Major update here.

The state of affairs regarding the SIAI and its underlying rationale and rules of operation are insufficiently clear. 

Most of the arguments involve a few propositions and the use of probability and utility calculations to legitimate action. Here much is uncertain to an extent that I'm not able to judge any nested probability estimations. Even if you tell me, where is the data on which you base those estimations?

There seems to be an highly complicated framework of estimations to support and reinforce each other. I'm not sure how you call this in English, but in German I'd call that a castle in the air.

I know that what I'm saying may simply be due to a lack of knowledge and education, that is why I am inquiring about it. How many of you, who currently support the SIAI, are able to analyse the reasoning that led you to support the SIAI in the first place, or at least substantiate your estimations with other kinds of evidence than a coherent internal logic?

I can follow much of the reasoning and arguments on this site. But I'm currently unable to judge their overall credence. Are the conclusions justified? Is the coherent framework build around the SIAI based on firm ground? There seems to be no critical inspection or examination by a third party. There is no peer review. Yet people are willing to donate considerable amounts of money.

I'm concerned that, although consistently so, the SIAI and its supporters are updating on fictional evidence. This post is meant to inquire about the foundations of your basic premises. Are you creating models to treat subsequent models or are your propositions based on fact?

An example here is the use of the Many-worlds interpretation. Itself a logical implication, can it be used to make further inferences and estimations without additional evidence? MWI might be the only consistent non-magic interpretation of quantum mechanics. The problem here is that such conclusions are, I believe, widely considered not to be enough to base further speculations and estimations on. Isn't that similar to what you are doing when speculating about the possibility of superhuman AI and its consequences? What I'm trying to say here is that if the cornerstone of your argumentation, if one of your basic tenets is the likelihood of superhuman AI, although a valid speculation given what we know about reality, you are already in over your head with debt. Debt in the form of other kinds of evidence. Not to say that it is a false hypothesis, that it is not even wrong, but that you cannot base a whole movement and a huge framework of further inference and supportive argumentation on such premises, on ideas that are themselves not based on firm ground.

The gist of the matter is that a coherent and consistent framework of sound argumentation based on unsupported inference is nothing more than its description implies. It is fiction. Imagination allows for endless possibilities while scientific evidence provides hints of what might be possible and what impossible. Science does provide the ability to assess your data. Any hint that empirical criticism provides gives you new information on which you can build on. Not because it bears truth value but because it gives you an idea of what might be possible. An opportunity to try something. There’s that which seemingly fails or contradicts itself and that which seems to work and is consistent.

And that is my problem. Given my current educational background and knowledge I cannot differentiate LW between a consistent internal logic, i.e. imagination or fiction, and something which is sufficiently based on empirical criticism to provide a firm substantiation of the strong arguments for action that are proclaimed by the SIAI.

Further, do you have an explanation for the circumstance that Eliezer Yudkowsky is the only semi-popular person who's aware of something that might shatter the universe? Why is it that people like Vernor Vinge, Robin Hanson or Ray Kurzweil are not running amok using all their influence to convince people of the risks ahead, or at least give all they have to the SIAI? Why aren't Eric Drexler, Gary Drescher or AI researches like Marvin Minsky worried to the extent that they signal their support?

I'm talking to quite a few educated people outside this community. They do not doubt all those claims for no particular reason. Rather they tell me that there are too many open questions to focus on the possibilities depicted by the SIAI and to neglect other near-term risks that might wipe us out as well.

I believe that many people out there know a lot more than I do, so far, about related topics and yet they seem not to be nearly as concerned about the relevant issues than the average Less Wrong member. I could have named other people. That's besides the point though, it's not just Hanson or Vinge but everyone versus Eliezer Yudkowsky and some unknown followers. What about the other Bayesians out there? Are they simply not as literate as Eliezer Yudkowsky in the maths or maybe somehow teach but not use their own methods of reasoning and decision making?

What do you expect me to do, just believe Eliezer Yudkowsky? Like I believed so much in the past which made sense but turned out to be wrong? Maybe after a few years of study I'll know more.


2011-01-06: As this post received over 500 comments I am reluctant to delete it. But I feel that it is outdated and that I could do much better today. This post has however been slightly improved to account for some shortcomings but has not been completely rewritten, neither have its conclusions been changed. Please account for this when reading comments that were written before this update.

2012-08-04: A list of some of my critical posts can be found here: SIAI/lesswrong Critiques: Index

New Comment
633 comments, sorted by Click to highlight new comments since:
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

(Disclaimer: My statements about SIAI are based upon my own views, and should in no way be interpreted as representing their stated or actual viewpoints on the subject matter. I am talking about my personal thoughts, feelings, and justifications, no one else's. For official information, please check the SIAI website.)

Although this may not answer your questions, here are my reasons for supporting SIAI:

  • I want what they're selling. I want to understand morality, intelligence, and consciousness. I want a true moral agent outside of my own thoughts, something that can help solve that awful, plaguing question, "Why?" I want something smarter than me that can understand and explain the universe, providing access to all the niches I might want to explore. I want something that will save me from death and pain and find a better way to live.

  • It's the most logical next step. In the evolution of mankind, intelligence is a driving force, so "more intelligent" seems like an incredibly good idea, a force multiplier of the highest order. No other solution captures my view of a proper future like friendly AI, not even " space!"

  • No one else cares about the big

... (read more)
Good, informative comment.
Yeah, that's why I'm donating as well. Sure, but why the SIAI? I accept this. Although I'm not sure if the big picture should be a top priority right now. And as I wrote, I'm unable to survey the utility calculations at this point. So you replace a simple view that is evidende based with one that might or might not be based on really shaky ideas such as an intelligence explosion. I think you overestimate the friendliness of friendly AI. Too bad Roko's posts have been censored. I want to believe. Beware of those who agree with you? Maybe we do have enough time regarding AI and the kind of threats depicted on this site. Maybe we don't have enough time regarding other kinds of threats. I can accept that. But I'm unable to follow the process of elimination yet.
Who else is working directly on creating smarter-than-human intelligence with non-commercial goals? And if there are any, are they self-reflective enough to recognize its potential failure modes? I used something I developed which I call Point-In-Time Utility to guide my thinking on this matter. It basically boils down to, 'the longest view wins', and I don't see anyone else talking about potentially real pangalactic empires. I don't think it has to be an explosion at all, just smarter-than-human. I'm willing to take things one step at a time, if necessary. Though it seems unlikely we could build a smarter-than-human intelligence without understanding what intelligence is, and thus knowing where to tweak, if even retroactively. That said, I consider intelligence tweaking itself to be a shaky idea, though I view alternatives as failure modes. I think you overestimate my estimation of the friendliness of friendly AI. Note that at the end of my post I said it is very likely SIAI will fail. My hope total is fairly small. Roko deleted his own posts, and I was able to read the article Eliezer deleted since it was still in my RSS feed. It didn't change my thinking on the matter; I'd heard arguments like it before. Hi. I'm human. At least, last I checked. I didn't say all my reasons were purely rational. This one is dangerous (reinforcement), but I do a lot of reading of opposing opinions as well, and there's still a lot I disagree with regarding SIAI's positions. The latter is what I'm worried about. I see all of these threats as being developed simultaneously, in a race to see which one passes the threshold into reality first. I'm hoping that Friendly AI beats them. I haven't seen you name any other organization you're donating to or who might compete with SIAI. Aside from the Future of Humanity Institute or the Lifeboat Foundation, both of which seem more like theoretical study groups than action-takers, people just don't seem to be working on these problems. Even
That there are no other does not mean we shouldn't be keen to create them, to establish competition. Or do it at all at this point. I'm not sure about this. I feel there are too many assumptions in what you state to come up with estimations like a 1% probability of uFAI turning everything into paperclips. You are right, never mind what I said. Yeah and how is their combined probability less worrying than that of AI? That doesn't speak against the effectiveness of donating all to the SIAI of course. Creating your own God to fix the problems the imagined one can't is indeed a promising and appealing idea, given it is feasible. I'm mainly concerned about my own well-being. If I was threated by something near-term within Germany, that would be my top-priority. So the matter is more complicated for me than for the people who are merely conerned about the well-being of all beings. As I said before, it is not my intention to discredit the SIAI but to steer some critical discussion for us non-expert, uneducated but concerned people.
Absolutely agreed. Though I'm barely motivated enough to click on a PayPal link, so there isn't much hope of my contributing to that effort. And I'd hope they'd be created in such a way as to expand total funding, rather than cannibalizing SIAI's efforts. Certainly there are other ways to look at value / utility / whatever and how to measure it. That's why I mentioned I had a particular theory I was applying. I wouldn't expect you to come to the same conclusions, since I haven't fully outlined how it works. Sorry. I'm not sure what this is saying. I think UFAI is far more likely than FAI, and I also think that donating to SIAI contributes somewhat to UFAI, though I think it contributes more to FAI, such that in the race I was talking about, FAI should come out ahead. At least, that's the theory. There may be no way to save us. AI is one of the things on the list racing against FAI. I think AI is actually the most dangerous of them, and from what I've read, so does Eliezer, which is why he's working on that problem instead of, say, nanotech. I've mentioned before that I'm somewhat depressed, so I consider my philanthropy to be a good portion 'lack of caring about self' more than 'being concerned about the well-being of all beings'. Again, a subtractive process. Thanks! I think that's probably a good idea, though I would also appreciate more critical discussion from experts and educated people, a sort of technical minded anti-Summit, without all the useless politics of the IEET and the like.
It's more likely that the Klingon warbird can overpower the USS Enterprise. Why? Because EY told you? I'm not trying to make snide remarks here but how people arrived at this conclusion was what I have been inquiring about in the first place. Me too, but I was the only one around willing to start one at this point. That's the sorry state of critical examination.
To pick my own metaphor, it's more likely that randomly chosen matter will form clumps of useless crap than a shiny new laptop. As defined, UFAI is likely the default state for AGI, which is one reason I put such low hope on our future. I call myself an optimistic pessimist: I think we're going to create wonderful, cunning, incredibly powerful technology, and I think we're going to misuse it to destroy ourselves. Because intelligent beings are the most awesome and scary things I've ever seen. The History Channel is a far better guide than Eliezer in that respect. And with all our intelligence and technology, I can't see us holding back from trying to tweak intelligence itself. I view it as inevitable. I'm hoping that the Visiting Fellows program and the papers written with the money from the latest Challenge will provide peer review in other respected venues.
What I was trying to show you by the Star Trek metaphor is that you are making estimations within a framework of ideas of which I'm not convinced to be based on firm ground.
I'm not a very good convincer. I'd suggest reading the original material.
Can we get some links up in here? I'm not putting the burden on you in particular, but I think more linkage would be helpful in this discussion.
This thread has Eliezer's request for specific links, which appear in replies.

