Jan_Rzymkowski

• Jan_RzymkowskiJun 22, 2014, 7:50 PM

  −3 points

  in reply to: V_V’s comment on: Will AGI sur­prise the world?

  Sounds very reasonable.

• Jan_RzymkowskiJun 22, 2014, 7:46 PM

  7 points

  in reply to: Mark_Friedenbach’s comment on: Will AGI sur­prise the world?

  I’m not willing to engage in a discussion, where I defend my guesses and attack your prediction. I don’t have sufficient knowledge, nor a desire to do that. My purpose was to ask for any stable basis for AI dev predictions and to point out one possible bias.

  I’ll use this post to address some of your claims, but don’t treat that as argument for when AI would be created:

  How are Ray Kurzweil’s extrapolations an empiric data? If I’m not wrong, all he takes in account is computational power. Why would that be enough to allow for AI creation? By 1900 world had enough resources to create computers and yet it wasn’t possible, because the technology wasn’t known. By 2029 we may have proper resources (computational power), but still lack knowledge on how to use them (what programs run on that supercomputers).

  I’m not sure what you’re saying here. That we can assume AI won’t arrive next month because it didn’t arrive last month, or the month before last, etc.? That seems like shaky logic.

  I’m saying that, I guess, everybody would agree that AI will not arrive in a month. I’m interested on what basis we’re making such claim. I’m not trying to make an argument about when will AI arrive, I’m genuinely asking.

  You’re right about comforting factor of AI coming soon, I haven’t thought of that. But still, developement of AI in near future would probably mean that its creators haven’t solved the friendliness problem. Current methods are very black-box. More than that, I’m a bit concerned about current morality and governement control. I’m a bit scared, what may people of today do with such power. You don’t like gay marriage? AI can probably “solve” that for you. Or maybe you want financial equality of humanity? Same story. I would agree though that it’s hard to tell where would our preferences point to.

  If you assume the worst case that we will be unable to build AGI any faster than direct neural simulation of the human brain, that becomes feasible in the 2030′s on technological pathways that can be foreseen today.

  Are you taking in account that to this day we don’t truly understand biological mechanism of memory forming and developement of neuron connections? Can you point me to any predictions made by brain researchers about when we may expect technology allowing for full scan of human connectome and how close are we to understanding brain dynamics? (Creating of new synapses, control of their strenght, etc.)

  Once you are able to simulate the brain of a computational neuroscientist and give it access to its own source code, that is certainly enough for a FOOM.

  I’m tempted to call that bollocks. Would you expect a FOOM, if you’d give to a said scientist a machine telling him which neurons are connected and allowing to manipulate them? Humans can’t even understand nematoda’s neural network. You expect them to understand whole 100 billion human brain?

  Sorry for the above, it would need a much longer discussion, but I really don’t have strength for that.

  I hope it would be in any way helpful.

• Jan_RzymkowskiJun 22, 2014, 12:17 PM

  2 points

  in reply to: Mark_Friedenbach’s comment on: Will AGI sur­prise the world?

  This whole debate makes me wonder , if we can have any certainity for AI predictions. Almost all is based on personal opinions, highly susceptible to biases. And even people with huge knowledge about these biases aren’t safe. I don’t think anyone can trace their prediction back to empiric data, it all comes from our minds’ black boxes, to which biases have full access and which we can’t examine with our conciousness.

  While I find Mark’s prediction far from accurate, I know it might be just because I wouldn’t like it. I like to think that I would have some impact on AGI research, that some new insights are needed rather than just pumping more and more money in SIRI-like products. Developement of AI in next 10-15 years would mean that no qualitative research were needed and that all what is to be done is honing current technology. It would also mean there was time for thorough developement of friendliness and we may end up with AI catastroph.

  While I guess human level AI to rise in about 2070s, I know I would LIKE if it happened in 2070s. And I base this prediction on no solid base.

