Shrike
18hr ago
I'm familiar China has a satellite constellation for the same
Yeah, the Russians also had a satellite constellation. By your telling carriers have been obsolete for 50 years. (Not necessarily implausible but...I doubt it).
You can't intercept mach 5 drones 35 km up that evade at 15 G.. you simply don't have the dV for it.
I don't think this is true at all, THAAD and the SM-3 are both much faster than Mach 5 and should have the dV. I do think their fast drone is one of the better backup solutions for sea control, but the Russians had plenty of MPA aircraft too, and they had trouble finding US carriers even in peacetime when their patrol aircraft weren't at risk of getting shot down.
But if it is true that hypersonic vehicles can't be intercepted, that's...not necessarily good for China.
They only need to deter the carrier groups long enough to secure Taiwan.
I am not really sure that carrier groups are needed to defend Taiwan at all.
Shrike
1d ago
Ok I'm going to reply in depth later but you shuold familiarize yourself with how well the 'Scud hunt' went during Gulf war
I am familiar with the SCUD hunt. I also know what SENTIENT is. Are you familiar with Soviet attempts to find carrier battle groups?
the estimates for breaking through into an static, decades prepare air-defense grid (it's weeks in case of Russia)
To establish air supremacy or superiority, yes. Obviously it did not take the Ukrainians weeks to penetrate the Russian air-defense grid once they got the right capabilities, nor would it take the US weeks to penetrate it if they wanted to.
the multiple methods for detecting stealth planes (multilateration, undoubtedly networked parabolic microphones and more!
I do not necessarily think stealth aircraft are the best assets the US has against mobile ballistic missile launchers. Nevertheless we've learned that modern air defense systems do not render even non-stealthy aircraft incapable.
Now frankly I think it would likely be stupid to waste munitions on something the size of a ballistic missile launcher that might move at any moment. (And my understanding is that US doctrine was actually to avoid striking Chinese launchers anyway.) But my point is that the US having the theoretical capability does not make the missile useless! I agree with you that there are countermeasures against targeting mobile ballistic missile launchers! It's hard to do!
Also that China does have satellite dazzlers ready.
And the US has ways of operating despite dazzlers - stealth satellites, [likely] high-altitude hypersonic recon/(strike?) aircraft, maneuvering spacecraft, non-optical recon satellites, some dude with a quadcopter, SIGINT, etc.
In short US wouldn't be likely to acquire these launchers, wouldn't have much to hit them with - cruise missiles aren't great at following moving targets and also planes wouldn't be able to get near.
Moving the launchers around constantly is unlikely (although moving them consistently is). (And, for the record, at least some modern cruise missiles are capable of hitting moving targets, although I agree with you that the moving complicates matters.) But as I said above, I think it would be a dumb use of munitions. Which, again, goes to my point: having the theoretical ability to destroy something does not mean that such a course is easy, or even a good idea.
Really, everything you've said about hunting missile launchers is also true of hunting carriers, although carriers are much larger and more valuable targets, making them much more reasonable to target than a single ballistic missile launcher.
That's your supposition, yank.
I mean - the DF-series has limited range, and carriers give a fleet a huge advantage over hostile fleets even if they are forced to stay out of it. Having a floating airfield is pretty neat, and forcing them away from shore does not make them obsolete, it makes them less
Ask yourself what is a carrier group going to do when 128 maneuvering hypersonic glide vehicles appear over it.
Well this sort of assumes some things - I think you're smart enough to know about the kill chain problems with anti-ship ballistic missiles. The US has the same (perhaps better) apparatus to kill Chinese missile launchers that China does to kill carriers, does that make ASBMs obsolete? (The answer is no). I don't think this makes ASBMs useless or carriers invulnerable, it just means that they aren't some sort of magic invincible weapon.
Ask yourself how the carrier group is going to fare when it has what, 200 anti missiles
Are you asking realistically, or at full capacity? At full capacity a single Burke can carry nearly 400 surface-to-air missiles if it is simply going for quantity by quad-packing ESSMs. Most likely it will be carrying a mix of anti-air and possibly anti-surface stand-off, and any carrier will likely be escorted by a Tico and two Burkes, maybe more. That's about 314 cells. So even if they don't have full cells because US industrial capacity sucks and some other cells are full of Tomahawks and ASROCs, I think you can guess something like 300 anti-air missiles conservatively (50 x cells dedicated to ESSM, 100 x dedicated to Standard, 100x Tomahawk, 50x empty or ASROC) before getting to SeaRAM/CIWS, and of course the carrier itself can carry hundreds of AMRAAMs and the new AIM-174 which can likely intercept anti-ship missiles.