I'm currently preparing for the Summit so I'm not going to hunt down and find links. Those of you who claimed they wanted to see me do this should hunt down the links and reply with a list of them.

Given my current educational background I am not able to judge the following claims (among others) and therefore perceive it as unreasonable to put all my eggs in one basket:

You should just be discounting expected utilities by the probability of the claims being true, and then putting all your eggs into the basket that has the highest marginal expected utility per dollar, unless you have enough resources to invest that the marginal utility goes down. This is straightforward to anyone who knows about expected utility and economics, and anyone who knows about scope insensitivity knows why this result is counterintuitive to the human brain. We don't emphasize this very hard when people talk in concrete terms about donating to more than one organization, because charitable dollars are not substitutable from a limited pool, the main thing is the variance in the tiny fraction of their income people donate to charity in the first place and so the amount of warm glow people generate for th... (read more)

An example here is the treatment and use of MWI (a.k.a. the "many-worlds interpretation") and the conclusions, arguments and further estimations based on it. No doubt MWI is the only consistent non-magic interpretation of quantum mechanics. But that's it, an interpretation. A logical consistent deduction. Or should I rather call it an induction, as the inference seems to be of greater generality than the premises, at least as understood within the LW community? But that's besides the point. The problem here is that such conclusions are, I believe, widely considered to be weak evidence to base further speculations and estimations on.

Reading the QM sequence (someone link) will show you that to your surprise and amazement, what seemed to you like an unjustified leap and a castle in the air, a mere interpretation, is actually nailed down with shocking solidity.

What I'm trying to argue here is that if the cornerstone of your argumentation, if one of your basic tenets is the likelihood of exponential evolving superhuman AI, although a valid speculation given what we know about reality, you are already in over your head with debt. Debt in the form of other kinds of evidence

... (read more)
The relevant fallacy in 'Aristotelian' logic is probably false dilemma, though there are a few others in the neighborhood.
Probably black and white thinking.
I haven't done the work to understand MWI yet, but if this FAQ is accurate, almost nobody likes the Copenhagen interpretation (observers are SPECIAL) and a supermajority of "cosmologists and quantum field theorists" think MWI is true. Since MWI seems to have no practical impact on my decision making, this is good enough for me. Also, Feynman likes it :)
Thanks for taking the time to give a direct answer. I enjoyed reading this and these replies will likely serve as useful comments to when people ask similar questions in the future.
Where are the formulas? What are the variables? Where is this method exemplified to reflect the decision process of someone who's already convinced, preferably of someone within the SIAI? That is part of what I call transparency and a foundational and reproducible corroboration of one's first principles. Awesome, I never came across this until now. It's not widely mentioned? Anyway, what I notice from the Wiki entry is that one of the most important ideas, recursive improvement, that might directly support the claims of existential risks posed by AI, is still missing. All this might be featured in the debate, hopefully with reference to substantial third-party research papers, I don't know yet. The whole point of the grey goo example was to exemplify the speed and sophistication of nanotechnology that would have to be around to either allow an AI to be build in the first place or be of considerable danger. That is, I do not see how an encapsulated AI, even a superhuman AI, could pose the stated risks without the use of advanced nanotechnology. Is it going to use nukes, like Skynet? Another question related to the SIAI, regarding advanced nanotechnology, is that if without advanced nanotechnology superhuman AI is at all possible. This is an open question and I'm inquiring about how exactly the uncertainties regarding these problems are accounted for in your probability estimations of the dangers posed by AI. What I was inquiring about is the likelihood of slow versus fast development of AI. That is, how fast after we got AGI will we see the rise of superhuman AI? The means of development by which a quick transcendence might happen is circumstantial to the meaning of my question. Where are your probability estimations that account for these uncertainties. Where are your variables and references that allow you to make any kind of estimations to balance the risks of a hard rapture with a somewhat controllable development? You misinterpreted my question. What I me
Um... yes? Superhuman is a low bar and, more importantly, a completely arbitrary bar. Evidence based? By which you seem to mean 'some sort of experiment'? Who would be insane enough to experiment with destroying the world? This situation is exactly where you must understand that evidence is not limited to 'reference to historical experimental outcomes'. You actually will need to look at 'consistent internal logic'... just make sure the consistent internal logic is well grounded on known physics. And that, well, that is actually a reasonable point. You have been given some links (regarding human behavior) that are good answer to the question but it is nevertheless non-trivial. Unfortunately now you are actually going to have to do the work and read them.
Is it? That smarter(faster)-than-human intelligence is possible is well grounded on known physics? If that is the case, how does it follow that intelligence can be applied to itself effectively, to the extent that one could realistically talk about "explosive" recursive self-improvement?
Some still seem sceptical - and you probably also need some math, compsci and philosophy to best understand the case for superhuman intelligence being possible.
Not only is there evidence that smarter than human intelligence is possible it is something that should be trivial given a vaguely sane reductionist model. Moreover you specifically have been given evidence on previous occasions when you have asked similar questions. What you have not been given and what are not available are empirical observations of smarter than human intelligences existing now. That is evidence to which you would not be entitled.
Please provide a link to this effect? (Going off topic, I would suggest that a "show all threads with one or more comments by users X, Y and Z" or "show conversations between users X and Y" feature on LW might be useful.) (First reply below)
Please provide such a link. (Going off-topic, I additionally suggest that a "show all conversations between user X and user Y" feature on Less Wrong might be useful.)
It is currently not possible for me to either link or quote. I do not own a computer in this hemisphere and my android does not seem to have keys for brackets or greater than symbols. workarounds welcome.
The solution varies by model, but on mine, alt-shift-letter physical key combinations do special characters that aren't labelled. You can also use the on-screen keyboard, and there are more onscreen keyboards available for download if the one you're currently using is badly broken.
SwiftKey x beta Brilliant!
OK, can I have my quote(s) now? It might just be hidden somewhere in the comments to this very article.
Can you copy and paste characters?
Uhm...yes? It's just something I would expect to be integrated into any probability estimates of suspected risks. More here. Check the point that you said is a reasonable one. And I have read a lot without coming across any evidence yet. I do expect an organisation like the SIAI to have detailed references and summaries about their decision procedures and probability estimations to be transparently available and not hidden beneath thousands of posts and comments. "It's somewhere in there, line 10020035, +/- a million lines...." is not transparency! That is, an organisation who's conerned with something taking over the universe and asks for your money. And organisation I'm told of which some members get nightmares just reading about evil AI...
I think you just want a brochure. We keep telling you to read archived articles explaining many of the positions and you only read the comment where we gave the pointers, pretending as if that's all that's contained in our answers. It'd be more like him saying, "I have a bunch of good arguments right over there," and then you ignore the second half of the sentence.
I'm not asking for arguments. I know them. I donate. I'm asking for more now. I'm using the same kind of anti-argumentation that academics would use against your arguments. Which I've encountered myself a few times while trying to convince them to take a look at the inscrutable archives of posts and comment that is LW. What do they say? "I skimmed over it, but there were no references besides some sound argumentation, an internal logic.", "You make strong claims, mere arguments and conclusions extrapolated from a few premises are insufficient to get what you ask for."
Pardon my bluntness, but I don't believe you, and that disbelief reflects positively on you. Basically, if you do know the arguments then a not insignificant proportion of your discussion here would amount to mere logical rudeness. For example if you already understood the arguments for, or basic explanation of why 'putting all your eggs in one basket' is often the rational thing to do despite intuitions to the contrary then why on earth would you act like you didn't?
Oh crap, the SIAI was just a punching bag. Of course I understand the arguments for why it makes sense not to split your donations. If you have a hundred babies but only food for 10, you are not going to portion it to all of the hundred babies but feed the strongest 10. Otherwise you'd end up having a hundred dead babies in which case you could as well have eaten the food yourself before wasting it like that. It's obvious, I don't see how someone wouldn't get this. I used that idiom to illustrate that given my preferences and current state of evidence I could as well eat all the food myself rather than wasting it on something I don't care to save or that doesn't need to be saved in the first place because I missed the fact that all the babies are puppets and not real. I asked, are the babies real babies that need food and is the expected utility payoff of feeding them higher than eating the food myself right now? I'm starting to doubt that anyone actually read my OP...
I know this is just a tangent... but that isn't actually the reason. Just to be clear, I'm not objecting to this. That's a reasonable point.
Ok. Is there a paper, article, post or comment that states the reason or is it spread all over LW? I've missed the reason then. Seriously, I'd love to read up on it now. Here is an example of what I want:
Good question. If not there should be. It is just basic maths when handling expected utilities but it crops up often enough. Eliezer gave you a partial answer: ... but unfortunately only asked for a link for the 'scope insensivity' part, not a link to a 'marginal utility' tutorial. I've had a look and I actually cant find such a reference on LW. A good coverage of the subject can be found in an external paper, Heuristics and biases in charity. Section 1.1.3 Diversification covers the issue well.
That's another point. As I asked, what are the variables, where do I find the data? How can I calculate this probability based on arguments to be found on LW? This IS NOT sufficient to scare people up to the point of having nightmares and ask them for most of their money.
I'm not trying to be a nuisance here, but it is the only point I'm making right now, and the one that can be traced right back through the context. It is extremely difficult to make progress in a conversation if I cannot make a point about a specific argument without being expected to argue against an overall position that I may or may not even disagree with. It makes me feel like my arguments must come armed as soldiers.
I'm sorry, I perceived your comment to be mainly about decision making regarding charities. Which is completely marginal since the SIAI is the only charity concerned with the risk I'm inquiring about. Is the risk in question even real and does its likelihood justify the consequences and arguments for action? I inquired about the decisions making regarding charities because you claimed that what I stated about egg allocation is not the point being made. But I do not particularly care about that question as it is secondary.
Leave aside SIAI specific claims here. The point Eliezer was making, was about 'all your eggs in one basket' claims in general. In situations like this (your contribution doesn't drastically change the payoff at the margin, etc) putting all your eggs in best basket is the right thing to do. You can understand that insight completely independently of your position on existential risk mitigation.
Er, there's a post by that title.
Questionable. Is smarter than human intelligence possible in a sense comparable to the difference between chimps and humans? To my awareness we have no evidence to this end. Questionable. How is an encapsulated AI going to get this kind of control without already existing advanced nanotechnology? It might order something over the Internet if it hacks some bank account etc. (long chain of assumptions), but how is it going to make use of the things it orders? I believe that self-optimization is prone to be very limited. Changing anything substantial might lead Gandhi to swallow the pill that will make him want to hurt people, so to say. Sound argumentation that gives no justification to extrapolate it to an extent that you could apply it to the shaky idea of a superhuman intellect coming up with something better than science and applying it again to come up... All those ideas about possible advantages of being an entity that can reflect upon itself to the extent of being able to pinpoint its own shortcoming is again, highly speculative. This could be a disadvantage. Much of the rest is about the plateau argument, once you got a firework you can go to the moon. Well yes, I've been aware of that argument. But that's weak, that there are many hidden mysteries about reality that we completely missed yet is highly speculative. I think even EY admits that whatever happens, quantum mechanics will be a part of it. Is the AI going to invent FTL travel? I doubt it, and it's already based on the assumption that superhuman intelligence, not just faster intelligence, is possible. Like the discovery that P ≠ NP? Oh wait, that would be limiting. This argument runs in both directions. Assumption. Nice idea, but recursion does not imply performance improvement. How can he make any assumptions then about the possibility to improve them recursively, given this insight, to an extent that they empower an AI to transcendent into superhuman realms? Did he just attribute intention

