Richard Hollerith, 15 miles north of San Francisco. hruvulum@gmail.com
rhollerith_dot_com
Karma: 2,202
I’m unlikely to go back and read all of it
OK, good to know; let us know if your probability changes :)
OK, well, here is where I got the information in my previous comment—somewhere in the first 47 minutes of this.
I assume that Lukashenka is probably happier being a sovereign dictator than mere governor.
I heard that Lukashenka lobbied for a merger of Russia and Belarus because he liked his chances of eventually becoming leader of the combined entity.
During the Cold War, no one believed the economic figures published by the Soviet Union. I am puzzled as to why the figures published by the Chinese Communist party are often taken at face value now. Yes, I concede that the average Chinese resident really is wealthier than the average resident of India. But I don’t necessarily believe the GDP figures that say he or she is wealthier than the average Mexican, just to pick a country for the sake of illustration that according to the World Bank has a GDP per capita a little lower than China’s. And I’m much less certain than “the internet” that the average Chinese person will ever be wealthier than the average Mexican since Mexico is much less likely to descend into political chaos than China is and since Mexico has significant petroleum reserves and access to cheap Texan natural gas.
The main argument for the potential of China IMO is the staggering number of people with IQs over 130, many of whom have degrees in science or engineering. But whereas scientific and engineering knowledge transfer fairly easily to any sufficiently bright population anywhere in the world, the ability to organize a society such that the scientists and engineers can contribute to societal goals is not so easy, and Russia is a good example of this because although Russia certainly can produce great scientists (just restricting ourselves to the 20th Century, we find Solomonoff, Kolmogorov and Grigori Perelman) the society is lacking in many ways (e.g., the quality of life of the average person is low; e.g., even if you’re wealthy, the chances of your being treated unjustly by the government or the justice system is much higher than it is in many countries; e.g., Russia does not contribute to the flourishing of the world anywhere near as much as the West does).
Again, I am not making a prediction about how powerful China will be 10 or 20 years from now; I’m merely saying that I’m much less certain that it will continue to grow in capability than the average internet commentator is.
China has put an extraordinary amount of resources into developing their own chip fabrication capacities, and so far has failed pretty pathetically and now IMO it is probably too late: China will more likely than not never acquire the capability to design and fab GPUs and TPUs as efficient as those currently being designed by Nvidia and Google respectively and currently being fabbed by TSMC and whoever is fabbing Google’s TPUs.
The ability to design and fab such chips is currently a “world ability”: the fabbing relies on litho machines made by in the Netherlands, which in turn depends heavily on lenses made by Zeiss in Germany and other advanced products from Japan, the US and probably other countries. The fabbing relies on very expensive other products that like the litho machines are made by just a handful of companies, almost all of which are vulnerable to pressure from the US government (especially if other governments generally aligned with the US tend to agree with the USGov’s reasons). Designing the chips (the layouts) requires software available from only a handful of suppliers again probably all vulnerable to pressure from the USGov. It is currently a world ability and more likely than not will remain a world ability and not an ability of any one bloc (e.g., the Chinese government and its friends) unless we consider the US and Europe and entities vulnerable to pressure from those 2 as one bloc.
Note the “more likely than not” above! I’m not saying that I can reliably predict the relative strength of China in the future! It’s just that I am not moved by the numerous pronouncements that the rise of China is inevitable. To take just one example, Peter Zeihan says that if China were subjected to the same sanctions imposed on Russia after Feb 24 of this year, there would be rolling blackouts within a few months, trucks running out of gas and abandoned on its highways and in about a year half of the population will have died of starvation: in contrast to Russia, which produces more fossil fuels, fertilizer and basic foodstuffs like wheat than it needs, China is heavily dependent on imports for those things.
Zeihan goes on to say that although China has always been aware of the possibility that US and Europe governments would impose sanctions on it, it was utterly shocked to learn of the possibility that major corporations would in the future refuse to do business with China without even being forced to by a government. He says he has heard from sources inside China that such an eventuality was never considered by Chinese planners and now they feel they must re-evaluate their entire geopolitical strategy.