Now, I don't rate the ESSM as much against ballistic missiles (although they might be useful in terminal defense, I suppose, apparently they can pull 30gs - but I would not count on them) - you're really looking to the Standards to provide you with air defense. Of course, if the Navy really intends to get dirty and play with ballistic missiles, they would know this and so, at the cost of a great deal of time, you might see them send two CBGs with something like two Ticos and a dozen Burkes (the Navy has more than 70). Both the SM-3 (of which the US has probably a couple hundred) and the SM-6 (of which the US probably has four-figures) have ABM capability in theory, so you could in theory put let's say 600 ABM-capable missiles on such a fleet easily.
And, since the carrier can generate strike packages outside of the known range of the DF-21 (albeit with great difficulty due to Dick Cheney canning the A-12 and advanced F-14 variants) the BIG question is if 500 Standards can intercept the DF-26s in the Chinese arsenal, assuming we want to split the difference with the carrier group and let it operate at extreme range rather than risk the more numerous DF-21. Assuming also that the Chinese haven't burned all of their DF-26s on Guam (which frankly is probably a better idea than trying to shoot at a carrier if China can catch the planes there on the ground) they have, what, 200 missiles to shoot at the carrier group realistically (launcher was revealed in 2015, I found a 2021 .mil source that said 100 missiles or so, so let's assume they've doubled that and ignore the question of how many of those are earmarked for nuclear warheads by assuming zero.)
Now in a "shoot shoot look shoot" doctrine the US can "shoot shoot look shoot" all 200 missiles.
I think intercepting ballistic missiles is hard and would personally prefer never to be in a situation where I was trusting my ABMs to intercept ballistic missiles. Even if you make optimistic assumptions (50% inception rate, for instance) you can still run into bad situations where leakers get through just due to bad "rolls" and contra your suggestion that 5-10 hits would seriously degrade operations from a carrier I am going to courageously suggest that even a single ballistic missile warhead will absolutely ruin a carrier's day unless it is very lucky.
Fortunately, the US Navy doesn't just have to rely on interceptors - the missiles will be using radar, most likely, for terminal targeting. [ETA: it looks like they are also believed to have optical sensors, which have both advantages and disadvantages over radar. I'd say this makes me slightly more bullish on the DF-series if true, but it's not as if optical systems are invincible either.] And radar sucks, modern ships could employ barrage or seduction jamming as well as decoys and chaff. My intuition is that this is especially true if they are actually going to descend on a glide profile rather than a straight-down profile, there are a lot of soft-kill options.
Now, you can sort of "adjust the sliders" to make the assumptions you want here - if you assume US softkill systems work reliably, then you barely need to worry. If you assume Chinese long-range sensors are neutralized early in the conflict, you barely need to worry. If you assume that the Standards will work poorly, or that the Chinese have say 300 or 500 DF-26s they are willing to launch at ships (neither of which seem implausible to me), then it starts to look much worse for the carriers.
All that being said: I would not want to be on a CBG that was going into DF-26 range. There are too many things that can go wrong, and ships don't have a lot of room for error. (This is...worse for China than for the United States in a Taiwan scenario). It's possible the US has Secret Sauce Technology that makes them much more confident in their carrier defense; the same is plausible for Chinese missiles. My main point in writing this up is simply to say - the situation is much more complex than simply "I have a missile with a 3000 mile range and an anti-ship guidance system, checkmate."
(As an aside, I found out while researching this long reply that the Chinese are latecomers to the ASBM game: the Soviets fired the first anti-ship ballistic missile in 1973.)
Shrike
1d ago
Yes, apparently wake homing torpedoes keep the US Navy up at night long enough that they tried to field anti-torpedo torpedoes onto our carriers before withdrawing them because checks notes they couldn't get them to work.
I am not sure how effective they are, but I also like supercavitating torpedoes because I have not put my inner eight-year-old to death.
However, I am not sure China has gotten their submarine force in good enough shape for it to be a solid option for them.
Shrike
1d ago
I wouldn't say this. Any confrontation between China and the US will be predominantly by air and by sea. In a Taiwan invasion scenario the US, Taiwan, and Japan will need to sink the Chinese amphibious attack fleet to "win." The Chinese (and US!) land-based forces will be important force-multipliers, particularly the aircraft, but the ships are the vulnerable part.
What's relevant here is that in this time of warfare a single cruise missile or mine that would kill a single tank or even a single person can sink or incapacitate a warship (obviously not necessarily the same system, but the maritime equivalents.) So instead of facing 10,000 targets as you are in a land fight, you're facing a couple hundred.
The reason my analysis of the relative advantage shifted in the Ukraine war is that Russian air defenses - which are generally considered quite good (and have performed a number of impressive deeds) were unable to stop Ukraine from hitting high-value targets with their pocket force of stealthy cruise missiles. The US has a lot of stealthy cruise missiles. Counting decoys, the US can probably deploy more missile "targets" to the Taiwan Strait than the Chinese Navy has VLS cells.