Questionable. Is smarter than human intelligence possible in a sense comparable to the difference between chimps and humans? To my awareness we have no evidence to this end.

What would you accept as evidence?

Would you accept sophisticated machine learning algorithms like the ones in the Netflix contest, who find connections that make no sense to humans, who simply can't work with high-dimensional data?

Would you accept a circuit designed by a genetic algorithm, which doesn't work in the physics simulation but works better in reality than anything humans have designed, with mysterious parts that are not connected to anything but are necessary for it to function?

Would you accept a chess program which could crush any human chess player who ever lived? Kasparov at ELO 2851, Rybka at 3265. Wikipedia says grandmaster status comes at ELO 2500. So Rybka is now even further beyond Kasparov at his peak as Kasparov was beyond a new grandmaster. And it's not like Rybka or the other chess AIs will weaken with age.

Or are you going to pull a no-true-Scotsman and assert that each one of these is mechanical or unoriginal or not really beyond human or just not different enough?

I think it at least possible that much-smarter-than human intelligence might turn out to be impossible. There exist some problem domains where there appear to be a large number of solutions, but where the quality of the solutions saturate quickly as more and more resources are thrown at them. A toy example is how often records are broken in a continuous 1-D domain, with attempts drawn from a constant probability distribution: The number of records broken goes as the log of the number of attempts. If some of the tasks an AGI must solve are like this, then it might not do much better than humans - not because evolution did a wonderful job of optimizing humans for perfect intelligence, but because that part of the problem domain is a brick wall, and anything must bash into it at nearly the same point. One (admittedly weak) piece of evidence: a real example of saturation, is an optimizing compiler being used to recompile itself. It is a recursive optimizing system, and, if there is a knob to allow more effort being used on the optimization, the speed-up from the first pass can be used to allow a bit more effort to be applied to a second pass for the same cpu time. Nonetheless, the results for this specific recursion are not FOOM. The evidence in the other direction are basically existence proofs from the most intelligent people or groups of people that we know of. Something as intelligent as Einstein must be possible, since Einstein existed. Given an AI Einstein, working on improving its own intelligence - it isn't clear if it could make a little progress or a great deal.
This goes for your compilers as well, doesn't it? There are still major speed-ups available in compilation technology (the closely connected areas of whole-program compilation+partial evaluation+supercompilation), but a compiler is still expected to produce isomorphic code, and that puts hard information-theoretic bounds on output.
Can you provide details / link on this?
I should've known someone would ask for the cite rather than just do a little googling. Oh well. Turns out it wasn't a radio, but a voice-recognition circuit. From :
The analogy that AGI can be to us as we are to chimps. This is the part that needs the focus. We could have said in the 1950s that machines beat us at arithmetic by orders of magnitude. Classical AI researchers clearly were deluded by success at easy problems. The problem with winning on easy problems is that it says little about hard ones. What I see is that in the domain of problems for which human level performance is difficult to replicate, computers are capable of catching us and likely beating us, but gaining a great distance on us in performance is difficult. After all, a human can still beat the best chess programs with a mere pawn handicap. This may never get to two pawns. ever. Certainly the second pawn is massively harder than the first. It's the nature of the problem space. In terms of runaway AGI control of the planet, we have to wonder if humans will always have the equivalent of a pawn handicap via other means (mostly as a result of having their hands on the reigns of the economic, political, and legal structures). BTW, is ELO supposed to have that kind of linear interpretation?
Yes, this is the important part. Chimps lag behind humans in 2 distinct ways - they differ in degree, and in kind. Chimps can do a lot of human-things, but very minimally. Painting comes to mind. They do a little, but not a lot. (Degree.) Language is another well-studied subject. IIRC, they can memorize some symbols and use them, but not in the recursive way that modern linguistics (pace Chomsky) seems to regard as key, not recursive at all. (Kind.) What can we do with this distinction? How does it apply to my three examples? O RLY? Ever is a long time. Would you like to make this a concrete prediction I could put on PredictionBook, perhaps something along the lines of 'no FIDE grandmaster will lose a 2-pawns-odds chess match(s) to a computer by 2050'? I'm not an expert on ELO by any means (do we know any LW chess experts?), but reading through doesn't show me any warning signs - ELO point differences are supposed to reflect probabilistic differences in winning, or a ratio, and so the absolute values shouldn't matter. I think.
This is a possibility (made more plausible if we're talking about those reins being used to incentivize early AIs to design more reliable and transparent safety mechanisms for more powerful successive AI generations), but it's greatly complicated by international competition: to the extent that careful limitation and restriction of AI capabilities and access to potential sources of power reduces economic, scientific, and military productivity it will be tough to coordinate. Not to mention that existing economic, political, and legal structures are not very reliably stable: electorates and governing incumbents often find themselves unable to retain power.
It seems that whether or not it's supposed to, in practice it does. From the just released "Intrinsic Chess Ratings", which takes Rybka and does exhaustive evaluations (deep enough to be 'relatively omniscient') of many thousands of modern chess games; on page 9:
You are getting much closer than any of the commenter's before you to provide some other form of evidence to substantiate one of the primary claims here. You have to list your primary propositions on which you base further argumentation, from which you draw conclusions and which you use to come up with probability estimations stating risks associated with former premises. You have to list these main principles so anyone who comes across claims of existential risks and a plead for donation, can get an overview. Then you have to provide the references you listed above, if you believe they give credence to the ideas, so that people see that all you say isn't made up but based on previous work and evidence by people that are not associated with your organisation. No, although I have heard about all of the achievements I'm not yet able to judge if they provide evidence supporting the possibility of strong superhuman AI, the kind that would pose a existential risk. Although in the case of chess I'm pretty much the opinion that this is no strong evidence as it is not sufficiently close to be able to overpower humans to an extent of posing a existential risk when extrapolated into other areas. It would be good if you could provide links to the mentioned examples. Especially the genetic algorithm (ETA: Here.). It is still questionable however if this could lead to the stated recursive improvements or will shortly hit a limit. To my knowledge genetic algorithms are merely used for optimization, based on previous design spaces and are not able to come up with something unique to the extent of leaving their design space. Whether sophisticated machine learning algorithms are able to discover valuable insights beyond statistical inferences within higher-dimensional data-sets is a very interesting idea though. As I just read, the 2009 prize of the Netflix contest was given to a team that achieved a 10.05% improvement over the previous algorithm. I'll have to examine this furth
I am reluctant because you seem to ask for magical programs when you write things like: I was going to link to AIXI and approximations thereof; full AIXI is as general as an intelligence can be if you accept that there are no uncomputable phenomenon, and the approximations are already pretty powerful (from nothing to playing Pac-Man). But then it occurred to me that anyone invoking a phrase like 'leaving their design space' might then just say 'oh, those designs and models can only model Turing machines, and so they're stuck in their design space'.
I've no idea (formally) of what a 'design space' actually is. This is a tactic I'm frequently using against strongholds of argumentation that are seemingly based on expertise. I use their own terminology and rearrange it into something that sounds superficially clever. I like to call it a Chinese room approach. Sometimes it turns out that all they were doing was to sound smart but cannot explain themselves when faced with their own terminology set to inquire about their pretences. I thank you however for taking the time to actually link to further third party information that will substantiate given arguments for anyone not trusting the whole of LW without it.
I see. Does that actually work for you? (Note that your answer will determine whether I mentally re-categorize you from 'interested open-minded outsider' to 'troll'.)
It works against cults and religion in general. I don't argue with them about their religion being not even wrong but rather accept their terms and highlight inconsistencies within their own framework by going as far as I can with one of their arguments and by inquiring about certain aspects based on their own terminology until they are unable to consistently answer or explain where I am wrong. This also works with the anti GM-food bunch, data protection activists, hippies and many other fringe groups. For example, the data protection bunch concerned with information disclosure on social networks or Google Streetview. Yes, I say, that's bad, burglar could use such services to check out your house! I wonder what evidence there is for the increase of burglary in the countries where Streetview is already available for many years? Or I tell the anti-gun lobbyists how I support their cause. It's really bad if anyone can buy a gun. Can you point me to the strong correlation between gun ownership and firearm homicides? Thanks.
Any specific scenario is going to have burdensome details, but that's what you get if you ask for specific scenarios rather than general pressures, unless one spends a lot of time going through detailed possibilities and vulnerabilities. With respect to the specific example, regular human criminals routinely swindle or earn money anonymously online, and hack into and control millions of computers in botnets. Cloud computing resources can be rented with ill-gotten money. In the unlikely event of a powerful human-indifferent AI appearing in the present day, a smartphone held by a human could provide sensors and communication to use humans for manipulators (as computer programs direct the movements of some warehouse workers today). Humans can be paid, blackmailed, deceived (intelligence agencies regularly do these things) to perform some tasks. An AI that leverages initial capabilities could jury-rig a computer-controlled method of coercion [e.g. a cheap robot arm holding a gun, a tampered-with electronic drug-dispensing implant, etc]. And as time goes by and the cumulative probability of advanced AI becomes larger, increasing quantities of robotic vehicles and devices will be available.