There is a danger when spending a lot of time reading comments on the internet—or when reading the output of our newspapers and similar institutions—to allow the sheer repetition of a claim thousands of times from thousands of individual authors to give the claim more credence than it deserves. The vast majority of the claims I have seen as to the probable eclipsing of the US by China either makes no argument in support of the claim or the included argument is very unimpressive. Yes, there are experts I respect, like John Mearsheimer, who make the claim (though I haven’t heard him make an argument for it: he just claims it, but he is Mearsheimer, so I listen), but there are other geopolitical experts I respect, like George Friedman, who believe that China will be lucky if 15 years from now it hasn’t split into 2 or 3 countries—and that the Chinese government knows that and preventing it is their main focus. Friedman also says that the Chinese navy is still not capable enough to protect the ships that bring it oil from the Persian Gulf from pirates and privateers if the US navy stepped down from its role of protecting the world’s shipping from pirates. “Privateers”: if there ever is a really serious oil shortage, the Indian government might follow a centuries-old tradition of defining the takeover of oil tankers bound to China as “piracy”, but rather “privateering”, and consequently not illegal and in fact maybe it is rewarded, and according to Friedman, there is nothing the Chinese navy can do about that: they simply lack the ability to project force that far from port. Friedman also says that the Japanese navy would more likely than not prevail over the Chinese navy if they fought today. He says that that is partly because Japan’s admirals were trained by admirals who were trained by admirals who actually fought a major war with what remains the dominant naval technology relevant to a major war, namely, the aircraft carrier—and according to him, that that counts for a lot.
The reason I’m commenting is that this has relevance to managing the risk of AI: in particular, in a situation as dire as the current global situation around AI research, we should be paying attention to those futures in which our ability to control and manage the global situation around AI has drastically increased even if right now those futures seem unlikely. In particular, there is a chance that the leaders of China will in the future do something at least as risky and ill-fated as Putin’s decision to invade Ukraine, and the response to that by the West will make it so that we don’t have to worry about AI researchers in China because China will be too preoccupied with feeding itself and just keeping the country supplied with electricity and whatever China uses to heat its homes. Or maybe Southern China or Shanghai will make a serious attempt to secede from Beijing’s rule, in response to which Beijing will lock the country down kinda like they did for Covid, with the same effects on the Chinese economy that the Covid lock-downs had, but these lock-downs will go on for years, and again then we don’t have to worry about the danger posed by Chinese AI researchers.
So if young people concerned about AI risk move to Washington and London and Paris and start a career in the national-security bureaucracy there with the goal of eventually convincing their government to impose a regime of “compute governance” similar to the existing regime of nuclear non-proliferation, then maybe their efforts will be in vain because the center of mass of irresponsible AI research (currently somewhere in the Atlantic between the US and Britain) will just move to China. But maybe their efforts won’t be in vain because China will stop being able to host the intensity of scientific research and technological development necessary to contribute significantly to the AI danger. And maybe Russia will continue its decline with the result that regardless of the wishes and the ambitions of its government, it becomes unable to contribute significantly to the AI danger (which of course its government does not perceive as particularly dangerous).
Huh. I always thought that if we were magically/hypothetically given the option to limit the death and destruction caused by the next drawn-out competition among great powers to that caused by the Cold War (which includes of course the Korean War, the Vietnam War and the Soviet occupation of Afghanistan) we should jump at the chance to choose that option.
I always thought that the Cold War turned out remarkably benignly given the weapons available to the warring sides and given previous conflicts between great powers.
Do I read you correctly as believing that humanity has entered a more enlightened era where outcomes as bad as what happened to people who got caught up in the Cold War are unlikely?
Hitler probably liked his wife, kids and dog, but when it came to the end game we all know how that went.
Hitler never wanted kids and most historians say he didn’t have any although there are claims he had a son out of wedlock. (The claimed son survived WWII as did Hitler’s sister.) He did marry once -- 40 hours before committing suicide at the end of WWII.
Stalin married and had a kid, then his wife died of TB or typhus, then he married someone else and had more kids. He had control over nukes during the final 4 years of his rule, but never used them.
40 miles away, the only way the explosions can badly hurt you is via fallout. It is relatively easy for informed people with, say, 72 hours of warning of the attack to protect themselves from fallout. They will be spending most of those 72 hours digging a trench, topping the trench with plywood or felled trees or such, heaping about 18 inches (maybe make it 24 inches for people in your situation) of dirt on top of that wood, then putting a sheet of plastic on top of that so that rain cannot wash the fallout into the trench.