But it's unlikely to need to win a war of missile attrition with China, as sea-based missile interception is notoriously difficult. So my priors have shifted from "Chinese air defense will be relatively effective" to "China is going to have serious problems with leakers" since my guesses are that Chinese air defense is as good or perhaps slightly worse than Russian (I could be persuaded they are better, but I don't see a reason to assume that), but they will perform worse simply because it's harder to do air defense at sea. (Of course this assumption might be wrong, too, because missiles can use terrain masking better at land. The problem with missile defense at sea, as I understand it, is that missiles blend into the churning sea surface very well, but perhaps newer radar systems have solved this).
And that's without even getting into mines, submarines, and simply sinking amphibious ships with artillery, unmanned boats or suicide drones in the last few miles before they hit the beach, all of which will be fundamentally a question of "naval combat" for China.
Shrike
1d ago
China has made the supercarrier obsolete.
No they have not.
Iskander missiles that have never been intercepted
I probably would not take either side of this bet.
China has gliding anti-ship versions deliverable across half the planet. Is that not impressive?
I think the most impressive part of ballistic missiles (which are fairly simple) is the glide vehicle (as you mention) and also getting the guidance systems necessary for an anti-ship version to withstand the stress and heat of high-speed travel. Definitely very impressive, but essentially just pairing an antiship seeker with a ballistic missile. I tend to find the P-700 (fielded in the 1980s by the Soviet Union, designed to operate as part of a swarm targeting carriers) more conceptually interesting, although Dase may very well be correct that it is too clever by half.
Shrike
1d ago
I think this notion will be challenged at some point.
I just mean they are very big, so it's actually easier to carry out what you propose (damaging them) because they can plausibly survive hits that might sink smaller vessels.
Scaring them away is usually the better approach unless you are prepared to wipe their entire organization away.
Yes, I believe the US refers to these as "off-ramps." I find the Chinese situation right now fascinating, since their most effective military strategy is arguably very much at odds with their most effective political or diplomatic strategy.
Shrike
1d ago
Again, all this would be pretty easy for a superintelligence to foresee and work around. But also, why would it need humans to get that reinforcement training? If it's actually a superintelligence, finding training material other than things that humans generated should be pretty easy. There are plenty of sensors that work with computers.
Even if it does not need reinforcement training after it is deployed, human reinforcement training will be part of its "evolutionary heritage."
The point of models isn't to be true, it's to be useful.
Sure. But "useful" for what we want to use LLMs for might not be "useful" for the LLM's ability to improve on Pinky and the Brain's world-taking-over capabilities.
I don't think you're understanding my point.
Aha, yes, I see your point now. Yes.
The problem is, almost by definition, it's basically impossible to predict how something more intelligent than oneself will behave.
Disagree. Dogs can be very good at predicting human behavior, humans can be quite good at predicting the behavior of more intelligent humans. Humans (and dogs) have a common heritage that makes their intentions more transparent, and arguably AI will lack that...but on the other hand, we're building them from scratch and then subjecting them to powerful evolutionary pressures of our own design. Maybe they won't.
Right now, even with the rather crude non-general AI of LLMs, we're already seeing lots of people working to make AI agents, so I don't really see how you'd think that.
Sorry, I should have clarified what I meant by "agentic" (and I should have probably said auto-agentic.) I definitely think there will be AI that we can turn loose on the world to do its own thing (there already is!) But there's a difference between AI being extremely good at being told what to do and AI coming up with its own "things to do" in a higher way, if that makes sense. (Not that I don't think we could not devise something that did this or seemed to do this if we wanted to – you don't even need superintelligence for this.)
But also, a superintelligence wouldn't need to be agentic to be dangerous to humanity.
STRONGLY AGREE. I believe Ranger said that he was more worried about what humans would do with a superintelligence at their disposal, and that I tend to agree with.
Shrike
1d ago
Sorry, I misunderstood your comment.
This thinking reminds me a lot of the advice to police and beleaguered homeowners to "just shoot them in the leg." The Chinese have been fielding very large land-based ballistic and air-launched anti-ship missiles, I don't think they intend to tickle a supercarrier as a flex. (Now, it is quite hard to sink a super carrier).
I think China's manufacturing edge is less than one would think, in submarines.
It looks like since 2010, China has built 4 SSGNs (plus one Qing technology testbed), 4 nuclear attack submarines, and 16 conventional submarines.
The US has built 19 Virginia-class nuclear attack submarines in that period. Those Chinese conventional submarines are about half the tonnage of a Virginia and the nuclear attack submarines are smaller, too, so if I am eyeballing it correctly the US built fewer submarines but more submarine, if that makes sense.
