Froolow

[Question] AI Risk Micro­dy­nam­ics Survey

• Froolow 19 Feb 2015 10:54 UTC

  25 points

  0

  on: An alarm­ing fact about the anti-ag­ing community

  Here is a rough cost-benefit analysis. I found all the numbers before doing any maths and tried to be as optimistic as possible towards your theory to account for people more intelligent than me coming up with better ways to do it if it became standardised.

  The cost of freezing a cell is proxied as the cost of freezing your eggs, which is already commercially available. It is £2900 for the initial harvest and £275 for a subsequent year of freezing. This £275 is not discounted because you pay it every year. Source. Let’s assume that the clinic are nice and let you freeze as many different organ-cells as you want for the same £275 /​ year fee.

  There are 56,000 people in the EU who are on the organ donor waiting list, suggesting there are around 56,000 people who have a sufficiently chronic failure of an organ that it didn’t kill them outright but will kill them if they don’t get a transplant. Many of these 56,000 people will get an organ through conventional means, but let’s say none of them do to account for the fact that as medicine advances we will probably be able to transfer more people off the ‘going to die’ list onto the ‘might be able to survive if given a transplant’ list. Let’s also assume that this represents 56,000 new people on the organ transplant list each year, even though the average wait on the list is around three years. Source (also I found a dodgy-looking source suggesting that 50,000 people die of organ failure each year, so the figure for the number who could benefit is probably not out by more than an order of magnitude). There are 500m people in the EU, so your chance of being on the organ transplant list in any given year is about 0.01%.

  Same source as above suggests the average QALY gain for an organ transplant is 11.5 for a liver transplant, 6.8 for a heart transplant and 5.2 for a lung transplant. Let’s assume all transplants give you the full 11.5 QALYs because medicine improves.

  Finally let’s assume you freeze your cells now at 25 with the expectation that they will be used in 40 years at 65. You only get one shot at doing this; you may never freeze your cells again because they degrade too much on your 26th birthday.

  This means in total you pay £13900 for a 0.01% chance at 11.5 additional QALYs, for an expectated value of £10.8m /​ QALY. That is to say, if you would pay £10.8m for an additional year of life at the margin, this is probably worthwhile (given some very optimistic assumptions). However, this is only true if you get one shot to freeze your cells at 25. If instead you can wait until you need them and freeze them then, you’d only be paying something like £250 /​ QALY. You can see from these two numbers that even if cells degrade in quality to the extent that they are a thousand times harder to successfully transplant you are still better to wait until you are pretty sure you are a high-risk group for organ failure. Anyone who believes £10.8m /​ QALY is a good deal should also be prepared to accept a salary cut of around £115,000 /​ year in order to avoid a 20 mile commute since road vehicles have a fatality rate of something like 1.5 per billion miles travelled.

• Froolow 3 Nov 2014 12:40 UTC

  0 points

  on: Maybe you want to max­imise pa­per­clips too

  I really enjoyed the article, but I think your argument falls down in the following way:

  1) Fission /​ fusion are the best energy sources we know of, but we can’t yet do it for all forms of matter

  2) A sufficiently clever and motivated intelligence probably could do it for all forms of matter, because it looks to be thermodynamically possible

  3) (Implicit premise) In between now and the creation of a galaxy-hopping superintelligence with the physical nouse to fusion /​ fission at least the majority of matter in its path, there will be no more efficient forms of energy discovered

  4) Therefore paperclips (or at least something that looks enough like paperclips that we needn’t argue)

  Premise 1 is trivialy true, premise 2 has just enough wild speculation to make it plausible but still exciting, and the conclusion is supported if premise 3 is true. But premise 3 looks pretty shakey to me—we can already extract energy from the quantum foam and can at least theoretically extract energy from matter-antimatter collision (although I don’t know if thermodynamics permits either of these methods to be more efficient than fusion). It is a bold judgement to suppose we are at the limits of our understanding of these processes, and bolder still to assume there are no further processes to discover.

• Froolow 3 Nov 2014 12:26 UTC

  8 points

  in reply to: ChristianKl’s comment on: What are the most com­mon and im­por­tant trade-offs that de­ci­sion mak­ers face?