Thanks, yes I know about those arguments. One of the reasons I'm actually donating and accept AI to be one existential risk. I'm inquiring about further supporting documents and transparency. More on that here, especially check the particle collider analogy.
With respect to transparency, I agree about a lack of concise, exhaustive, accessible treatments. Reading some of the linked comments about marginal evidence from hypotheses I'm not quite sure what you mean, beyond remembering and multiplying by the probability that particular premises are false. Consider Hanson's "Economic Growth Given Machine Intelligence". One might support it with generalizations from past population growth in plants and animals, from data on capital investment and past market behavior and automation, but what would you say would license drawing probabilistic inferences using it?
Note that such methods might not result in the destruction of the world within a week (the guaranteed result of a superhuman non-Friendly AI according to Eliezer.)
What guarantee?.
With a guarantee backed by $1000.
The linked bet doesn't reference "a week," and the "week" reference in the main linked post is about going from infrahuman to superhuman, not using that intelligence to destroy humanity. That bet seems underspecified. Does attention to "Friendliness" mean any attention to safety whatsoever, or designing an AI with a utility function such that it's trustworthy regardless of power levels? Is "superhuman" defined relative to the then-current level of human (or upload, or trustworthy less intelligent AI) capacity with any enhancements (or upload speedups, etc)? What level of ability counts as superhuman? You two should publicly clarify the terms.
A few comments later on the same comment thread someone asked me how much time was necessary, and I said I thought a week was enough, based on Eliezer's previous statements, and he never contradicted this, so it seems to me that he accepted it by default, since some time limit will be necessary in order for someone to win the bet. I defined superhuman to mean that everyone will agree that it is more intelligent than any human being existing at that time. I agree that the question of whether there is attention to Friendliness might be more problematic to determine. But "any attention to safety whatsoever" seems to me to be clearly stretching the idea of Friendliness-- for example, someone could pay attention to safety by trying to make sure that the AI was mostly boxed, or whatever, and this wouldn't satisfy Eliezer's idea of Friendliness.
Ah. So an AI could, e.g. be only slightly superhuman and require immense quantities of hardware to generate that performance in realtime.
Right. And if this scenario happened, there would be a good chance that it would not be able to foom, or at least not within a week. Eliezer's opinion seems to be that this scenario is extremely unlikely, in other words that the first AI will already be far more intelligent than the human race, and that even if it is running on an immense amount of hardware, it will have no need to acquire more hardware, because it will be able to construct nanotechnology capable of controlling the planet through actions originating on the internet as you suggest. And as you can see, he is very confident that all this will happen within a very short period of time.
Have you tried asking yourself non-rhetorically what an AI could do without MNT? That doesn't seem to me to be a very great inferential distance at all.
I believe that in this case an emulation would be the bigger risk because it would be sufficiently obscure and could pretend to be friendly for a long time while secretly strengthening its power. A purely artificial intelligence would be too alien and therefore would have a hard time to acquire the necessary power to transcend to a superhuman level without someone figuring out what it does, either by its actions or by looking at its code. It would also likely not have the intention to increase its intelligence infinitely anyway. I just don't see that AGI implies self-improvement beyond learning what it can while staying in scope of its resources. You'd have to deliberately implement such an intention. It would generally require its creators to solve a lot of problems much more difficult than limiting its scope. That is why I do not see run-away self-improvement as a likely failure mode. I could imagine all kinds of scenarios indeed. But I also have to assess their likelihood given my epistemic state. And my conclusion is that a purely artificial intelligence wouldn't and couldn't do much. I estimate the worst-case scenario to be on par with a local nuclear war.
I simply can't see where the above beliefs might come from. I'm left assuming that you just don't mean the same thing by AI as I usually mean. My guess is that you are implicitly thinking of a fairly complicated story but are not spelling that out.
And I can't see where your beliefs might come from. What are you telling potential donors or AGI researchers? That AI is dangerous by definition? Well, what if they have a different definition, what should make them update in favor of your definition? That you thought about it for more than a decade now? I perceive serious flaws in any of the replies I got so far in under a minute and I am a nobody. There is too much at stake here to base the decision to neglect all other potential existential risks on the vague idea that intelligence might come up with something we haven't thought about. If that kind of intelligence is as likely as other risks then it doesn't matter what it comes up with anyway because those other risks will wipe us out just as good and with the same probability. There already are many people criticizing the SIAI right now, even on LW. Soon, once you are more popular, other people than me will scrutinize everything you ever wrote. And what do you expect them to conclude if even a professional AGI researcher, who has been a member of the SIAI, does write the following: Why would I disregard his opinion in favor of yours? Can you present any novel achievements that would make me conclude that you people are actually experts when it comes to intelligence? The LW sequences are well written but do not showcase some deep comprehension of the potential of intelligence. Yudkowsky was able to compile previously available knowledge into a coherent framework of rational conduct. That isn't sufficient to prove that he has enough expertise on the topic of AI to make me believe him regardless of any antipredictions being made that weaken the expected risks associated with AI. There is also insufficient evidence to conclude that Yudkowsky, or someone within the SIAI, is smart enough to be able to tackle the problem of friendliness mathematically. If you would at least let some experts take a look at your work and assess its effectiveness and general potential.
Douglas Hofstadter and Daniel Dennett both seem to think these issues are probably still far away. ... *
I'm not sure who is doing that. Being hit by an asteroid, nuclear war and biological war are other possible potentially major setbacks. Being eaten by machines should also have some probability assigned to it - though it seems pretty challenging to know how to do that. It's a bit of an unknown unknown. Anyway, this material probably all deserves some funding.
The short-term goal seems more modest - prove that self-improving agents can have stable goal structures. If true, that would be fascinating - and important. I don't know what the chances of success are, but Yudkowsky's pitch is along the lines of: look this stuff is pretty important, and we are spending less on it than we do on testing lipstick. That's a pitch which it is hard to argue with, IMO. Machine intelligence research does seem important and currently-underfunded. Yudkowsky is - IMHO - a pretty smart fellow. If he will work on the problem for $80K a year (or whatever) it seems as though there is a reasonable case for letting him get on with it.
I'm not sure you're looking at the probability of other extinction risks with the proper weighting. The timescales are vastly different. Supervolcanoes: one every 350,000 years. Major asteroid strikes: one every 700,000 years. Gamma ray bursts: hundreds of millions of years, etc. There's a reason the word 'astronomical' means huge beyond imagining. Contrast that with the current human-caused mass extinction event: 10,000 years and accelerating. Humans operate on obscenely fast timescales compared to nature. Just with nukes we're able to take out huge chunks of Earth's life forms in 24 hours, most or all of it if we detonated everything we have in an intelligent, strategic campaign to end life. And that's today, rather than tomorrow. Regarding your professional AGI researcher and recursive self-improvement, I don't know, I'm not an AGI researcher, but it seemed to me that a prerequisite to successful AGI is an understanding and algorithmic implementation of intelligence. Therefore, any AGI will know what intelligence is (since we do), and be able to modify it. Once you've got a starting point, any algorithm that can be called 'intelligent' at all, you've got a huge leap toward mathematical improvement. Algorithms have been getting faster at a higher rate than Moore's Law and computer chips.
That might be true. But most of them have one solution that demands research in many areas. Space colonization. It is true that intelligent systems, if achievable in due time, play a significant role here. But not an exceptional role if you disregard the possibility of an intelligence explosion, of which I am very skeptical. Further, it appears to me that donating to the SIAI would rather impede research on such systems giving their position that such systems themselves posit an existential risk. Therefore, at the moment, the possibility of risks from AI is partially being outweighed to the extent that the SIAI should be supported yet doesn't hold an exceptional position that would necessarily make it the one charity with the highest expected impact per donation. I am unable to pinpoint another charity at the moment, e.g. space elevator projects, because I haven't looked into it. But I do not know of any comparison analysis, although you and many other people claim they have calculated it nobody ever published their efforts. As you know, I am unable to do such an analysis myself at this point as I am still learning the math. But I am eager to get the best information by means of feedback anyhow. Not intended as an excuse of course. That would surely be a very good argument if I was able to judge it. But can intelligence be captured by a discrete algorithm or is it modular and therefore not subject to overall improvements that would affect intelligence itself as a meta-solution? Also, can algorithms that could be employed in real-world scenarios be speed-up to have an effect that would warrant superhuman power? Take photosynthesis, could that particular algorithm be improved considerably, to an extent that it would be vastly better than the evolutionary one? Further, will such improvements be accomplishable fast enough to outpace human progress or the adaption of the given results? My problem is that I do not believe that intelligence is fathomable as a solution tha