  Can anybody point me to any near-empiric data concerning, when AGI may be developed? Anything more solid than hunch of even most prominent AI researcher? Applying Moore’s law seems a bit magical, it without doubt has some Bayesian effect, but with little certainity.

  The best thing I can think of is that we all can agree, that AI is not be developed tomorrow. Or in a month. Why do we think that? It seems like coming from some very reliable empiric data. If we can identify factor, which make us near-certain AI is not be created in a span of few months from now, maybe upon closer look, it may provide us with some less shaky predictions for further future.

• Jan_RzymkowskiJun 19, 2014, 11:13 AM

  0 points

  in reply to: Kyre’s comment on: Paper­clip Max­i­mizer Re­vis­ited

  Yeah. Though actually it’s more of a simplified version of a more serious problem.

  One day you may give AI precise set of instructions, which you think would make good. Like find a way of curing diseases, but without harming patients, and without harming people for the sake of research and so on. And you may find that your AI is perfectly friendly, but it wouldn’t yet mean it actually is. It may simply have learned human values as a mean of securing its existence and gaining power.

  EDIT: And after gaining enough power it may as well help improve human health even more or reprogram human race to think unconditionaly that diseases were eradicated.

• Jan_RzymkowskiJun 19, 2014, 1:31 AM

  −4 points

  in reply to: solipsist’s comment on: [LINK] Elon Musk in­ter­ested in AI safety

  But Musk starts with mentioning “Terminator”. There’s plenty of sf literature showing much more accuratly danger of AI, though none of them as widely known as “Terminator”.

  That AI may have unexpected dangers seems too vague to me, to expect Musk to think along lines of LWers.

Paper­clip Max­i­mizer Re­vis­ited

• Jan_RzymkowskiJun 19, 2014, 12:53 AM

  1 point

  in reply to: ShardPhoenix’s comment on: [LINK] Elon Musk in­ter­ested in AI safety

  It’s not only unlikely—what’s much worse, is that it points to wrong reasons. It suggests that we should fear AI trying to take over the world or eliminating all people, as if AI would have incentive to do that. It stems from nothing more, but anthropomorphisation of AI, imagining it as some evil genius.

  This is very bad, because smart people can see that those reasonings are flawed and get impression that these are the only arguments against unbounded developement of AGI. While reverse stupidity isn’t smart, it’s much harder to find good reasons why we should solve AI friendliness, when there are lots of distracting strawmans.

  It was me from half a year ago. I used to think, that anybody, who fears AI may bring harm, is a loony. All the reasons I heard from people were that AI wouldn’t know emotions, AI would try to harmfully save people from themselves, AI would want to take over the world, AI would be infected by virus or hacked or that AI would be just outright evil. I can easily debunk all of above. And then I read about Paperclip Maximizer and radically changed my mind. I might got to that point much sooner, if not for all the strawman distractions.

• Jan_RzymkowskiJun 18, 2014, 11:52 PM

  −4 points

  on: [LINK] Elon Musk in­ter­ested in AI safety

  Ummm… He points to “Terminator” movie. Doesn’t that mean he’s just going along usual “AI will revolt and enslave the human race… because it’s evil!” rather than actually realising what existential risk involving AI is?

  I started to use it as a good rule of thumb. When somebody mentions Skynet, he’s probably not worth listening to. Skynet really isn’t a reasonable scenario for what may go wrong with AI.

• Jan_RzymkowskiJun 18, 2014, 7:32 PM

  2 points

  in reply to: NancyLebovitz’s comment on: Ra­tion­al­ist Sport

  While yoga seems like a salutary way of spending time, I woudn’t call that sport. Clear win-states and competition seems crutial to sport.

  And that’s why sport for rationalists is someting so hard to come up with and so valuable—it needs to combine the happiness from the effort to be better than others, while battling the sense of superiority, which often comes with winning.

  Sense of group superiority is to me the most revolting thing about most sports.