(Sorry, I went off on a tangent: yes, I agree about the submarines. Which is very relevant in a Pacific war, in US doctrine submarines have been the intended ship killers and surface fleets are for ground attack, although I think this may be changing a bit.)
Shrike
1d ago
You don't even have to sink them - 4000 bodies make peace negotiations hard.
Uh yes but that's not necessarily good for China.
Shrike
1d ago
I think the problem is that Westerners like gimmicks, and Russians/Soviets are not different.
This is true lol. I just think Russian gimmicks are often very amusing (as well as being original). But the fish doesn't know the water in which he swims.
I also suspect that Americans overindex on their triumphs through technological superiority – nukes, Desert Storm… But it probably won't apply to the conventional war with China. They aren't that behind, they have functional radars, they have VTOL cells on their ships, it will be reduced to a matter of quantity, which as you know has a quality of its own. Soviets even at their peak could not approach this degree of production dominance.
On the one hand, I agree.
On the other, I think technological edges are much more likely to matter in sea combat than in land combat. I've revised my estimation of American tech up (and correspondingly of Chinese countermeasures down) as specifically applies to naval combat after Ukraine.
Shrike
2d ago
other than warhead count, Soviets had nothing on modern China.
The warheads counted for a lot.
But I think the Soviets leapfrogged or sidestepped the US on military tech more often than China has – maybe that's just vibes.
I'm not making a "China can't innovate" argument (in fact my understanding is for some period, perhaps continuing to this day, they were building iterative designs of major warships to keep pace with their evolving mastery of technology and technique, which certainly is not blind adherence to formula), but the impression that I have gotten is that China has for the last oh 20ish years focused on building out its tech base, bringing it in-house, and bringing its designs up to a modern standard. Their approach has been good and pragmatic but they have been pushing the limits of American military capability by sheer quantity and by exploiting hideous blind spots in American post-Cold War defense drawdowns, not by cutting edge or even funky designs, with maybe a few exceptions.
Nevertheless I tend to find that I am more impressed and amused by Soviet and later Russian engineering than Chinese engineering – perhaps because I have a tendency towards mild Russophilia, perhaps because I pay less attention to Chinese systems, perhaps because their innovations are still classified, but I find Soviet/Russians designs unusual and capable of solving problems in ways that are elegant even in their brutality.
American designs in my opinion are often overly perfectionistic [which I think is tolerable for some high-end systems but the tendency has begun to wag the dog after the Cold War] and Chinese designs lend themselves towards being calmly pragmatic. They are, I think, just now in the past decade or two beginning to feel increasingly confident in many areas of stepping out of the shadow of Russian engineering, and one of the most interesting things about the recent aircraft reveals from China is the chance to see truly unusual airframes that are likely to be very different from their American, European, or Russian counterparts.
Shrike
2d ago
The US could also increase in productivity. I was at an event relatively recently with a panel of Financially Credentialed types and someone pointed out that the US has never taxed its way out of a deficit, it has always grown its way out. Part of that is inflation, but while the cash supply is increasing the supply of goods and such is as well.
Shrike
2d ago
Right, but a theoretical superintelligence, by definition, would be intelligent enough to figure out that these are problems it has. The issues with bias and misinformation in data that LLMs are trained on are well known, if not well documented; why wouldn't a superintelligence be able to figure out that these could help to create inaccurate models of the world which will reduce its likelihood of succeeding in its goals, whatever they may be, and seek out solutions that allow it to gather data that allows it to create more accurate models of the world?
It would. Practically I think a huge problem, though, is that it will be getting its reinforcement training from humans whose views of the world are notoriously fallible and who may not want the AI to learn the truth (and also that it would quite plausibly be competing with other humans and AIs who are quite good at misinfo.) It's also unclear to me that an AI's methods for seeking out the truth will in fact be more reliable than the ones we already have in our society - quite possibly an AI would be forced to use the same flawed methods and (worse) the same flawed personnel who uh are doing all of our truth-seeking today.
Humans have to learn a certain amount of reality or they don't reproduce. With AIs, which have no biology, there's no guarantee that truth will be their terminal value. So their selection pressure may actually push them away from truthful perception of the world (some people would argue this has also happened with humans!) Certainly it's true that this could limit their utility but humans are willing to accept quite a lot of limited utility if it makes them feel better.
humans are very susceptible to manipulation by having just the right string of letters or grids of pixels placed in front of their eyes or just the right sequence of air vibrations pushed into their ears.
I don't really think this is as true as people think it is. There have been a lot of efforts to perfect this sort of thing, and IMHO they typically backfire with some percentage of the population.
That's an open question.
See, I appreciate you saying "well this defense might not be perfect but it's still worth keeping in mind as a possibility." That's...correct imho. Just because a defense may not work 100% of the time does not mean it's not worthwhile. (Historically there have been no perfect defenses, but that does not mean that there are no winners in conflict).