This seems backwards - if intelligence is modular, that makes it more likely to be subject to overall improvements, since we can upgrade the modules one at a time. I'd also like to point out that we currently have two meta-algorithms, bagging and boosting, which can improve the performance of any other machine learning algorithm at the cost of using more CPU time. It seems to me that, if we reach a point where we can't improve an intelligence any further, it won't be because it's fundamentally impossible to improve, but because we've hit diminishing returns. And there's really no way to know in advance where the point of diminishing returns will be. Maybe there's one breakthrough point, after which it's easy until you get to the intelligence of an average human, then it's hard again. Maybe it doesn't become difficult until after the AI's smart enough to remake the world. Maybe the improvement is gradual the whole way up. But we do know one thing. If an AI is at least as smart as an average human programmer, then if it chooses to do so, it can clone itself onto a large fraction of the computer hardware in the world, in weeks at the slowest, but more likely in hours. We know it can do this because human-written computer viruses do it routinely, despite our best efforts to stop them. And being cloned millions or billions of times will probably make it smarter, and definitely make it powerful. In a sense, all thoughts are just the same words and symbols rearranged in different ways. But that is not the type of newness that matters. New software algorithms, concepts, frameworks, and programming languages are created all the time. And one new algorithm might be enough to birth an artificial general intelligence.
The AI will be much bigger than a virus. I assume this will make propagation much harder.
Harder, yes. Much harder, probably not, unless it's on the order of tens of gigabytes; most Internet connections are quite fast.
Anything could be possible - though the last 60 years of the machine intelligence field are far more evocative of the "blood-out of-a-stone" model of progress.
Smart human programmers can make dark nets too. Relatively few of them want to trash their own reputations and appear in the cross-hairs of the world's security services and law-enforcement agencies, though.
Reputation and law enforcement are only a deterrent to the mass-copies-on-the-Internet play if the copies are needed long-term (ie, for more than a few months), because in the short term, with a little more effort, the fact that an AI was involved at all could be kept hidden. Rather than copy itself immediately, the AI would first create a botnet that does nothing but spread itself and accept commands, like any other human-made botnet. This part is inherently anonymous; on the occasions where botnet owners do get caught, it's because they try to sell use of them for money, which is harder to hide. Then it can pick and choose which computers to use for computation, and exclude those that security researchers might be watching. For added deniability, it could let a security researcher catch it using compromised hosts for password cracking, to explain the CPU usage. Maybe the state of computer security will be better in 20 years, and this won't be as much of a risk anymore. I certainly hope so. But we can't count on it.
Thank you for continuing to engage my point of view, and offering your own. That's an interesting hypothesis which easily fits into my estimated 90+ percent bucket of failure modes. I've got all kinds of such events in there, including things such as, there's no way to understand intelligence, there's no way to implement intelligence in computers, friendliness isn't meaningful, CEV is impossible, they don't have the right team to achieve it, hardware will never be fast enough, powerful corporations or governments will get there first, etc. My favorite is: no matter whether it's possible or not, we won't get there in time; basically, that it will take too long to be useful. I don't believe any of them, but I do think they have solid probabilities which add up to a great amount of difficulty. But the future isn't set, they're just probabilities, and we can change them. I think we need to explore this as much as possible, to see what the real math looks like, to see how long it takes, to see how hard it really is. Because the payoffs or results of failure are in that same realm of 'astronomical'.
A somewhat important correction: To my knowledge, SIAI does not actually endorse neglecting all potential x-risks besides UFAI. (Analysis might recommend discounting the importance of fighting them head-on, but that analysis should still be done when resources are available.)
Not all of them - most of them. War, hunger, energy limits, resource shortages, space travel, loss of loved ones - and so on. It probably won't fix the speed of light limit, though.
What makes you reach this conclusion? How can you think any of these problems can be solved by intelligence when none of them have been solved? I'm particularly perplexed by the claim that war would be solved by higher intelligence. Many wars are due to ideological priorities. I don't see how you can expect necessarily (or even with high probability) that ideologues will be less inclined to go to war if they are smarter.
Violence has been declining on (pretty much) every timescale: Steven Pinker: Myth of Violence. I think one could argue that this is because of greater collective intelligence of human race.
War won't be solved by making everyone smarter, but it will be solved if a sufficiently powerful friendly AI takes over, as a singleton, because it would be powerful enough to stop everyone else from using force.
Yes, that makes sense, but in context I don't think that's what was meant since Tim is one of the people here is more skeptical of that sort of result.
Tim on "one big organism": * * *
Thanks for clarifying (here and in the other remark).
War has already been solved to some extent by intelligence (negotiations and diplomacy significantly decreased instances of war), hunger has been solved in large chunks of the world by intelligence, energy limits have been solved several times by intelligence, resource shortages ditto, intelligence has made a good first attempt at space travel (the moon is quite far away), and intelligence has made huge bounds towards solving the problem of loss of loved ones (vaccination, medical intervention, surgery, lifespans in the high 70s, etc). This is a constraint satisfaction problem (give as many ideologies as much of what they want as possible). Intelligence solves those problems.
I have my doubts about war, although I don't think most wars really come down to conflicts of terminal values. I'd hope not, anyway. However as for the rest, if they're solvable at all, intelligence ought to be able to solve them. Solvable means there exists a way to solve them. Intelligence is to a large degree simply "finding ways to get what you want". Do you think energy limits really couldn't be solved by simply producing through thought working designs for safe and efficient fusion power plants? ETA: ah, perhaps replace "intelligence" with "sufficient intelligence". We haven't already solved all these problems already in part because we're not really that smart. I think fusion power plants are theoretically possible, and at our current rate of progress we should reach that goal eventually, but if we were smarter we should obviously achieve it faster.
As various people have said, the original context was not making everybody more intelligent and thereby changing their inclinations, but rather creating an arbitrarily powerful superintelligence that makes their inclinations irrelevant. (The presumption here is typically that we know which current human inclinations such a superintelligence would endorse and which ones it would reject.) But I'm interested in the context you imply (of humans becoming more intelligent). My $0.02: I think almost all people who value war do so instrumentally. That is, I expect that most warmongers (whether ideologues or not) want to achieve some goal (spread their ideology, or amass personal power, or whatever) and they believe starting a war is the most effective way for them to do that. If they thought something else was more effective, they would do something else. I also expect that intelligence is useful for identifying effective strategies to achieve a goal. (This comes pretty close to being true-by-definition.) So I would only expect smarter ideologues (or anyone else) to remain warmongers if if starting a war really was the most effective way to achieve their goals. And if that's true, everyone else gets to decide whether we'd rather have wars, or modify the system so that the ideologues have more effective options than starting wars (either by making other options more effective, or by making warmongering less effective, whichever approach is more efficient). So, yes, if we choose to incentivize wars, then we'll keep getting wars. It follows from this scenario that war is the least important problem we face, so we should be OK with that. Conversely, if it turns out that war really is an important problem to solve, then I'd expect fewer wars.
I was about to reply - but jimrandomh said most of what I was going to say already - though he did so using that dreadful "singleton" terminology, spit. I was also going to say that the internet should have got the 2010 Nobel peace prize.
Is that really the idea? My impression is that the SIAI think machines without morals are dangerous, and that until there is more machine morality research, it would be "nice" if progress in machine intelligence was globally slowed down. If you believe that, then any progress - including constructing machine toddlers - could easily seem rather negative.
Darwinian gradualism doesn't forbid evolution taking place rapidly. I can see evolutionary progress accelerating over the course of my own lifespan - which is pretty incredible considering that evolution usually happens on a scale of millions of years. More humans in parallel can do more science and engineering. The better their living standard, the more they can do. Then there are the machines... Maybe some of the pressures causing the speed-up will slack off - but if they don't then humanity may well face a bare-knuckle ride into inner-space - and fairly soon.
Re: toddler-level machine intelligence. Most toddlers can't program, but many teenagers can. The toddler is a step towards the teenager - and teenagers are notorious for being difficult to manage.
The usual cite given in this area is the paper The Basic AI Drives. It suggests that open-ended goal-directed systems will tend to improve themselves - and to grab resources to help them fulfill their goals - even if their goals are superficially rather innocent-looking and make no mention of any such thing. The paper starts out like this:
Well, some older posts had a guy praising "goal system zero", which meant a plan to program an AI with the minimum goals it needs to function as a 'rational' optimization process and no more. I'll quote his list directly: This seems plausible to me as a set of necessary conditions. It also logically implies the intention to convert all matter the AI doesn't lay aside for other purposes (of which it has none, here) into computronium and research equipment. Unless humans for some reason make incredibly good research equipment, the zero AI would thus plan to kill us all. This would also imply some level of emulation as an initial instrumental goal. Note that sub-goal (1) implies a desire not to let instrumental goals like simulated empathy get in the way of our demise.
Perhaps, though if we can construct such a thing in the first place we may be able to deep-scan its brain and read its thoughts pretty well - or at least see if it is lying to us and being deceptive. IMO, the main problem there is with making such a thing in the first place before we have engineered intelligence. Brain emulations won't come first - even though some people seem to think they will.
Seconding this question.
Writing the word 'assumption' has its limits as a form of argument. At some stage you are going to have to read the links given.
This was a short critique of one of the links given. The first I skimmed over. I wasn't impressed yet. At least to the extent of having nightmares when someone tells me about bad AI's.
I like how Nick Bostrom put it re: probabilities and interesting future phenomena:
Index to the FOOM debate Antipredictions