If the fallout is very heavy, you have to stay in the trench/tunnel for 3 weeks. You can venture out for an hour after week one to scrounge for supplies if you want.
You can make the trench/tunnel in the yard of your home if you have a yard. In contrast, the only way for residents of, e.g., Berkeley, Manhattan or San Francisco to protect themselves that doesn’t involve hiring an expert and probably spending 100s of 1000s of dollars and many weeks of warning of the attack is to move themselves many miles away from their home before the start of the attack.
The really ugly aspect of the situation is that if you’ve prepared by digging yourself a fallout shelter and stocking it with food and water, you have to deal somehow with desperate unprepared people—great numbers of them if you’re in the wrong place. That is why, e.g., Switzerland made sure that everyone would be prepared, by having communal fallout shelters in its cities and requiring home owners outside the city to maintain fallout shelters in their homes, but even Switzerland stopping doing that after the end of the Cold War.
My probability of Putin’s going “fuck it” is about 17 times lower than it would be if Putin didn’t have living children and grandchildren.
The part of the US government that decides the details of US economic sanctions believes that Putin’s relationships with his children is warm enough that substantial parts of his wealth might be held in their names.
I had trouble with the quotes (from previous centuries): I read the first 5 or so of them, and never became confident I knew what any of them meant. It is possible (but not likely) that my ability to read seventeenth-century english is no better than my ability to read French or German. (I’m a native English speaker. I would guess that I’ve spent at least as much time learning to read English from at least 300 years ago as I’ve spent learning to read French, i.e., not much time.) Because the quotes are a large fraction of the blog post, I felt I needed to give up on it even though I was very eager to learn about the topic.
That was approximately 24 hours ago (when I found the blog post on Hacker News). Just now as part of my preparation of this comment, I notice that the final 2 paragraphs of the post contain no quotes from previous centuries, and of course I had no trouble understanding them. My suggestion for improving the article is to move those 2 final paragraphs up to before the start of the quotes, so that people like me who give up on the article because of the old-English quotes see them.
It is unclear to me whether I’m just unusually bad at reading old English for a native English speaker or whether you are overestimating your audience’s ability to read old English.
In an attack with a single nuke, the BART tunnels seem like a pretty good place to go.
In an attack by Russia or China (or Russia and China), it’s not—unless the people in the tunnels are low in number and have some way to prevent additional people from entering.
In normal times, the tunnels are “actively ventilated”, meaning fans are used to bring air in from the outside, but an attack by Russia or China would surely prevent electricity from getting to the fans (because destroying the US’s generating and electrical-distribution capacity would be one of the goals of the attack) so CO2 and heat would build up in the tunnels if there are people in the tunnels. Of the 2 dangers, I’m more afraid of the heat (though if the tunnel were in Fairbanks, AK, I’d be more afraid of CO2): the ground is a very poor conductor of heat. Either danger seems likely to kill most of the people in the tunnels before it becomes safe enough to leave the tunnels (2 or 3 weeks).
I think that the subway systems of Moscow and St Petersburg have ventilation systems designed to keep operating after a nuclear attack (because the Russian government takes nuclear war more seriously than governments in the US do) although I welcome more information on that point: I’m particularly interested in details such as whether the air intakes for those 2 subway systems point down (which is the easiest way to prevent fallout from falling into the air intakes and thereby landing close enough to sheltering people to cause their deaths).
I spent an hour or 2 recently Googling for any signs that BART’s ventilation system has been designed to keep operating if the electrical grid goes down and I didn’t find anything. I know from having seen some of them that BART’s air intakes weren’t designed to prevent fallout from falling down into them. (I am focusing on air intakes because designing them right strike me as a particularly cheap intervention relative to how much grief they can prevent in case of an attack.)
The tunnels through the tops of the Berkeley and Oakland hills (for roads and maybe train tracks) are much more survivable on my models than the BART tunnels because there is some wind or breeze most of the time in those hills. In general, even when there is no breeze in the flatlands, there is usually a breeze at elevation in the Bay Area.