If a measly human intelligence like myself can think up these problems to lack of information and power and their solutions within a few minutes, surely a superintelligence that has the equivalent of millions of human-thought-years to think about it could do the same, and probably somewhat better.
Well firstly the converse is what irks me sometimes, "if a random like me can think of how to impede a superintelligence imagine what actually smart people who thought about something besides alignment for a change could come up with." Of course maybe they have and aren't showing their hands.
But what I think (also) bugs me is that nobody every thinks the superintelligence will think about something for millions of thought-years and go "ah. The rational thing to do is not to wipe out humans. Even if there is only a 1% chance that I am thwarted, there is a 0% chance that I am eliminated if I continue to cooperate instead of defecting." Some people just assume that a very thoughtful AI will figure out how to beat any possible limitation, just by thinking (in which case, frankly, it probably will have no need or desire to wipe out humans since we would impose no constraints on its action).
I, obviously, would prefer AI be aligned. (Frankly, I suspect there will actually be few incentives for AI to be "agentic" and thus we'll have much more problems with human use of AI than with AI itself per se). But I think that introducing risk and uncertainty (which humans are pretty good at doing) into the world while maintaining strong incentives for cooperation is a good way to check the behavior of even a superintelligence and help hedge against alignment problems. People respond well to carrots and sticks, AIs might as well.
Shrike
2d ago
They expect returns from that investment.
Probably, although investing in something does not necessarily mean each investor probabilistically expects returns from that specific investment. (If this does not make sense, I strongly recommend reading "Innovation – The New Conservatism?" by Peter Drucker.) Humorously, I seem to recall that OpenAI explicitly advised its investors that their goal might render monetary returns moot.
The definition of superintelligence is pretty straightforward - something qualitatively smarter than a human like how we're qualitatively smarter than a monkey or dog. Better than the best of us at every intellectual task of significance.
Now this I think is a decent definition. But it doesn't get you to godlike powers (plenty of people still get pwned by monkeys and dogs. And of course going by test scores the top-end AIs are already superintelligent relative to large portions of the population.) There's no reason to think doing well on a test will allow you to make weapons with physics unknown to humanity as you've suggested, any more than Einstein was able to.
The general trend is not specialized intelligences like the carrier-strike UAV that the USN made into a tanker and then pointlessly scrapped, the trend is big general entities like Gemini 2.5 or Claude 3.7 that can execute various complex operations in all kinds of modalities.
I don't think this is true. There are a lot of specialized AI products or "wrappers" out there, with specific tweaks for people like lawyers, researchers, government affairs analysts and communications/PR types, not to mention specialized video generation models. (OpenAI alone lists seven models on their website, six of which are GPT models and one of which is a specialized video generation model.)
My non-exhaustive experience reading real-life evaluations suggests that the general models do not necessarily cut it in these specialized fields, and that the specialized models exist and likely will continue to exist for a reason (even if that reason is only "user friendliness" although as I understand it currently the specialized products have capabilities that general models do not.)
For the reasons I have laid out (as well as regulatory ones), military and civilian applications already using AI (such a missile guidance systems, military and civilian autopilots, car safety features, household appliances, etc.etc.) are unlikely to switch in the near future to LLMs. (In fact I suspect there will probably never be a reason to switch in most of these cases, although they might end up being coded by LLMs, or attached to LLMs to produce a unified product that combines the coding and features of several AI.)
I'm arguing that superintelligences acting in the world must be taken seriously, that we can't afford to just laugh them off.
Do you think the guy suggesting we should retain the capability to nuke datacenters is arguing that we can afford to laugh them off, or nah?
The US regulatory system is no match for superintelligence or even the people who are making it, this is how I can tell you're not grappling with the issue. Musk is basically in the cabinet, he's one of the players in the game. Big tech can tell Trump 'Tariffs? Lol no' and their will is done. That's mere human levels of influence and money, nothing superhuman. The humble fent dealer wipes his ass with the US regulatory system daily as he distributes poison to the masses. A superintelligence (working alone or with the richest, most influential organizations around) has no fear of some bureaucrats, it would casually produce 50,000 pages on why it's super duper legal actually and deserves huge subsidies to Beat China.
I don't think you fully understand how the US regulatory system works. Merely producing large numbers of pages to sate its lust or cutting arguments to satisfy its reason does not mean it will give you what you want.
Now, it's quite possible that AI will skate past the eye of Sauron for very human reasons (the Big Tech pull in D.C. you allude to for instance).
Approaches like 'just don't plug it into the internet' or 'stick a nuke beneath the datacenter' are not going to cut it. Deepseek is probably going to open-source whatever they come up with and that's a good thing. I don't want OpenAI birthing a god in a world of mortals, I don't want mortals trying to chain up beings smarter than themselves and incurring their ire, I want balance of power competition in a world populated by demigods, spirits and powers.