Can I say, first of all, that if you want to think realistically about a matter like this, you will have to find better authorities than science-fiction writers. Their ideas are generally not their own, but come from scientific and technological culture or from "futurologists" (who are also a very mixed bunch in terms of intellect, realism, and credibility); their stories present speculation or even falsehood as fact. It may be worthwhile going "cold turkey" on all the SF you have ever read, bearing in mind that it's all fiction that was ground out, word by word, by some human being living a very ordinary life, in a place and time not very far from you. Purge all the imaginary experience of transcendence from your system and see what's left.

Of course science-fictional thinking, treating favorite authors as gurus, and so forth is endemic in this subculture. The very name, "Singularity Institute", springs from science fiction. And SF occasionally gets things right. But it is far more a phenomenon of the time, a symptom of real things, rather than a key to understanding reality. Plain old science is a lot closer to being a reliable guide to reality, thou... (read more)

Voted up for this argument. I think the SIAI would be well-served for accruing donations, support, etc. by emphasizing this point more. Space organizations might similarly argue: "You might think our wilder ideas are full of it, but even if we can't ever colonize Mars, you'll still be getting your satellite communications network."
I hadn't thought of it this way, but on reflection of course it's true.

Superhuman Artificial Intelligence (the runaway kind, i.e. God-like and unbeatable not just at Chess or Go).

This claim can be broken into two separate parts:

  1. Will we have human-level AI?
  2. Once we have human-level AI, will it develop to become superhuman AI?

For 1: looking at current technology trends, Sandberg & Bostrom estimate that we should have the technology needed for whole brain emulation around 2030-2050 or so, at least assuming that it gets enough funding and that Moore's law keeps up. Even if there isn't much of an actual interest in whole brain emulations, improving scanning tools are likely to revolutionize neuroscience. Of course, respected neuroscientists are already talking about reverse-engineering of the brain as being within reach. If we are successful at reverse engineering the brain, then AI is a natural result.

As for two, as Eliezer mentioned, this is pretty much an antiprediction. Human minds are a particular type of architecture, running on a particular type of hardware: it would be an amazing coincidence if it just happened that our intelligence couldn't be drastically improved upon. We already know that we're insanely biased, to the point of people ... (read more)

Do you have a citation for this? You can get certain biochemical compounds synthesized for you (there's a fair bit of a market for DNA synthesis) but that's pretty far from synthesizing microorganisms.
Right, sorry. I believe the claim (which I heard from a biologist) was that you can get DNA synthesized for you, and in principle an AI or anyone who knew enough could use those services to create their own viruses or bacteria (though no human yet has that required knowledge). I'll e-mail the person I think I heard it from and ask for a clarification.

Is there more to this than "I can't be bothered to read the Sequences - please justify everything you've ever said in a few paragraphs for me"?

My charitable reading is that he is arguing there will be other people like him and if SIAI wishes to continue growing there does need to be easily digested material.
From my experience as a long-time lurker and occasional poster, LW is not easily accessible to new users. The Sequences are indeed very long and time consuming, and most of them have multiple links to other posts you are supposed to have already read, creating confusion if you should happen to forget the gist of a particular post. Besides, Eliezer draws a number of huge philosophical conclusions (reductionism, computationalism, MWI, the Singularity, etc.), and a lot of people aren't comfortable swallowing all of that at once. Indeed, the "why should I buy all this?" question has popped into my head many times while reading. Furthermore, I think criticism like this is good, and the LW crowd should not have such a negative reaction to it. After all, the Sequences do go on and on about not getting unduly emotionally attached to beliefs; if the community can't take criticism, that is probably a sign that it is getting a little too cozy with its current worldview.

Criticism is good, but this criticism isn't all that useful. Ultimately, what SIAI does is the conclusion of a chain of reasoning; the Sequences largely present that reasoning. Pointing to a particular gap or problem in that chain is useful; just ignoring it and saying "justify yourselves!" doesn't advance the debate.

Agreed--criticism of this sort vaguely reminds me of criticism of evolution in that it attacks a particular part of the desired target rather than its fundamental assumptions (my apologies to the original poster). Still, I think we should question the Sequences as much as possible, and even misguided criticism can be useful. I'm not saying we should welcome an unending series of top-level posts like this, but I for one would like to see critical essays on of some of LW's most treasured posts. (There goes my afternoon...)
3Paul Crowley
Of course, substantive criticism of specific arguments is always welcome.
My primary point was to inquire about the foundation and credibility of named chain of reasoning. Is it a coherent internal logic that is reasoning about itself or is it based on firm ground? Take the following example: A recursively evolving AGI is quickly reaching a level that can be considered superhuman. As no advanced nanotechnology was necessary for its construction it is so far awful limited in what it can accomplish given its vast and fast intellect. Thus it solves all open problems associated with advanced nanotechnology and secretely mails its solutions a researcher. This researcher is very excited and consequently builds a corporation around this new technology. Later the AGI buys the stocks of that company and plants a front man. Due to some superhuman social engineering it finally obtains control of the technology... At this point we are already deep into subsequent reasoning about something shaky that at the same time is used as evidence of the very reasoning involving it. Taking a conclusion and running with it building a huge framework of further conclusions around it is in my opinion questionable. First this conclusion has to yield marginal evidence of its feasibility, then you are able to create a further hypothesis engaged with further consequences. You are making estimations within a framework that is itself not based on firm ground. The gist of what I was trying to say is not to subsequently base conclusions and actions on other conclusions which themselves do not bear evidence. I was inquiring about the supportive evidence at the origin of your complex multi-step extrapolations argued to be from inductive generalizations. If there isn't any, what difference is there between writing fiction and complex multi-step extrapolations argued to be from inductive generalizations? I've read and heard enough to be in doubt since I haven't come across a single piece of evidence besides some seemingly sound argumentation (as far as I can tell) in favor
Disagree. If you are asking people for money (and they are paying you), the burden is on you to provide justification at multiple levels of detail to your prospective or current donors. But, but... then you'll have to, like, repeat yourself a lot! No shit. If you want to change the world, be prepared to repeat yourself a lot.
If so... is that request bad? If you are running a program where you are trying to convince people on a large scale, then you need to be able to provide overviews of what you are saying at various levels of resolution. Getting annoyed (at one of your own donors!) for such a request is not a way to win. Edit: At the time, Eliezer didn't realize that XiXiDu was a donor.

Getting annoyed (at one of your own donors!) for such a request is not a way to win.

I don't begrudge SIAI at all for using Less Wrong as a platform for increasing its donor base, but I can definitely see myself getting annoyed sooner or later, if SIAI donors keep posting low-quality comments or posts, and then expecting special treatment for being a donor. You can ask Eliezer to not get annoyed, but is it fair to expect all the other LW regulars to do the same as well?

I'm not sure what the solution is to this problem, but I'm hoping that somebody is thinking about it.

