there was 36256,3 ha of farmland that was technically not fallow but wasn't worked by its owners. I could buy 3200 ha of farmland in Hakassija for the price of a gaming PC if I wanted, why would I ever be interested in a single hectare in the middle of nowhere?

I could buy 3200 ha of farmland in Hakassija for the price of a gaming PC if I wanted

What's the catch ? Climate's not good for farming there ?

More comments

OK so they press the 'create paradise' button but they'd still keep the master key to themselves. Then, if there are any disputes or troubles in paradise...

Even with vast resources, there will still be questions of allocation. Posthumans will probably reproduce very quickly or have very high material aspirations or both.

I think the problems of such a future age would be so divorced from those of the present day that it would be difficult to predict from their present-day positions and motives and histories whether Sam Altman or Xi Jingping or anyone else would be particularly likely abuse the keys to the kingdom.

Indeed, I don't trust anyone to control the lamp with the genie in it. We should not be creating powerful entities with alien mindsets, certainly not letting anyone have monopoly access to them.

Okay, but my only point is that I don't think it makes a real difference for better or worse whether China creates AGI or whether the US does, whereas a lot of people think it does make a real difference.

Who says there are two buttons?

The risk with a geopolitical rival isn’t that they’re mustache twirling psychos. It’s if there’s only one button, and it does boring normal international stuff, except better. Stuff like bombings or economic pressure. If the AI is just going to give someone an advantage, I want that someone to be me.

Well, I guess there’s also the risk that it has zero buttons, and just starts killing. Then it doesn’t really matter who made it. Unless you think some countries are more likely to be careful than others.

Lesser AI systems, yes. But superintelligence, if both the doomers and optimists are to be believed, will have such a power that it will be capable of creating a paradise, in which case it doesn't matter who gets it, because that's what every human on earth would ask it to do with the exception of a few lunatics who are vastly unlikely to be the ones in the position to make the decision. Or it will kill everyone because we couldn't figure out how to get it to follow instructions, in which case it doesn't matter who gets it.

what is an anti-AI US going to do about it?

Start a nuclear war, according to Yud (in case the OP was vague I absolutely don't agree that "create totalitarian world government to stop AI" is in any way a good idea). Otherwise the US would have to convince Chinese leadership that AI research is tantamount to pressing a suicide button.

Even if I thought there was a 99% chance of AI destroying humanity, creating a massive totalitarian world-state that tracks all private behavior and is willing to start nuclear wars to enforce its power doesn't seem like an improvement. What Yud is proposing is probably not possible, but if it is possible it's one of the worst futures I can imagine for the human race.

A "massive totalitarian world-state that tracks all private behavior" is not being called for here, nor is it likely. What's being proposed is much more like existing arms control, biosafety treaties between existing countries. Manufacturing and use of large-scale GPU clusters would be regulated in much the same way other large-scale industrial processes are regulated. "The evil fascist totalitarian state is coming for you" is a staple of the political imagination, but we're very far away from the nazis or soviets today, and this closes the gap by .00001%.

Would there be such a major change here? Anti-mass-destruction-proliferation sentiment becoming strong enough to prompt airstrikes on suspected violators is a thing that first happened more than 40 years ago ... prompting some outrage, but also creating strange bedfellows, even bringing some of the original lukewarm critics down the same rathole eventually.

probably not possible

Not even close. Anti-nuclear-proliferation military leadership could easily point to a history with nuked cities and much-much-bigger-nuke tests, and their actions were still on the continuum from "controversial" to "outright mistaken". Imagine how crazy they'd have looked in a world where nuclear energy had so far only ever given us power plants, where nobody had ever seen a nuclear warhead! But any unfriendly AGI with decent I isn't going to give us any early close calls to point to; there'd be no point in it doing anything destructive until success was certain.

creating a massive totalitarian world-state that tracks all private behavior and is willing to start nuclear wars to enforce its power doesn't seem like an improvement

No one is asking for that.