If I saw credible signs that a major nuclear attack was coming or if an attack had occurred leaving me uninjured enough to walk, I’d head to the Cal Park Tunnel (in Marin County, about a mile from my home) with as much water and food as I can carry. (I’d carry about 93% of the weight as water, 7% as food.) It is 1100 feet long according to some web site (which seems about right to me). I’ve been in the tunnel often; I notice that even when the breeze outside is barely perceptible, there’s definitely plenty movement of air inside (in a constant direction at a constant velocity, which of course is what you want for ventilation).
Now some of you are asking: won’t the breeze or wind blow the fallout into the tunnel? Good question! Most of the fallout mass is in the form of “aggregations” between the size of a pea and a grain of sand. Basically what it does is fall out of the sky, then remain on whatever horizontal surface it lands on. It’s not that the fallout that is in the form of particles small enough to remain suspended in the air for a while is completely harmless—if during the hour or 3 during which fallout is falling out of the sky, you can seal off the tunnel or whatever other shelter you are in, do so; failing that, if you have a mask, wear it—but if I recall correctly my reading of the book Nuclear War Survival Skills, the majority of the danger from the radioactive material that remains suspended in the air or is blown around by winds comes from some process in which beta radiation affects the mucous membranes, particularly in the nose, and is only a significant danger in dry weather whereas of course the air in the Bay Area is moist most of the time.
I [have] the right to think whatever thoughts I [want], including arbitrarily inappropriate sexual thoughts
Although I agree with that, my experience has been that if I fantasize a lot about a woman (i.e., at least one “fantasizing session” ending in intense pleasure), it becomes significantly less likely that she will response positively to me.
Having had the fantasies changes my behavior in some way that the woman tends to notice (although I myself do not notice).
It took me many, many years to notice this cause-and-effect relationship. (Why so long? Maybe because pleasure is the motivation in motivated cognition.) But by now I realize that (at least for me) the effect is strong enough that I have basically adopted a rule of refraining from fantasizing a lot about a woman unless I know for sure I will neither want nor need to interact with her in the future. “Interact”: regularly pass her in a hallway, talk to her in any capacity.
I adopted this rule because I need all the help I can get in being liked or at least consistently tolerated by the women in my environment, not because I recognize any moral imperative to refrain from fantasizing about anybody.
I am a straight man in his 60s.
IMPORTANT ADDENDUM.
The OP writes of sexual thoughts that last less than 5 seconds. That is definitely not at all what I refer to above as “fantasizing a lot”.
I think most adult women realize that most men run a quick sexual evaluation of every single woman they interact with (and that this quick evaluation often involves imagining what it would be like to have sex with her). My guess is that a lot of women are perceptive enough that they can detect this evaluation as it is happening, but are forgiving of it—or at least it does not worry them or annoy them—as long as it does not go on and on. 5 seconds is fine if my life experience is any guide.
Also, my guess is that most men would not be able to stop running this evaluation during those initial 5 seconds or so without employing measures that are very expensive for the man in time, attention or opportunity costs.
Before it gains the ability to transform itself in the ways you speculate about, humanity will probably perish, killed by some AGI research project, so it is a little inefficient to worry about transhumans at this point.
I don’t need anything to encourage me to use Lesswrong, but if you have something to discourage me from using it, although I don’t have a need for it now, I am worried enough about the future that I would keep a link to it on my hard drive.
A false statement can cause a reasoner’s beliefs to become more accurate.
Suppose for example that Alice believes falsely that there is an invisible dragon in her garage, but then Bob tells her falsely that all dragons, invisible or not, cannot tolerate the smell of motor oil. Alice decides to believe that, notes that there is a big puddle of motor oil in the center of her garage (because her car leaks oil) and stops believing there is an invisible dragon in her garage.
But by your definition of deception, what Bob told Alice just now is not deceptive because it made Alice’s beliefs more accurate, which is all that matters by your definition.
It would be reasonable for Alice to want Bob never to lie to her even when the lie would make her beliefs more accurate, but there is no way for Alice to specify that desire with your formalism. And no way to for Alice to specify the opposite desire, namely, the fact that a lie would be okay with her as long as it makes her beliefs more accurate. And I cannot see a way to improve your definition to allow her to specify that desire.