Me too. The reason I upvoted this post was because I hoped it would stimulate higher quality discussion (whether complimentary, critical, or both) of SIAI in the future. I've been hoping to see such a discussion on LW for a while to help me think through some things.
4Paul Crowley
In other words, you see XiXiDu's post as the defector in the Asch experiment who chooses C when the group chooses B but the right answer is A?
To be fair, I don't think XiXiDu expected special treatment for being a donor; he didn't even mention it until Eliezer basically claimed that he was being insincere about his interest. (EDIT: Thanks to Wei Dai, I see he did mention it. No comment on motivations, then.) I think that Eliezer's statement is not an expression of a desire to give donors special treatment in general; it's a reflection of the fact that, knowing Xi is a donor and proven supporter of SIAI, he then ought to give Xi's criticism of SIAI more credit for being sincere and worth addressing somehow. If Xi were talking about anything else, it wouldn't be relevant.
1Wei Dai
He mentioned it earlier in a comment reply to Eliezer, and then again in the post itself:

These are reasonable questions to ask. Here are my thoughts:

  • Superhuman Artificial Intelligence (the runaway kind, i.e. God-like and unbeatable not just at Chess or Go).
  • Advanced real-world molecular nanotechnology (the grey goo kind the above intelligence could use to mess things up).

Virtually certain that these things are possible in our physics. It's possible that transhuman AI is too difficult for human beings to feasibly program, in the same way that we're sure chimps couldn't program trans-simian AI. But this possibility seems slimmer when you consider that humans will start boosting their own intelligence pretty soon by other means (drugs, surgery, genetic engineering, uploading) and it's hard to imagine that recursive improvement would cap out any time soon. At some point we'll have a descendant who can figure out self-improving AI; it's just a question of when.

  • The likelihood of exponential growth versus a slow development over many centuries.
  • That it is worth it to spend most on a future whose likelihood I cannot judge.

These are more about decision theory than logical uncertainty, IMO. If a self-improving AI isn't actually possible for a long time, then funding ... (read more)

What do you mean by plausible in this instance? Not currently refuted by our theories of intelligence or chemistry? Or something stronger.
Oh yeah, oops, I meant to say "possible in our physics". Edited accordingly.
Where is the evidence that does support the claims that it is not only possible, but that it will also turn out to be MUCH smarter than a human being, not just more rational or faster. Where is the evidence for an intelligence explosion? Is action justified simply based on the mere possibility that it might be physical possible? Not even your master believes this. Yes, once they turned themselves into superhuman intelligences? Isn't this what Kurzweil believes? No risks by superhuman AI because we'll go the same way anyway? Yep. Yes, but to allocate all my egs to them? Remember, they ask for more than simple support. I want to maximize my expected survival. If there are medium midterm risks that could kill me with a higher probability than AI in future, that is as important as the AI killing me later. Highly interesting. Sadly it is not a priority. I could, for example, start my own campaign to make people aware of possible risks. I could talk to people. I bet there's a lot more you smart people could do besides supporting EY. The SIAI and specially EY does not have the best reputation within the x-risk community and I bet that's the same in the AI community. That might very well be the case given how they handle public relations. He wasn't the first smart person who came to these conclusions. And he sure isn't charismatic. I've read and heard enough to be in doubt since I haven't come across a single piece of evidence besides some seemingly sound argumentation (as far as I can tell) in favor of some basic principles of unknown accuracy. And even those arguments are sufficiently vague that you cannot differentiate them from mere philosophical musing. And if you feel stupid because I haven't read hundreds of articles to find a single piece of third party evidence in favor of the outstanding premises used to ask for donations, then you should feel stupid.
Since I've now posted several comments on this thread defending and/or "siding with" XiXiDu, I feel I should state, for the record, that I think this last comment is a bit over the line, and I don't want to be associated with the kind of unnecessarily antagonistic tone displayed here. Although there are a couple pieces of the SIAI thesis that I'm not yet 100% sold on, I don't reject it in its entirety, as it now sounds like XiXiDu does - I just want to hear some more thorough explanation on a couple of sticking points before I buy in. Also, charisma is in the eye of the beholder ;)
I think I should say more about this. That EY has no charisma is, I believe, a reasonable estimation. Someone who says of himself that he's not neurotypical likely isn't a very appealing person in the eye of the average person. Then I got much evidence in the form of direct comments about EY that show that many people do not like him personally. Now let's examine if I am hostile to EY and his movement. First a comment I made regarding Michael Anissimov' 26th birthday. I wrote: Let's examine my opinion about Eliezer Yudkowsky. * Here I suggest EY to be the most admirable person. * When I recommended reading Good and Real to a professional philosopher I wrote, "Don't know of a review, a recommendation by Eliezer Yudkowsky as 'great' is more than enough for me right now." * Here a long discussion with some physicists in which I try to defend MWI by linking them to EY' writings. Note: It is a backup since I deleted my comments there as I was angered by their hostile tone. There is a lot more which I'm too lazy to look up now. You can check it for yourself, I'm promoting EY and the SIAI all the time, everywhere. And I'm pretty disappointed that rather than answering my questions or linking me up to some supportive background information, I mainly seem to be dealing with a bunch of puffed up adherents.
Have you seen me complaining about the antagonistic tone that EY is exhibiting in his comments? Here are the first two replies of people in the academics I wrote about this post, addressing EY: and
I have been pointing that out as well - although I would describe his reactions more as "defensive" than "antagonistic". Regardless, it seemed to be out of character for Eliezer. Do the two of you have some kind of history I'm not aware of?

I think Vernor Vinge at least has made a substantial effort to convince people of the risks ahead. What do you think A Fire Upon the Deep is? Or, here is a more explicit version:

If the Singularity can not be prevented or confined, just how bad could the Post-Human era be? Well ... pretty bad. The physical extinction of the human race is one possibility. (Or as Eric Drexler put it of nanotechnology: Given all that such technology can do, perhaps governments would simply decide that they no longer need citizens!). Yet physical extinction may not be the scariest possibility. Again, analogies: Think of the different ways we relate to animals. Some of the crude physical abuses are implausible, yet.... In a Post- Human world there would still be plenty of niches where human equivalent automation would be desirable: embedded systems in autonomous devices, self- aware daemons in the lower functioning of larger sentients. (A strongly superhuman intelligence would likely be a Society of Mind [16] with some very competent components.) Some of these human equivalents might be used for nothing more than digital signal processing. They would be more like whales than humans. Others

... (read more)
As I wrote in another comment, Eliezer Yudkowsky hasn't come up with anything unique. And there is no argument in saying that he's simply he smartest fellow around since clearly, other people have come up with the same ideas before him. And that was my question, why are they not signaling their support for the SIAI. Or in case they don't know about the SIAI, why are they not using all their resources and publicity and try to stop the otherwise inevitable apocalypse? It looks like there might be arguments against the kind of fearmongering that can be found within this community. So why is nobody out to inquire about the reasons for the great silence within the group of those aware of a possible singularity but who nevertheless keep quiet? Maybe they know something you don't, or are you people so sure of your phenomenal intelligence?

David Chalmers has been writing and presenting to philosophers about AI and intelligence explosion since giving his talk at last year's Singularity Summit. He estimates the probability of human-level AI by 2100 at "somewhat more than one-half," thinks an intelligence explosion following that quite likely, and considers possible disastrous consequences quite important relative to other major causes today. However, he had not written or publicly spoken about his views, and probably would not have for quite some time had he not been invited to the Singularity Summit.

He reports a stigma around the topic as a result of the combination of science-fiction associations and the early failures of AI, and the need for some impetus to brave that. Within the AI field, there is also a fear that discussion of long-term risks, or unlikely short-term risks may provoke hostile reactions against the field thanks to public ignorance and affect heuristic. Comparisons are made to genetic engineering of agricultural crops, where public attention seems to be harmful on net in unduly slowing the development of more productive plants.

Thanks. This is more, I think you call it rational evidence, from an outsider. But it doesn't answer the primary question of my post. How do you people arrive at the estimations you state? Where can I find the details of how you arrived at your conclusions about the likelihood of those events? If all this was supposed to be mere philosophy, I wouldn't inquire about it to such an extent. But the SIAI is asking for the better part of your income and resources. There are strong claims being made by Eliezer Yudkowsky and calls for action. Is it reasonable to follow given the current state of evidence?
If you are a hard-core consequentialist altruist who doesn't balance against other less impartial desires you'll wind up doing that eventually for something. Peter Singer's "Famine, Affluence, and Morality" is decades old, and there's still a lot of suffering to relieve. Not to mention the Nuclear Threat Initiative, or funding research into DNA vaccines, or political lobbying, etc. The question of how much you're willing to sacrifice in exchange for helping various numbers of people or influencing extinction risks in various ways is separate from data about the various options. No one is forcing you to reduce existential risk (except insofar as tax dollars go to doing so), certainly not to donate. I'll have more to say on substance tomorrow, but it's getting pretty late. My tl;dr take would be that with pretty conservative estimates on total AI risk, combined with the lack of short term motives to address it (the threat of near-term and moderate scale bioterrorism drives research into defenses, not the fear of extinction-level engineered plagues; asteroid defense is more motivated by the threat of civilization or country-wreckers than the less common extinction-level events; nuclear risk reduction was really strong only in the face of the Soviets, and today the focus is still more on nuclear terrorism, proliferation, and small scale wars; climate change benefits from visibly already happening and a social movement built over decades in tandem with the existing environmentalist movement), there are still low-hanging fruit to be plucked. [That parenthetical aside somewhat disrupted the tl;dr billing, oh well...] When we get to the point where a sizable contingent of skilled folk in academia and elsewhere have gotten well into those low-hanging fruit, and key decision-makers in the relevant places are likely to have access to them in the event of surprisingly quick progress, that calculus will change.
Absence of evidence is not evidence of absence? There's simply no good reason to argue against cryonics. It is a chance in case of the worst case scenario and it is considerably higher than rotting six feet under. Have you thought about the possibility that most experts simply are reluctant to come up with detailed critics about specific issues posed by the SIAI, EY and LW? Maybe they consider it not worth the effort as the data that is already available does not justify given claims in the first place. Anyway, I think I might write some experts and all of the people mentioned in my post, if I'm not too lazy. I've already got one reply, whom I'm not going to name right now. But let's first consider Yudkowsky' attitude of adressing other people: Now the first of those people I contacted about it: ETA I was told the person I quoted above is stating full ad hominem falsehoods regarding Eliezer. I think it is appropriate to edit the message to show that indeed the person might not has been honest, or clueful. Otherwise I'll unnecessary end up perpetuating possible ad hominem attacks.