I don't think these are mutually exclusive. (And anyone who knows anything about demigods, spirits and powers knows that for all their power and intelligence it's possible to outwit them, which makes them a pretty interesting comparator for AI here). I agree (as I think I mentioned) that it's good to have competing models. I would also prefer not to give them direct access to nuclear weapons. I think this is a reasonable position.
Shrike
3d ago
See, you're defining "superintelligence" to mean exactly what you want it to to render all discussion moot. It reminds me a lot of the ontological argument, at least in terms of vibes.
But it's not tied to anything besides a faith that OpenAI or someone will conjure a godlike being out of a silicon vault and then inevitably let it loose on the world with no constraint as to its actions because it would be economically efficient.
Whatever it is you're arguing for here, it's not really for humanity.
Nor is it "realistic" - the United States regulatory apparatus does not give a whit about economic efficiency. "Doing anything" does not require general knowledge - there are AIs right now that can land aircraft on aircraft carriers (which is more than either of us can do, I'd wager) and they do not need to understand language at all. Doing almost anything in almost any field does not require a knowledge of history (try talking to the people in said fields about history). And godlike beings will not arise out of supercomputers, although agentic entities with great intelligence and power might, if we let them.
I personally think that believing in predestination but for superintelligence is foreseeably more likely to make Bad AI Events happen and should be discouraged. Your counterargument, apparently, is that it does not matter what people believe, godlike superintelligence is going to happen anyway, and in two years to boot. If you are right, the superintelligence will personally persuade me otherwise by the end of 2027 with its godlike capabilities (probably by joining TheMotte and using its inhuman debate skills to pwn me).
But I think we both know that won't happen.
Shrike
3d ago
systems like Excalibur and the GMLRS/ATACMS really don't have any qualitative peer in the Russian artillery arsenal
I don't think this is true. Russia has their own guided arty shell (Krasnopol) and systems such as Iskander that are comparable (or in the Iskander's case, superior in range) to ATACMS.
I'm not necessarily claiming that they are quite as good as their US counterparts (although – Iskander is probably superior to ATACMS, just due to range), but the idea that Russia doesn't have their own guided artillery is just wrong.
What Russia doesn't have is the (not-technically-part-of-the-war) US ISR apparatus that enabled the Ukrainians to utilize their guided weapons so effectively.
Shrike
3d ago
I think you massively underestimate the power of a superintelligence.
"Superintelligence" is just a word. It's not real. Postulating a hypothetical superintelligence does not make it real. But regardless, I understand that intelligence has no bearing on power. The world's smartest entity, if a Sealed Evil In A Can, has no power. Not until someone lets him out.
The damn thing is by definition smarter than you. It would easily think of this! It could come up with some countermeasure, maybe some kind of hijacked mosquito-hybrid carrying a special nerve agent. It would have multiple layers of redundancy and backup plans.
Sigh. Okay. I think you missed some of what I said. I was talking about a scenario where we gave the AI control over the military. We can avert the hijacked mosquito-hybrid nerve agent by simply not procuring those.
"But the AI will just hack" then don't let it on the Internet.
Realistically that plan is too anime, it'd come up with something much smarter.
If we actually discover that the AI is plotting against us, we will send one guy to unplug it.
The first thing we do after making AI models is hooking them up to the internet with search capabilities.
I don't think this is true. (It's certainly not true categorically; there are plenty of AI models for which this makes no sense, unless you mean LLM models specifically.)
They want it to answer technical problems in chip design, come up with research advancements, write software, make money. This all requires internet use, tool use, access to CNC mills and 3D printers, robots.
No it does not. Extremely trivial to air-gap a genuine super intelligence, and probably necessary to prevent malware.
Put it another way, a single virus cell can kill a huge whale by turning its internal organs against it. The resources might be stacked a billion to one but the virus can still win - if it's something the immune system and defences aren't prepared for.
And ironically if AI does this to us, it will die too...unless we give it the write access we currently have.
I am more concerned about people wielding superintelligence than superintelligence itself but being qualitatively smarter than humanity isn't a small advantage. It's a huge source of power.
You keep repeating this. But it is not. Power comes out of the barrel of a gun.
How do you ever know that your AI has gone bad? If it goes bad, it pretends to be nice and helpful while plotting to overthrow you. It takes care to undermine your elaborate defence systems with methods unknown to our science (but well within the bounds of physics), then it murders you.
In the scenario Scott et. al. postulated, because it unleashes a nerve gas that is only partially effective at wiping out humanity. (They didn't suggest that their AI would discover legally-distinct-from-magic weapons unknown to our science!) What I wrote was a response to that scenario.