Pushing the SotA on LLMs takes millions of dollars. Pinpointing and regulating institutional actors who are doing training at that level is a lot easier than trying than trying to stop each and every individual from running LLaMA-7B in their basement at home. The latter is not feasible, which is why no one is asking for it.

We already threaten nuclear war if Russian troops invade Latvia, a country that was part of the Soviet Union a mere 32 years ago. We already invade countries that we suspect of violating nuclear non-proliferation (Iraq). The whole (non-cynical) point of the giant globe-spanning American military is to deter and destroy threats to the civilizational order.

I also think people are underestimating the willingness of rival powers to agree to this. You think Russia and China like their odds in an AI arms race? If USG made a credible offer of “AI non-proliferation,” how do we know it wouldn’t be accepted? “Maybe don’t build god,” is only considered an unreasonable proposition by SF techbros.

We already threaten nuclear war if Russian troops invade Latvia, a country that was part of the Soviet Union a mere 32 years ago.

You might, but NATO states don't. Nuclear retaliation is reserved for nuclear attacks, while conventional forces for conventional attacks, the conflation of which ignores how MAD interacts, and doesn't interact, with conventional deterence.

We already invade countries that we suspect of violating nuclear non-proliferation (Iraq).

The United States also does not invade countries known to be violating nuclear proliferation, and suspected of violating nuclear proliferation, and of course those actually having nuclear weapon. Because, again, MAD.

The whole (non-cynical) point of the giant globe-spanning American military is to deter and destroy threats to the civilizational order.

No, the point of the globe-spanning American military is to advance American security interests for the American alliance network. The Americans do not generally invade their allies, or their enemies when the cost to the Americans is too high, or when their domestic political order has higher priorities than destroying other people.

I also think people are underestimating the willingness of rival powers to agree to this. You think Russia and China like their odds in an AI arms race?

Them not liking their odds in an AI arms race is why they would offer a treaty, and cheat, for the same reason they (and most major powers) cheat to very degrees on other limitation agreements.

Cheating is the expectation in genuinely limiting international agreements, whether it be loophole abuse, redefinition of contested items, or blanket denial.

If USG made a credible offer of “AI non-proliferation,” how do we know it wouldn’t be accepted?

Because we know many of the American's own citizens don't find the US Government credible, let alone many other countries, and especially the geopolitical adversaries.

Because, of course, the US has many means to cheat, and would have an even greater incentive to cheat the more other parties genuinely gave up on a potential competitive advantage / competitive parity.

suspect of violating nuclear non-proliferation (Iraq)

This was the canard that led to the invasion. In truth the intelligence would have been whatever was necessary to produce the desired outcome.

You think Russia and China like their odds in an AI arms race?

If the US is willing to cripple itself? You betcha. They'd agree and then go ahead and build their own AI.

You think Russia and China like their odds in an AI arms race? If USG made a credible offer of “AI non-proliferation,” how do we know it wouldn’t be accepted?

Credible means enforceable. Isn't this Dark Forest situation where malicious signatories can continue to build AI in secret? If so, even non-malicious signatories will feel compelled to develop AI in secret to avoid getting shivved in the back.

It's also worth noting there is, basically, one other country in the world that can really present a challenge in the AI Arms Race, and that's China. Does anyone here think that the US is willing and able to conduct airstrikes on Chinese territory on the chance they're violating the AINPT (AI Non-Proliferation Treaty)? Because I don't.

Right. My problem with AI doomers is their inability to imagine an alternative I would actually prefer to paperclipping. If this is the solution to the problem, I'd rather the problem not be solved.

Devil’s advocate, but “tracks all private behavior” isn’t necessary; only controlling by the flow of high-tier computing resources.

I know that the ability—or will—of a state to stop at that point is rather suspect.

"Only" controlling the flow of high-tier computing resources is really hard, especially when that control needs to be exercised across the entire globe. And the basic elements (GPUs) presumably won't be outlawed, just large clusters. And it's not enough to shut down 99% of clusters, since even a few slipping through the cracks is still a major threat if Yud's argument is correct. Absent panopticon-like surveillance and control, how would this be even remotely feasible?

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats