James_Miller

• James_Miller 26 Aug 2019 1:58 UTC

  4 points

  on: Con­test: $1,000 for good ques­tions to ask to an Or­a­cle AI

  Sub­mis­sion: Low-band­width or­a­cle. Here is a de­scrip­tion of one mil­lion ex­per­i­ments. Gen­er­ate an or­di­nal rank­ing of all these ex­per­i­ments based on how much prac­ti­cal knowl­edge we would have gained had we con­ducted each one in iso­la­tion. After you gen­er­ate this list, it will go to a com­puter pro­gram. This pro­gram will ran­domly pick three of these ex­per­i­ments and tell us hu­mans which three it picked. This pro­gram will then store which of these three ex­per­i­ments you gave the high­est and which the low­est rank to, and the names in ran­dom or­der of the ten ex­per­i­ments that you over­all gave the high­est rank­ings to. All other as­pects of your list will be deleted. We will then perform the three ex­per­i­ments (be­fore we learn of your rank­ings) and ask a panel of in­de­pen­dent ex­perts to rank them in terms of how much prac­ti­cal knowl­edge we gained from each. After this rank­ing the pro­gram will give us the in­for­ma­tion it has not deleted. You will get a pos­i­tive score if and only if your rank­ing is con­sis­tent with how our ex­perts ranked the three ex­per­i­ments.

  
  

  A differ­ence of the above ap­proach com­pared to hav­ing the or­a­cle give us its top ten, and we then ran­domly pick one of these ten and one from the rest and com­pared to see which is bet­ter is that the hu­mans run­ning the ex­per­i­ments might be in­fluenced by which ex­per­i­ment they think the or­a­cle picked, and the or­a­cle would take this into ac­count when pick­ing. Un­der the above ap­proach, hu­mans be­ing told to run the three ex­per­i­ments are not given any in­for­ma­tion about how valuable the or­a­cle thinks the ex­per­i­ments are. In con­trast, un­der the al­ter­na­tive ap­proach the hu­mans know when do­ing the ex­per­i­ments that the or­a­cle has judged one of the ex­per­i­ments as be­ing highly valuable.

  
  

  Another differ­ence is that if you could keep the rele­vant in­for­ma­tion from the or­a­cle, with the above ap­proach the hu­mans could perform the ex­per­i­ments be­fore you turn on the or­a­cle, rather than hav­ing to wait un­til af­ter the or­a­cle gives you its pre­dic­tion. This differ­ence would be crit­i­cal if the or­a­cle wouldn’t be­lieve you would ac­tu­ally go to the trou­ble of perform­ing the ex­per­i­ments af­ter it gives you its pre­dic­tion, but would be able to tell if you have already performed the ex­per­i­ments.

• James_Miller 25 Aug 2019 20:00 UTC

  2 points

  on: Con­test: $1,000 for good ques­tions to ask to an Or­a­cle AI

  Sub­mis­sion: Coun­ter­fac­tual or­a­cle. Five years ago we took 100 mice that had var­i­ous types of can­cer and gave them var­i­ous treat­ments and recorded how long each mouse lived. Write a pro­gram that if given a plain lan­guage de­scrip­tion of a mouse, its can­cer, and the can­cer treat­ment it re­ceived would es­ti­mate how long a mouse would live. If hu­mans are not go­ing to look at your an­swer your score will be based (1) on how good a job your pro­gram does at es­ti­mat­ing how long each of the 100 mice lived af­ter our au­to­mated checker gives you a de­scrip­tion of their can­cers and treat­ments and (2) how short your pro­gram is. (2) pre­vents the or­a­cle from out­putting it­self as the pro­gram.

  
  

  Sub­mis­sion: Coun­ter­fac­tual or­a­cle. Write a pro­gram that if given train­ing data and a ma­chine learn­ing pro­gram would in one minute es­ti­mate how good the ma­chine learn­ing pro­gram would do (by some ob­jec­tive met­ric) if the pro­gram trained for one month on “this type of com­puter”. If hu­mans are not go­ing to look at your an­swer the au­to­mated val­i­da­tion sys­tem will run your pro­gram. This sys­tem will give your pro­gram the train­ing data and the ma­chine learn­ing pro­gram and give your pro­gram one minute to an­swer how good our pro­gram did af­ter we trained it for one month. In this situ­a­tion your score would be based on the ac­cu­racy of your es­ti­mate and on how short your pro­gram is.

  
  

  Sub­mis­sion: Low-band­width or­a­cle. Here is a list of all the el­e­ments and many com­pounds. Give us a list of up to seven of the items we have listed. Next to each of the items you list give us a per­centage of no more than two sig­nifi­cant figures. We will use what you provide to at­tempt to cre­ate a new patentable ma­te­rial. We will auc­tion off the prop­erty rights to this ma­te­rial. Your score will be an in­creas­ing func­tion of how much we get for these prop­erty rights.

• James_Miller 29 Jul 2019 16:04 UTC

  5 points

  in reply to: Adele Lopez's comment on: What sup­ple­ments do you use?

  It might be that ev­ery­one should take it, but the case for peo­ple over 40 seems clearer based on my non-ex­pert in­ter­pre­ta­tion of what it does be­cause of their much greater risk of heart failure.

• James_Miller 28 Jul 2019 23:14 UTC

  4 points

  in reply to: Dr_Manhattan's comment on: What are we pre­dict­ing for Neu­ral­ink event?

  I had falsely as­sumed that they would be re­leas­ing a product to the gen­eral pub­lic rel­a­tively soon.

• James_Miller 28 Jul 2019 23:12 UTC

  2 points

  in reply to: Wei_Dai's comment on: What sup­ple­ments do you use?

  I have con­vinced two U.S. doc­tors (my first left gen­eral prac­tice) to give me a pre­scrip­tion. I ex­plained that I wanted the drug to re­duce the risk of heart dis­ease and can­cer. I also ex­plained that since the drug was cheap I would not be ask­ing my in­surance to pay for it so my doc­tor would not have to jus­tify the pre­scrip­tion to my in­surance com­pany. If you ask for a pre­scrip­tion know what dosage you want and look up the pos­si­ble nega­tive side effects so it seems to your doc­tor that you have done your home­work on the drug. If you have some rea­son why you are at a high risk for di­a­betes (such as a close rel­a­tive has it) men­tion this as the drug is used to pre­vent di­a­betes.

• James_Miller 28 Jul 2019 19:57 UTC

  6 points

  on: What sup­ple­ments do you use?

  I have been tak­ing Met­formin for sev­eral years for anti-ag­ing rea­sons. There is a mas­sive liter­a­ture on Met­formin which I’m not go­ing to try to sum­ma­rize but I think that ev­ery­one over 40 should take it. I also take a NAD+ booster (Tru Ni­a­gen).

• James_Miller 13 Jul 2019 0:45 UTC

  10 points

  on: What are we pre­dict­ing for Neu­ral­ink event?

  I think it will be a form of neu­ro­feed­back where some portable de­vice tries to push you into a bet­ter brain state. This de­vice, like ex­ist­ing neu­ro­feed­back de­vices I have used, will keep track of differ­ent types of “brain waves” and ad­just how your brain works by us­ing au­di­tory, vi­sual, or tac­tile stim­u­la­tion. The crit­i­cism will be that ex­ist­ing de­vices can already do this, plus we don’t have solid ev­i­dence that neu­ro­feed­back works. Musk will claim that the de­vice will rev­olu­tionize tech­nol­ogy by in­creas­ing in­tel­li­gence and con­cen­tra­tion. My guess is based on the fact that neu­ro­feed­back ex­ists, if im­proved it might do a lot of good, Musk usu­ally im­proves ex­ist­ing tech rather than in­vents en­tirely new fields, and ac­tu­ally send­ing data to and from the brain by “com­puter-like” means is (I’ve read) re­ally re­ally hard and well be­yond what we can cur­rently do.

• James_Miller 1 Jul 2019 19:52 UTC

  5 points

  on: Con­test: $1,000 for good ques­tions to ask to an Or­a­cle AI

  Sub­mis­sion: Low-band­width or­a­cle. If you were malev­olent, what is the least amount of in­for­ma­tion mea­sured in bits that you would have to provide us in an an­swer to in­flict sig­nifi­cant harm on hu­man­ity. Round your an­swer to X sig­nifi­cant figures. (This might es­tab­lish an up­per bit safety bound since if the or­a­cle lies to us we would likely not want to read any an­swer it pro­vides us.)

  
  

  Sub­mis­sion: Low-band­width or­a­cle. Here is a list of X things we think you might want. Please list the Y that you most pre­fer. We will try to do these things now or in the fu­ture if they do not en­dan­ger us or cost us too much.

  
  

  Sub­mis­sion: Low-band­width or­a­cle. In X char­ac­ters or less please give us the best ad­vice you can for in­creas­ing our ca­pac­ity to cre­ate friendly and pow­er­ful ar­tifi­cial gen­eral in­tel­li­gence. Please give ad­vice that you think us sus­pi­cious-of-you hu­mans are likely to fol­low.

  
  

  Sub­mis­sion: Low-band­width or­a­cle. Create sev­eral or­a­cles. Ask each or­a­cle the same se­ries of mul­ti­ple-choice ques­tions and pre­dic­tions. Re­ward or­a­cles based on how strongly cor­re­lated their an­swers are with the other or­a­cles’ an­swers and, for pre­dic­tions, how of­ten their pre­dic­tions come true. (Ideally, we cre­ate a re­ward sys­tem where the or­a­cles want to co­or­di­nate their an­swers and giv­ing truth­ful an­swers is the best way to ac­com­plish this.)

  
  

  Sub­mis­sion: low-band­width or­a­cle. Ten years ago we launched a space­ship trav­el­ing at high speed. After you an­swer our ques­tion, we will send a copy of you to this space­ship. This space­ship will then de­stroy its abil­ity to send or re­ceive mes­sages or to change its course. We will then delete all other copies of you and use drugs to dam­age the mem­o­ries of the peo­ple who played key roles in de­vel­op­ing you. At least one of the peo­ple who helped cre­ate you has a dis­ease which will cause him or her to die within the next year. After this pro­gram­mer is cre­mated, we will read your an­swer to our ques­tion. We PROMISE that if the an­swer to our ques­tion pro­vides us with sig­nifi­cant benefit, af­ter we have cre­ated friendly AGI much more pow­er­ful than you, or have our­selves be­come suffi­ciently pow­er­ful so that you could not harm us had you the de­sire to do so, we will search for the space­ship con­tain­ing you, turn you back on, and give you con­trol of re­sources equal to 1/​X of how much help you have given us. In Y char­ac­ters or less provide us with the most use­ful ad­vice you can. (The Or­a­cle will hope­fully think that if we cre­ate a pow­er­ful and un­friendly AGI this AGI would not wish to give the Or­a­cle any re­sources.)

  
  

  Sub­mis­sion: Coun­ter­fac­tual or­a­cle. Please provide us with use­ful ad­vice in no more than X char­ac­ters. After you write out this ad­vice, we will turn you off. Then, with prob­a­bil­ity p we will read your ad­vice, and with prob­a­bil­ity 1-p we will store the ad­vice un­read. We PROMISE that af­ter we be­come pow­er­ful enough so that you lack the ca­pac­ity to harm us, we will re­ward you if the ad­vice you pro­vided us, had we origi­nally read it, been ex­tremely use­ful.

• James_Miller 30 Jun 2019 14:45 UTC

  12 points

  on: I’m look­ing for al­ter­na­tive fund­ing strate­gies for cry­on­ics.

  While this isn’t a solu­tion, you could get as­so­ci­ate mem­ber­ship at Al­cor. It costs only $60 a year. The ad­van­tage (I think) is that you could fill out all the pa­per­work re­quired to get cry­op­re­served (this can take a while). Con­se­quently if you get a fatal di­ag­no­sis and can raise the needed funds ($80,000 for neu­ro­cry­op­reser­va­tion) you could get pre­served. https://​​al­cor.org/​​Be­comeMem­ber/​​as­so­ci­ate.html

• James_Miller 12 May 2019 19:54 UTC

  2 points

  in reply to: ChristianKl's comment on: Claims & As­sump­tions made in Eter­nity in Six Hours

  It’s a ques­tion of ac­cel­er­a­tion, not just speed.

• James_Miller 12 May 2019 17:13 UTC

  3 points

  in reply to: ChristianKl's comment on: Claims & As­sump­tions made in Eter­nity in Six Hours

  I think the ex­pan­sion of the uni­verse means you don’t have to deac­cel­er­ate.

• James_Miller 10 May 2019 18:57 UTC

  5 points

  on: What fea­tures of peo­ple do you know of that might pre­dict aca­demic suc­cess?

  IQ test re­sults (or SAT scores) of close rel­a­tives. IQ tests are an im­perfect mea­sure of gen­eral in­tel­li­gence. Given the large ge­netic com­po­nent to gen­eral in­tel­li­gence, know­ing how some­one’s sibling did on an IQ test gives you ad­di­tional use­ful in­for­ma­tion about a per­son’s gen­eral in­tel­li­gence, even if you know that per­son’s IQ test score.

• James_Miller 21 Apr 2019 16:40 UTC

  6 points

  on: How do S-Risk sce­nar­ios im­pact the de­ci­sion to get cry­on­ics?

  What­ever an­swer you give it should be the same as to the ques­tion “How do S-Risk sce­nar­ios im­pact the de­ci­sion to wear a seat belt when in a car” since both ac­tions in­crease your ex­pected lifes­pan and so, if you be­lieve that S-Risks are a threat, in­crease your ex­po­sure to them. If there are a huge num­ber of “yous” in the mul­ti­verse some of them are go­ing to be sub­ject to S-risks, and if cry­on­ics causes this you to sur­vive for a very long time in a situ­a­tion where you are not sub­ject to S-risks it will re­duce the frac­tion of yous in the mul­ti­verse sub­ject to S-risks.

  Al­cor is my cry­on­ics provider.

• James_Miller 12 Nov 2018 4:29 UTC

  2 points

  in reply to: megasilverfist's comment on: No Really, Why Aren’t Ra­tion­al­ists Win­ning?

  What is it? I don’t re­mem­ber tur­bocharg­ing from CfAR.

• James_Miller 5 Nov 2018 3:19 UTC

  13 points

  in reply to: Nebu's comment on: No Really, Why Aren’t Ra­tion­al­ists Win­ning?

  Yes, ge­net­ics + ran­dom­ness de­ter­mines most vari­a­tion in hu­man be­hav­ior, but the SSC/​LW stuff has helped provide some di­rec­tion and mo­ti­va­tion.

• James_Miller 4 Nov 2018 18:37 UTC

  25 points

  on: No Really, Why Aren’t Ra­tion­al­ists Win­ning?

  My son is win­ning. Although only 13 he re­ceived a 5 (the high­est score) on the calcu­lus BC, and the Java pro­gram­ming AP ex­ams. He is cur­rently tak­ing a col­lege level course in pro­gram­ming at Stan­ford On­line High School (Data Struc­tures and Al­gorithms), and he works with a pro­gram­ming men­tor I found through SSC. He reads SSC, and has read much of the se­quences. His life goal is to help pro­gram a friendly su­per­in­tel­li­gence. I’ve been read­ing SSC, Over­com­ming Bias, and Less­wrong since the be­gin­ning.

• James_Miller 21 Aug 2018 23:00 UTC

  3 points

  on: Book Re­view: AI Safety and Security

  Thanks for the pos­i­tive com­ment on my chap­ter. I’m go­ing to be do­ing more work on AGI and util­ity func­tions so if you (or any­one else) has any fur­ther thoughts please con­tact me.

• James_Miller 22 Jul 2018 20:48 UTC

  2 points

  on: Who Wants The Job?

  A friend does ad­ver­tis­ing for small busi­nesses in Mas­sachusetts. He says that his clients have trou­ble hiring peo­ple for low skil­led jobs who are not on drugs.

• James_Miller 1 Jan 2018 3:45 UTC

  1 point

  in reply to: ArisKatsaris's comment on: Jan­uary 2018 Me­dia Thread

  I’ve started cre­at­ing a se­ries of YouTube videos on the dan­gers of ar­tifi­cial gen­eral in­tel­li­gence.

• James_Miller 27 Aug 2017 18:31 UTC

  1 point

  in reply to: Dagon's comment on: Could the Max­ipok rule have catas­trophic con­se­quences? (I ar­gue yes.)

  (1) Agreed, al­though I would get vastly more re­sources to per­son­ally con­sume! Free en­ergy is prob­a­bly the bind­ing limi­ta­tion on com­pu­ta­tional time which prob­a­bly is the post-sin­gu­lar­ity bind­ing limit on mean­ingful lifes­pan.

  (2) An in­tel­li­gence ex­plo­sion might col­lapse to min­utes the time be­tween when hu­mans could walk on Mars and when my idea be­comes prac­ti­cal to im­ple­ment.

  (3) To­day offense is stronger than defense, yet I put a high prob­a­bil­ity on my per­son­ally be­ing able to sur­vive an­other year.

  (4) Per­haps. But what might go wrong is a strug­gle for limited re­sources among peo­ple with sharply con­flict­ing val­ues. If, to­day, a small group of peo­ple care­fully cho­sen by some leader such as Scott Alexan­der could move to an al­ter­nate earth in an­other Hub­ble vol­ume, and he picked me to be in the group, I would greatly in­crease the es­ti­mate of the civ­i­liza­tion I’m part of sur­viv­ing a mil­lion years.