• Jan_RzymkowskiJun 17, 2014, 8:32 PM

  6 points

  in reply to: Stuart_Armstrong’s comment on: [LINK] The er­rors, in­sights and les­sons of fa­mous AI pre­dic­tions: preprint

  Now I think I shouldn’t mention hindsight bias, it doesn’t really fit here. I’m just saying that some events would be more probably famous, like: a) laymen posing extraordinary claim and ending up being right b) group of experts being spectacularly wrong

  If some group of experts met in 1960s and pose very cautious claims, chances are small that it would end up being widely known. And ending up in above paper. Analysing famous predictions is bound to end up with many overconfident predictions—they’re just more flashy. But it doesn’t yet mean most of predictions are overconfident.

• Jan_RzymkowskiJun 17, 2014, 5:33 PM

  4 points

  on: [LINK] The er­rors, in­sights and les­sons of fa­mous AI pre­dic­tions: preprint

  Isn’t this article highly susceptible to hindsight bias? For example, the reason authors analyse Dreyfus’s prediction is that, he was somewhat right. If he weren’t, authors woudn’t include that data-point. Therefore it skewes the data, even if it is not their intention.

  It’s hard to take valuable assessements from the text, when it would be naturally prone to highlight mistakes of the experts and correct predictions by laymen.

• Jan_RzymkowskiJun 15, 2014, 8:38 PM

  4 points

  in reply to: Skeptityke’s comment on: Group Ra­tion­al­ity Diary, June 1-15

  It reminds me greatly my making of conlangs (artificial languages). While I find it creative, it takes vehement amounts of time to just create a simple draft and an arduous work to make satisfactory material. And all I’d get is just two or three people calling it cool and showing just a small interest. And I always know I’ll get bored with that language in few days and never make as much as to translate simple texts.

  And yet every now and then I get an amazing idea and can’t stop myself from “wasting” hours, planning and writing about some conlang. And I end up being unsatisfied.

  I don’t think it is about Sunk Cost. It’s more about a form of addiction toward creative works. Some kind of vicious cycle, where brain engages in activity, that just makes you want more to do that activity. The more you work on it, the more you want to do it, until reaching saturation, when you just can’t look at it anymore.

• Jan_RzymkowskiJun 10, 2014, 7:53 PM

  0 points

  in reply to: NancyLebovitz’s comment on: Come up with bet­ter Tur­ing Tests

  Stuart, it’s not about control groups, but that such test actually would test negatively for blind, who are intelligent. Blind AI would also test negatively, so how is that useful?

  Actually physics test is not about getting closer to humans, but about creating something useful. If we can teach program to do physics, we can teach it to do other stuff. And we’re getting somewhere mid narrow and real AI.

• Jan_RzymkowskiJun 10, 2014, 1:27 PM

  3 points

  on: Come up with bet­ter Tur­ing Tests

  Ad 4. Elite judges is quite arbitrary. I’d rather iterate the test, each time choosing only those judges, who recognized program correctly or some variant of that (e.g. top 50% with most correct guesses). This way we select those, who go beyond simply conforming to a conversation and actually look for differences between program and human. (And as seen from transcripts, most people just try to have a conversation, rather than looking for flaws) Drawback is that, if program has set personality, judges could just stick to identifing that personality rather than human characteristics.

  Another approach might be that, the same pair program-human is examined by 10 judges consecutively, each spending 5 minutes with both. The twist is that judges can leave instructions for next judges. So if program fails to perform “If you want to prove you’re human, simply do nothing for 4 minutes, then re-type this sentence I’ve just written here, skipping one word out of 2”, than every judge after the one, who found that flaw, can use that and make right guess.

  My favourite method would be to give bot a simple physics textbook and then ask him to solve few physics test problems. Even if it wouldn’t be actual AI, it would still prove helluva powerful. Just toss it summarized knowledge on quantum physics and ask to solve for GUT. Sadly, most humans wouldn’t pass such high-school physics test.

  1. is actually original Turing Test.