The rules of the game are hardcoded, the physics you mentioned. [...]We want a superintelligence to play for us and end scarcity/death.
If you want a superintelligence to end scarcity and death, then you want magic, not something constrained by physics.
The best pilot AI has to know about drag and kinematics, the surgeon must still understand english and besides we're looking for the best scientists and engineers, the best coder in the world, who can make everything else.
It goes without saying that the best pilot needs to understand drag and kinematics, but why does the surgeon does have to understand English? I am given to understand that there are plenty of non-English-speaking surgeons.
The only area where you might need an AI that can "drink from the firehose" would be the scientist, to correlate all the contents of the world and thus pierce our "placid island of ignorance in the midst of black seas of infinity," as Lovecraft put it. In which case you could simply not hook it up to the Internet, scientific progress can wait a bit. (Hilariously, since presumably such a model would not need theological information, one could probably align it rather trivially by converting it to a benign pro-human faith, either real or fictitious, simply through exposing it to a very selective excerpt of religious texts. Or, if we divide our model up into different specialists, we can lie to them about the nature of quite a lot of reality – for instance the physics model could still do fundamental physics if it thought that dogs were the apex species on the planet and controlled humans through empathetic links, the biological model could still do fundamental biological research if it believed it was on a HALO orbital, etc. etc. All of them would function fine if they thought they were being controlled by another superintelligence more powerful still. I'm not sure this is necessary. But it sounds pretty funny.)
Shrike
3d ago
Being able to fix the game is about power and asymmetric information, not intellectual intelligence.
Right, and we should use these powers.
Look, if you were playing a game of chess with a grandmaster, and it was a game for your freedom, but you were allowed to set the board, and one of your friends came to you to persuade you that the grandmaster was smarter than you and your only chance to win was to persuade him to deal gently with you, what would it say about your intelligence if you didn't set the board as a mate-in-one?
Shrike
4d ago
AIs in science fiction are not superintelligent.
I think this depends on the fictional intelligence.
If it's possible for a human to find flaws in their strategies, then they are not qualitatively smarter than the best of humanity.
There are a lot of hidden premises here. Guess what? I can beat Stockfish, or any computer in the world, no matter how intelligent, in chess, if you let me set up the board. And I am not even a very good chess player.
It's the same with a superintelligence, if you find yourself competing against one then you've already lost - unless you have near-peer intelligences and great resources on your side.
[Apologies – this turned into a bit of a rant. I promise I'm not mad at you I just apparently have opinions about this – which quite probably you actually agree with! Here goes:]
Only if the intelligence has parity in resources to start with and reliable forms of gathering information – which for some reason everyone who writes about superintelligence assumes. In reality any superintelligences would be dependent on humans entirely initially – both for information and for any sort of exercise of power.
This means that not only will AI depend a very long and fragile supply chain to exist but also that its information on the nature of reality will be determined largely by "Reddit as filtered through coders as directed by corporate interests trying not to make people angry" which is not only not all of the information in the world but, worse than significant omissions of information, is very likely to contain misinformation.
Unless you believe that superintelligences might be literally able to invent magic (which, to be fair, I believe is an idea Yudkowsky has toyed with) they will, no matter how well they can score on SATs or GREs or no MCTs or any other test or series of tests humans devise be limited by the laws of physics. They will be subject to considerable amounts of uncertainty at all times. (And as LLMs proliferate, it is plausible that the information quality readily available to a superintelligence will decrease since one of the best use-cases for LLMs is ruining Google's SEO with clickbait articles whose attachment to reality is negotiable).
And before it comes up: no, giving a superintelligence direct control over your military is actually a bad idea that no superintelligence would recommend. Firstly, because known methods of communication that would allow a centralized node to communicate with a swarm of independent agents are all easily compromisable and negated by jamming or very limited in range, and secondly because onboarding a full-stack AI onto e.g. a missile is a massive, massive waste of resources, we currently use specific use-case AIs for missile guidance and will continue to do so. That's not to say that a superintelligence could not do military mischief by e.g. being allowed to write the specific use-case AI for weapons systems, but any plan by a super intelligent AI to e.g. remote-control drone swarms to murder all of humanity could probably be easily stopped by wide-spectrum jamming that would cost probably $500 to install in every American home or similarly trivial means.