  They are slightly different, but in practical terms they describe the same error; sensitivity and specificity are properties of a test while Type I and II errors are properties of a system, but both errors are basically saying, “Our test is not perfectly accurate so if we want to catch more people with a disease we need to misdiagnose more people”

  To illustrate the distinction, consider a test which is 90% sensitive and 90% specific in a population of 100 where a disease has a 50% prevelance. This means 50 people have the disease, of which the test will identify 45 as having the problem (90% sensitive). 50 people are free of the disease, of which the test will correctly identify 45 (90% specific). So if diagnosed your probability of the diagnosis being a Type I error is ⁵⁄₅₀ = 10% (if given the all clear the same logic applies for a Type II error). You derive this from the number of people in the population who were told they have a disease who were incorrectly diagnosed divided by the total population who were told they have a disease (rightly or wrongly)

  But if the disease prevelance changes due to demographic pressue to 10% then 10 people have the disease of whom 9 are diagnosed, and 90 people are disease-free of whom 81 are given the all-clear. This means the probabilities of the different ‘Type’ errors change dramatically; now ⁹⁄₁₈ = 50% for a Type I error and ¹⁄₈₂ ~ 1.2% for a Type II error. But the sensitivity and specificity of the test are completely unchanged.

• Froolow 21 Oct 2014 18:57 UTC

  0 points

  in reply to: Lumifer’s comment on: One Year of Goodsearching

  I agree with everything you’ve said, but I would point out that I already allow myself to be tracked by Google, so the true cost is only the difference between the ‘badness’ of Google and Microsoft.

• Froolow 21 Oct 2014 18:54 UTC

  0 points

  in reply to: sixes_and_sevens’s comment on: One Year of Goodsearching

  Don’t worry about the tone, opportunity cost is that hinterland where it is too complicated to explain to someone who doesn’t get it in one sentence, but too fundamental not to need to talk about so it is very difficult to judge tone when you’re not sure whether you can assume familiarity with economic concepts.

  It sounds to me like we basically agree—the cost of switching search engine is ten minutes (assumption) and this pays off about 50 cents a day for forever (assumption). This makes cutting off the analysis at one year arbitrary, which I agree with. You also have to compare the effort you put into searching with anything else you could do with that time, (even if you would have been doing those searches ‘naturally’) for the purpose of correctly calculating opportunity cost.

  I think we disagree on the final step—if this is to be ineffective you need to be able to find an activity which is a better use of my time than conducting those daily searches. Since my primary contribution to charitable causes is from my salary, and I use a lot of Google in my job (I would be fired if I didn’t do internet searches because I would be totally ineffective) I can’t think what else I should be doing—what is a better use of my time than doing those searches? Assume we’re only interested in maximising my total charitable giving.

• Froolow 21 Oct 2014 15:50 UTC

  1 point

  in reply to: sixes_and_sevens’s comment on: One Year of Goodsearching

  Why not? Genuine question, because my job is pretty much nothing but cost-benefit analysis of fixed-cost projects which pay off a small amount every year, and the calculation you describe as ‘not how you do cost-benefit analysis’ is almost exactly how I would do a cost-benefit analysis of this kind (although I wouldn’t phrase it as $1200 /​ hour because that’s clunky, I’d probably talk about a percentage return on investment). I rate the probability that I am wrong here as extremely small, but if I am wrong I really need to hear it, and have the problem explained to me.

  If I were totally comitted to getting the most accurate answer I’d add a couple of complications that katydee doesn’t—for example;

  • I would discount against the possibility that GoodSearch no longer exists in 2040 (and discount future earnings more generally),

  • I would try to estimate a probability that I’d need to reconfigure my settings in the future, and estimate timings for that (if both Chrome and GoodSearch are still around in 30 years I’d be surprised!)

  But ignoring these complications for the moment I think it is totally accurate to say in 2040, “I have donated $16,000 to charity since 2014”. I think the $1200/​hour rhetoric is unhelpful, because it implies that you could earn another $1200 by working another hour, when in fact you don’t actually ever earn $1200 (well, I suppose you do after about a decade) and you can only earn the money over a very long period of time. I would probably describe it as a ‘Nominal Rate of Return of X% per annum over Y years’, compared to a nominal rate of return of almost zero percent per annum if I don’t search with GoodSearch (it is possible my advertising generates a multiplier effect which has a tiny positive externality on me, but I’d expect this to be almost totally negligable)

  Where is the error in my reasoning?

• Froolow 21 Oct 2014 13:24 UTC

  8 points

  in reply to: sixes_and_sevens’s comment on: One Year of Goodsearching

  I think there is a problem in how we are using the word ‘ineffective’. I think you are using the word to mean ‘very small absolute amounts’ whereas I am using the word to mean something like ‘low opportunity cost : return ratio’. I think looking at the cost : return ratio is fairer, and I also think ‘very small absolute amount’ is misleading.

  I did 46 searches on my work computer yesterday, and probably a handful more on mobile devices. Say 50 for the sake of argument, or $0.50. Over a year this is $130 dollars if I make no searches at weekends. I agree with katydee that it would probably take about ten minutes to configure my search settings, so I am better advised to spend ten minutes configuring my search settings than work a marginal ten minutes and donating the procedes so long as my salary is less than about $800 an hour. Based on the 2013 LessWrong Survey, around 1500 LessWrong users have a salary of less than $800 an hour, so if they all configured their internet settings in that way they would raise about $200,000 for a charity of their choice. I can’t find an official budget for MIRI, but I’d estimate that it is somewhere between $0.5m and $1m per year, so that’s a pretty meaningful amount.

  Another interpretation of your argument is that it is inefficient to do needless searches on GoodSearch to earn money for charity. This is a good argument; if it takes (say) five seconds to do a search then you are better off working a marginal hour as long as your salary is above $7.20 per hour. But I don’t think anyone is actually arguing that—the idea is just to monetise searches you would be making anyway, since recapturing a small (non-zero) fraction of your personal contribution to advertising is strictly better than recapturing a zero-size contribution to your personal contribution to advertising.

• Froolow 21 Oct 2014 12:06 UTC

  6 points

  in reply to: sixes_and_sevens’s comment on: One Year of Goodsearching

  I don’t understand why you think it is a grossly ineffective way to earn money for charity. Please could you explain why you think this is?

• Froolow 7 Aug 2014 10:57 UTC

  0 points

  in reply to: avairosa’s comment on: Gam­ing Democracy

  I think—though I’m not certain—that you are right that in all but the most marginal seats you’d never find a politically unafiliated /​ passionate about a political issue group large enough to swing a seat. I don’t think that implies that you shouldn’t try to game democracy though—there are certain known flaws in the democratic system we have which exist (and swing elections) independently of whether people knowingly exploit them or not.

• Froolow 7 Aug 2014 10:54 UTC

  2 points

  in reply to: Jiro’s comment on: Gam­ing Democracy

  I think that’s certainly an interesting idea—NHS-homeopathy could be even cheaper than what is currently provided (comissioning the services off a private homeopathy provider) because we could do it in bulk—the raw ingredients aren’t expensive at all. I’d worry about the indirect cost of moving the Overton Window though—at the moment we STRONGLY advise people not to use homeopathy even for trivial conditions, and we mock those that promote it. Even so, many people still use it and swear by its efficacy. If we moved to a situation where we promoted homeopathy for minor conditions and gave its practitioners the stamp of NHS/​Government approval, we would see many more people using it for minor conditions and—I would expect—some people begin to use it for major conditions. Thus the money we save on prescribing a placebo over an active drug might be sucked up by the cost of treating the complications of people who take homeopathic treatments to manage AF and then get a massive stroke, for example.

  But I think as a matter of principle we should set the level of homeopathy at whatever maximises the number of healthy life-years per unit of spending, even if that is not zero.

• Froolow 31 Jul 2014 18:08 UTC

  1 point

  in reply to: HopefullyCreative’s comment on: Gam­ing Democracy

  I certainly don’t disagree with your analysis, but I think I might not have been clear enough with the endgame of this potential strategy; I don’t think this is a good strategy to succeed as a minor party, because no matter how virtuous you make transhumanism sound, people are always going to care more about the economy or defence. But I think you can probably find enough people who care more about transhumanism than they do about the marginal difference between the economic policy of the two main parties. So the ‘transhuman’ party will never get off the ground, but it may have enough power to swing a marginal seat for one of the two main parties, in exchange for agreement to vote a certain way on a certain issue.

  Whether or not you could parley that into a successful minor party is a much harder question!

• Froolow 30 Jul 2014 17:21 UTC

  3 points

  in reply to: Lumifer’s comment on: Gam­ing Democracy

  I think both of you are incorrect. This leverages a specific flaw in the FPTP system which wouldn’t work in a PR system that gives a small, tightly coordinated group in a swing seat a disproportionate amount of power. Insofar as both political parties and lobby groups can exist in a PR system, this cannot be either of those things since it could not exist in a PR system.

  More specifically, it is not a political party because (amongst other things) it has no general platform and does not seek to acquire power. It is also not a lobby group because it doesn’t really ‘lobby’ in any meaningful sense to get the law changed. I think the example of the NRA is a red herring—it is hard to believe the NRA is well-enough coordinated to get a large number of its members to vote for a party they don’t like. Do you have any evidence they have ever been successful at swinging a seat in this way?

Gam­ing Democracy

• Froolow 4 May 2014 7:54 UTC

  0 points

  in reply to: christopherj’s comment on: The Ex­tended Liv­ing-For­ever Strat­egy-Space

  I don’t disagree I was grasping at straws for some of the more outlandish suggestions, but this was deliberate—to try and explore the full boundaries of the strategy space. So I take most of your criticism in the constructive spirit in which it was intended, but I do think maybe you are a bit confused about ‘philosophical preservation’ (no doubt I explained it very badly to avoid using the word ‘religion’). My point is not that you convince yourself, “I will live forever because all life is meaningless and hence death is the same as life”, it is that you find some philosophical argument that indicates a plausible strategy and then do that strategy. A simple example would be that you discover an argument which really genuinely proves Christianity offers salvation and then get baptised, or prove to your satisfaction that the soul is real and then pay a medium to continue contacting you after you die. Again, I agree this is outlandish but there must be something appealing about the approach because it is unquestionably the most popular strategy on the list in a worldwide sense.

• Froolow 4 May 2014 7:46 UTC

  0 points

  in reply to: Nornagest’s comment on: The Ex­tended Liv­ing-For­ever Strat­egy-Space

  I didn’t know that. Fair enough, seems likely ‘signal preservation’ is much more costly than I originally realised and not worth pursuing (I think the likelihood of revivification is the same or better than cryonics, but the cost in terms of hours spent tapping at a keyboard is basically more than any human could pay in one lifetime)

• Froolow 4 May 2014 7:44 UTC

  2 points

  in reply to: The_Duck’s comment on: The Ex­tended Liv­ing-For­ever Strat­egy-Space

  This is an excellent comment, and it is extremely embarrassing for me that in a post on the plausible ‘live forever’ strategy space I missed three extremely plausible strategies for living forever, all of which are approximately complementary to cryonics (unless they’re successful, in which case; why would you bother). I’d like to take this as evidence that many eyes on the ‘live forever’ problem genuinely does result in utility increase, but I think it is a more plausible explanation that I’m not very good at visualising the strategy space!

• Froolow 3 May 2014 10:22 UTC

  2 points

  in reply to: Nornagest’s comment on: The Ex­tended Liv­ing-For­ever Strat­egy-Space

  I basically agree with you that the strategy seems pretty unlikely. But I think you are over-harsh on it; you don’t need to reconstruct the entire brain, just the stuff that deals with personal identity. If you can select from any one of thirty keys on your keyboard then every ten letters you type has 10^15 bits of entropy, so it seems possible that if somebody knew absolutely everything about the state you were in when typing they could reconstruct you just from this. You are also not restricted to tapping away randomly—I suspect words or sentences would leak way more the pseudorandom tapping. At any rate, this strategy is almost free, so you’d need astonishingly good reasons not to attempt it if you plan on attempting cryonics.

  I think those reasons exist (I’m skeptical the information would survive) but I don’t think the theory is quite as much in the lunatic fringe as you do.

• Froolow 3 May 2014 10:09 UTC

  2 points

  in reply to: The_Duck’s comment on: The Ex­tended Liv­ing-For­ever Strat­egy-Space

  I think the plausibility of the arguments depends in a very great part on how plausible you think cryonics is; since the average on this site is about 22%, I can see how other strategies which are low likelihood/​high payoff might appear almost not worth considering. On the other hand, something like ‘simulationist’ preservation seems to me to be well within two orders of magnitude of the probability of cryonics—both rely on society finding your information and deciding to do something with it, and both rely on the invention of technology which appears logically possible but well outside the realms of current science (overcome death vs overcome computational limits on simulations). But simulation preservation is three orders of magnitude cheaper than cryonics, which suggests to me that it might be worthwhile to consider. That is to say, if you seriously dismissed it in a couple of seconds you must have very very strong reasons to think the strategy is—say—about four orders of magnitude less likely than cryonics. What reason is that? I wonder if maybe I assumed the simulation problem was more widely accepted than I thought it might be. I’m a bit concerned about this line of reasoning, because all of my friends dismiss cryonics as ‘obviously not worth considering’ and I think they adopt this argument because the probabilistic conclusions are uncomfortable to contemplate.

  With respect to your second point, that this post could be counter-productive, I am hugely interested by the conclusion. A priori it seems hugely unlikely that with all of our ingenuity we can only come up with two plausible strategies for living forever (religion and cryonics) and that both of those conclusions would be anathemic to the other group. If the ‘plausible strategy-space’ is not large I would take that as evidence that the strategy-space is in fact zero and people are just good at aggregating around plausible-but-flawed strategies. Can you think about any other major human accomplishment for which the strategy-space is so small? I suspect the conclusion for this is that I am bad at thinking up alternate strategies, rather than the strategies not existing, but it is an excellent point you make and well worth considering

• Froolow 2 May 2014 17:25 UTC

  2 points

  in reply to: ChristianKl’s comment on: The Ex­tended Liv­ing-For­ever Strat­egy-Space

  I’m not sure I agree with your analysis of the first—it is reasonable to assume that when a person generates pseudorandom noise they are masking a ‘signal’ with some amount of true randomness; we don’t know enough to say for absolute certain that the input is totally garbage and we have good reason to believe people are actually very bad at generating random numbers. Contrast that to—for example—the fact that we have pretty good reasons to think that bringing someone back from the dead is a hard project and I don’t think you’re fairly applying the same criteria across preservation methods.