In summary, although there might be some application, some special circumstance that you did not describe and that I have been unable to imagine, in which it suffices, your definition does not capture all the nuances of deception in human affairs, and I cannot see a way to make it do so without starting over.
But that is not surprising because formalizing things that matter to humans is really hard. Mathematics progresses mainly by focusing on things that are easy to formalize and resigning itself to having only the most tenuous connection to most of the things humans care about.
My guess is that we don’t have passenger or cargo VTOL airplanes because they would use more energy than the airplanes we use now.
It can be worth the extra energy cost in warplanes since it allows the warplanes to operate from ships smaller than the US’s supercarriers and to keep on operating despite the common military tactic of destroying the enemy’s runways.
Why do I guess that VTOLs would use more energy? (1) Because hovering expends energy at a higher rate than normal flying. (2) Because the thrust-to-weight ratio of a modern airliner is about .25 and of course to hover you need to get that above 1, which means more powerful gas-turbine engines, which means heavier gas-turbine engines, which means the plane gets heavier and consequently less energy efficient.
I do agree that there is a slight ‘creepiness vibe’ with the way the Sequences . . . are written.
IIRC back when he was writing the sequences, Eliezer said that he was psychologically incapable of writing in the typical dry academic manner. I.e., he wouldn’t be able to bring himself to do so even if he knew doing so would improve the reception of his writing.
Maybe he used the word “stuffy”.
The reason Eliezer’s 2004 “coherent extrapolated volition” (CEV) proposal is immune to Goodharting is probably because being immune to it was probably one of the main criteria for its creation. I.e., Eliezer came up with it through a process of looking for a design immune to Goodharting. It may very well be that all other published proposals for aligning super-intelligent AI are vulnerable to Goodharting.
Goodhart’s law basically says that if we put too much optimization pressure on criterion X, then as a side effect, the optimization process drives criteria Y and Z, which we also care about, higher or lower than we consider reasonable. But that doesn’t apply when criterion X is “everything we value” or “the reflective equilibrium of everything we value”.
The problem of course being that although the CEV plan is probably within human capabilities to implement (and IMHO Scott Garrabrant’s work is probably a step forward) unaligned AI is probably significantly easier to implement, so will likely arrive first.
Interesting. I am also curious what size of screen do you imagine (plan on) this keyboard being on? One of smartphone size?
I am surprised at how much text you are willing to enter on one’s on-screen keyboard if indeed you are planning to use this app on a smartphone. (I don’t own a smartphone, but am curious about them.)
If the alien civilization is no more advanced than we are, then it probably cannot send us a message that would have a drastic effect on us.
But if the aliens can create powerful AI, they can just send that.
Specifically, the alien message might start with the first 50 primes, followed by a computer program that finds prime numbers—written in some simple programming language. That is followed by a few more simple programs in the same programming language. Then comes the powerful AI (as a computer program in the same simple programming language). There are 100s of 1000s of people with the skills to use the examples in the first part of the message to get the AI in the second part of the message running on a computer here on Earth. 12 hours after such an alien message is published worldwide here on Earth, I expect that dozens of people would have the powerful AI running at good efficiency on their PCs. Obviously, I would prefer for SETI to retract their policy of publishing any alien message they might receive.
One simple program that is particularly illuminating for specifying a programming language is a simple interpreter for the same language: educational materials often call such a metacircular evaluator. Here is one example, but there are many. That example of course contains a lot of English prose intended to help the student understand the code. But there are 100s of 1000s of people who don’t need the English prose: the code is probably all information they would need to get any program written in the same programming language running at good efficiency on their PC. The only reason I need the qualifier “probably” in the previous sentence is that there is a chance that a single program won’t be enough: one or 2 more simple “sample” programs, e.g., for generating the Fibonacci sequence, might be required.
Keep in mind that the aliens are trying to make it as easy as possible for as many recipients as possible to run their code performantly.
Can the humans with the most skill in computing performantly run a complex computer program if all the information they have to go on is a few simple “sample programs” written in the same never-seen-before-by-humans programming language? -- that is the central issue. The answer is, yes, they can. Many of them can. And some of those many will be reckless enough to run the alien AI.