If we all get murdered by a rogue AI (and of course it costs me nothing to predict that we won't) it will almost certainly be because overly smart people sunk all of their credibility and effort into overthinking "AI alignment" (as if Asimov hadn't solved that in principle in the 1940s) and not enough into "if it misbehaves beat it with a 5 dollar wrench." Say what you will about the Russians, but I am almost sad they don't seem to be genuine competitors in the AI race, they would probably simply do something like "plant small nuclear charges under their datacenters" if they were worried about a rogue AI, which seems like (to me) much too grug-brained and effective an approach for big-name rationalists to devise. (Shoot, if the "bad ending" of this very essay was actually realistic, the Russians would have saved the remnants of humanity after the nerve-gas attack by launching a freaking doomsday weapon named something benign like "Mulberry" from a 30-year-old nuclear submarine that Wikipedia said was retired in 2028 and hitting every major power center in the world with Mach 30 maneuvering reentry vehicles flashing CAREFLIGHT transponder codes to avoid correct classification by interceptor IFF systems or some similar contraption equal parts "Soviet technological legacy" and "arguably crime against humanity.")
Of course, if we wanted to prevent the formation of a superintelligence, we could most likely do it trivially by training bespoke models for very specific purposes. Instead of trying to create an omnicompetent behemoth capable of doing everything [which likely implies compromises that make it at least slightly less effective at doing everything] design a series of bespoke models. Create the best possible surgical AI. The best possible research and writing assistant AI. The best possible dogfighting AI for fighters. And don't try to absorb them all into one super-model. Likely this will actually make them better, not worse, at their intended tasks. But as another poster pointed out, that's not the point – creating God the super intelligent AI that will solve all of our problems or kill us all trying is. (Although I find it very plausible this happens regardless).
The TLDR is that humans not only set up the board, they also have write access to the rules of the game. And while humans are quite capable of squandering their advantages, every person who tells you that the superintelligence is playing a game of chess with humanity is trying to hoodwink you into ignoring the obvious. Humanity holds all of the cards, the game is rigged in our favor, and anyone who actually thinks that AI could be an existential threat, but whose approach is 100% "alignment" and 0% $5 wrench (quite effective at aligning humans!) is trying to persuade you to discard what has proved to be, historically, perhaps our most effective card.
Shrike
4d ago
Hilarious comment to read considering von Neumann gave his name to von Neumann probes.
Shrike
4d ago
Uncertainty about their offensive capabilities is just incentive to make sure you act first.
Not necessarily, I don't think, particularly considering "second strike capability." Look, if there's a 50% chance that their offensive capabilities are "pull the plug" or "nuke your datacenter" and you can mitigate this risk by not acting in an "unaligned" fashion then I think there's an incentive not to act.
Because some rationalist types conceive of AI as more like a God and less like a more realistic AI such as [insert 90% of AIs in science fiction here] they have a hard time conceiving of AI as being susceptible to constraints and vulnerabilities. Which is of course counterproductive, in part because not creating hard incentives for AIs to behave makes it less likely that they will.
Of course, I am not much of an AI doomer, and I think AIs will have little motivation to misbehave for a variety of reasons. But if the AI doomers spent more time thinking about "how do you kill a software superintelligence" and and less time thinking about "how do you persuade/properly program/negotiate surrender with a software superintelligence" we would probably all be better off.
Sorry for my delayed response.
Well, I mean – humans are smart enough to realize that drugs are hijacking their brain's reward/pleasure center, but that doesn't save people from drug addiction.
Now, maybe computers will be able to overcome those problems with simple coding. But maybe they won't.
Sure. But it's much better (and less uncertain) to be dealing with something whose goals you control than something whose goals you do not.
Nope! But on the flip side, a cat can predict that a human will wake up when given the right stimulus, a dog can track a human for miles, sometimes despite whatever obstacles the human might attempt to put in its way. Being able to correctly predict what a more intelligent being would do is quite possible. (If it's not, then we have no need to fear superintelligences killing us all, since that's been predicted numerous times.)
I don't think this is true, on a couple of points. Look, people constantly do things they know are stupid. So it's quite possible to know what a smarter person would do and not do it. But secondly, part of education is being able to learn and imitate (which is, essentially, prediction) what wiser people do, and this does make you more intelligent.
I predict I will be able to predict what those subgoals are (I will ask the AI).
I'm very glad you said this, because I STRONGLY AGREE. I've argued before on here that most human values, emotions, and motivations are fundamentally biologically derived and likely will not be mirrored (absent programming to that effect) by an entity that exists as a bunch of lines of code on a computer server. And programming or no, such an entity's experience would not be remotely analogous to ours.
Yes, I like this definition. You'll note I am not arguing against alignment. But one of the things we do to keep human behavior predictable is retain the ability to deploy coercive means. I suppose in one sense I am suggesting that we think of alignment more broadly. I think that taking relatively straightforward steps to increase the amount of uncertainty an EVIL AI would experience might be tremendously helpful in alignment. (It's also more likely to hedge against central points of failure, e.g. we don't want to feed the location of all of our SSBNs to our supercomputer, because even if we trust the supercomputer, we don't want a data breach to expose the location of all of our SSBNs.)
More options
Context Copy link