  EDIT:

  1. is bad. It would exclude equally many actual AI and blind people as well. It is actually more general problem with Turing Test. It helps testing programs that mimic humans, but not AI in general. For text based AI, senses are alien. You could develop real intelligence, which would fail, when asked “How you like the smell of glass?”. Sure it can be taught that glass don’t smell, but it actually needs superhuman abilities. So while superintelligence can perfectly mimic human, human-level AI wouldn’t pass Turing Test, when asked about sensual stuff, just as humans would fail, when asked about nuances of geometry in four dimensions.

• Jan_RzymkowskiMay 13, 2014, 3:09 PM

  −2 points

  in reply to: Jiro’s comment on: Guardians of the Gene Pool

  You’re right. I got way too far with claiming equivalence.

  As for non-identity problem—I have trouble answering it. I don’t want to defend my idea, but I can think of an example when one brings up non-identity and comes to wrong conclusion: Drinking alcohol while pregnant can cause a fetus to develop a brain damage. But such grave brain damage means this baby is not the same one, that would be created, if his mother didn’t drink. So it is questionable that the baby would benefit from its mother abstinence.

• Jan_RzymkowskiMay 12, 2014, 10:24 PM

  8 points

  on: Univer­sal Fire

  Little correction:

  Phosphorus is highly reactive; pure phosphorus glows in the dark and may spontaneously combust. Phosphorus is thus also well-suited to its role in adenosine triphosphate, ATP, your body’s chief method of storing chemical energy.

  Actually, the above isn’t true. Reactivity is a property of a molecule, not of an element. Elemental phosphorus is prone to get oxidised with atmospheric oxygen, producing lots of energy. ATP is reactive, because anhydride bonds are fairly unstable—but none change of oxidation takes place. That it contains phosphorus, isn’t the actual reason for ATP to be an easy usable form of stroring energy. Salts of phosphoric acid also contain phosphorus, while being fairly unreactive. Thus the implication just doesn’t make sense.

• Jan_RzymkowskiMay 9, 2014, 11:07 PM

  −2 points

  on: Guardians of the Gene Pool

  “if you failed hard enough to endorse coercive eugenics”

  This might be found a bit too controversial, but I was tempted to come up with not-so-revolting coercive eugenics system. Of course it’s not needed, if there is technology for correcting genes, but let’s say we only have circa 1900 technology. It has nothing to do with the point of Elizer’s note, it’s ust my musing.

  Coervie eugenics isn’t strictly immoral itself. It is a way of protecting people not yet born from genetical flaws—possible diseases, etc. But even giving them less then optimal features—intelligence, strength, looks—is quite equivalent to making them stupidier, weaker, uglier. If you could give your child healthy and pleasent life, yet decide to strip him from that, you are hurting him—it’s not like his well-being is your property. But can you have YOUR child, while eugenics prevent you from breeding? Not in genetic sense, but it seems deeply flawed to base parent-child relation simply on genetic code. It’s upbringing that matters. Adopted child is in any meaningful way YOUR child. But there are two problems—you can’t really use “good genes” people for producing babies for “bad gene” people and “bad gene” mothers may have problem caring newborns without hormonal effect of birth. Way to make eugenics weaker, but overcome these problems, is to limit only mens’ breading. When a couple with “good gene man” wants children—let them. If couple with “bad gene man” wants children, then future mother is impregnated by some (possible hired) “good gene man”. Normally the couple have protected sex.

  It is by no means perfect. But the price for relative well-being of future people is only for a woman to have sex with not her husband, and for husband to be “cheated on”. While it seems quite unsettling, it’s mainly our cultural norm. While this might be unpleasant for both, it isn’t considerably worse then not being able to drink and smoke for woman through pregnancy. Therefore, such coercive eugenics would gradually improve gene pool, while not being considerable more evil then forbidding pregnant woman to smoke cigarettes.

  I don’t mean to say that such a system would be a good choice. But simply that it would be trading the rights of alive for the rights of not yet born.

  I apologize, if above was inappropriate.