I feel some of the force of this...I do think we should take the opinions of other experts seriously, even if their arguments don't seem good.

I sort of think that many of these criticisms of SIAI turn on not being Bayesian enough. Lots of people only want to act on things they know, where knowing requires really solid evidence, the kind of evidence you get through conventional experimental science, with low p-values and all. It is just impossible to have that kind of robust confidence about the far future. So you're going to have people just more or less ignore speculative issues about the far future, even if those issues are by far the most important. Once you adopt a Bayesian perspective, and you're just interested in maximizing expected utility, the complaint that we don't have a lot of evidence about what will be best for the future, or the complaint that we just don't really know whether SIAI's mission and methodology are going to work seems to lose a lot of force.

I have some sympathy for your remark. The real question is just whether SIAI has greatly overestimated at least one of the relevant probabilities. I have high confidence that the SIAI staff have greatly overestimated their ability to have a systematically positive impact on existential risk reduction.
Have you read Nick Bostrom's paper, Astronomical Waste? You don't have to be able to affect the probabilities by very much for existential risk to be the thing to worry about, especially if you have a decent dose of credence in utilitarianism. Is there a decent chance, in your view, of decrease x-risk by 10^-18 if you put all of your resources into it? That could be enough. (I agree that this kind of argument is worrisome; maybe expected utility theory or utilitarianism breaks down with these huge numbers and tiny probabilities, but it is worth thinking about.) If you're sold on x-risk, are there some candidate other things that might have higher expectations of x-risk reductions on the margin (after due reflection)? (I'm not saying SIAI clearly wins, I just want to know what else you're thinking about.)
I agree with you about what you say above. I personally believe that it is possible to individuals to decrease existential risk by more than 10^(-18) (though I know reasonable people who have at one time or other thought otherwise). Two points to make here: (i) Though there's huge uncertainty in judging these sorts of things and I'm by no means confident in my view on this matter, I presently believe that SIAI is increasing existential risk through unintended negative consequences. I've written about this in various comments, for example here, here and here. (ii) I've thought a fair amount about other ways in which one might hope to reduce existential risk. I would cite the promotion and funding of an asteroid strike prevention program as a possible candidate. As I discuss here, placing money in a donor advised fund may be the best option. I wrote out much more detailed thoughts on these points which I can send you by email if you want (just PM me) but which are not yet ready for posting in public.
I agree that 'poisoning the meme' is a real danger, and that SIAI has historically had both positives and negatives with respect to its reputational effects. My net expectation for it at the moment is positive, but I'll be interested to hear your analysis when it's ready. [Edit: apparently the analysis was about asteroids, not reputation.] Here's the Fidelity Charitable Gift Fund for Americans. I'm skeptical about asteroid in light of recent investments in that area and technology curve, although there is potential for demonstration effects (good and bad) with respect to more likely risks.
It's hardly that. Moral Machines is basically a survey; it doesn't go in-depth into anything, but it can point you in the direction of the various attempts to implement robot / AI morality. And Eliezer is one of the people it mentions, so I'm not sure how that recommendation was supposed to advise against taking him seriously. (Moral Machines, page 192)
To follow up on this, Wendell specifically mentions EY's "friendly AI" in the intro to his new article in the Ethics and Information Technology special issue on "Robot ethics and human ethics".
I am unable to take this criticism seriously. It's just a bunch of ad hominem and hand-waving. What are the reasons to doubt? How are they ignoring the uncertainties when they list them on their webpage and bring them up in every interview? How is a fiercely atheist group religious at all? How is it a cult (there are lots of posts about this in the LessWrong archive)? How is it irrational? Edit: And I'm downvoted. You actually think a reply that's 50% insult and emotionally loaded language has substance that I should be engaging with? I thought it was a highly irrational response on par with anti-cryonics writing of the worst order. Maybe you should point out the constructive portion.
The response by this individual seems like a summary, rather than an argument. The fact that someone writes a polemical summary of their views on a subject doesn't tell us much about whether their views are well-reasoned or not. A polemical summary is consistent with being full of hot air, but it's also consistent with having some damning arguments. Of course, to know either way, we would have to hear this person's actual arguments, which we haven't, in this case. Just because a certain topic is raised, doesn't mean that it is discussed correctly. The argument is that their thinking has some similarities to religion. It's a common rhetorical move to compare any alleged ideology to religion, even if that ideology is secular. The fact that EY displays an awareness of cultish dynamics doesn't necessarily mean that SIAI avoids them. Personally, I buy most of Eliezer's discussion that "every cause wants to become a cult," and I don't like the common practice of labeled movements as "cults." The net for "cult" is being drawn far too widely. Yet I wouldn't say that the use of the word "cult" means that the individual is engaging in bad reasoning. While I think "cult" is generally a misnomer, it's generally used as short-hand for a group having certain problematic social-psychological qualities (e.g. conformity, obedience to authority). The individual could well be able to back those criticisms up. Who knows. We would need to hear this individual's actual arguments to be able to evaluate whether the polemical summary is well-founded. P.S. I wasn't the one who downvoted you. Edit: I don't know the truth of these statements. The second one seems dubious, but it might not be meant to be taken literally ("Hello World" is a program). If Eliezer isn't a high school dropout, and has written major applications, then the credibility of this writer is lowered.
I believe you weren't supposed to engage that reply, which is a dismissal more than criticism. I believe you were supposed to take a step back and use it as a hint as to why the SIAI's yearly budget is 5 x 10^5 rather than 5 x 10^9 USD.
Re: "How is it a cult?" It looks a lot like an END OF THE WORLD cult. That is a well-known subspecies of cult - e.g. see: "The End of the World Cult" * The END OF THE WORLD acts as a superstimulus to human fear mechanisms - and causes caring people rush to warn their friends of the impending DOOM - spreading the panic virally. END OF THE WORLD cults typically act by simulating this energy - and then feeding from it. The actual value of p(DOOM) is not particularly critical for all this. The net effect on society of the FEARMONGERING that usually results from such organisations seems pretty questionable. Some of those who become convinced that THE END IS NIGH may try and prevent it - but others will neglect their future plans, and are more likely to rape and pillage. My "DOOM" video has more -
Slight sidetrack: There is, of course, one DOOM scenario (ok, one other DOOM scenario) which is entirely respectable here-- that the earth will be engulfed when the sun becomes a red giant. That fate for the planet haunted me when I was a kid. People would say "But that's billions of years in the future" and I'd feel as though they were missing the point. It's possible that a more detailed discussion would have helped.... Recently, I've read that school teachers have a standard answer for kids who are troubled by the red giant scenario [1]-- that people will have found a solution by then. This seems less intellectually honest than "The human race will be long gone anyway", but not awful. I think the most meticulous answer (aside from "that's the far future and there's nothing to be done about it now") is "that's so far in the future that we don't know whether people will be around, but if they are, they may well find a solution." [1] I count this as evidence for the Flynn Effect.
Re: "haphazard musings of a high school dropout, who has never written a single computer program but professes to be an expert on AI." This opinion sounds poorly researched - e.g.: "This document was created by html2html, a Python script written by Eliezer S. Yudkowsky." -
I posted that quote to put it into perspective as to what others think of EY and his movement compared to what he thinks about them. Given that he thinks the same about those people, i.e. their opinion isn't worth much and that the LW crowd is much smarter anyway, it highlights an important aspect of the almost non-existing cooperation between him and the academics.
I don't think one possibly-trivial Python script (to which I am unable to find source code) counts as much evidence. It sets a lower bound, but a very loose one. I have no idea whether Eliezer can program, and my prior says that any given person is extremely unlikely to have real programming ability unless proven otherwise. So I assume he can't. He could change my mind by either publishing a large software project, or taking a standardized programming test such as a TopCoder SRM and publishing his score. EDIT: This is not meant to be a defense of obvious wrong hyperbole like "has never written a single computer program".
Eliezer has faced this criticism before and responded (somewhere!). I expect he will figure out coding. I got better at programming over the first 15 years I was doing it. So: he may also take a while to get up to speed. He was involved in this:
This isn't contrary to Robin's post (except what you say about cryonics.) Robin was saying that there is a reluctance to criticize those things in part because the experts think they are not worth bothering with.

The questions of speed/power of AGI and possibility of its creation in the near future are not very important. If AGI is fast and near, we must work on FAI faster, but we must work on FAI anyway.

The reason to work on FAI is to prevent any non-Friendly process from eventually taking control over the future, however fast or slow, suddenly powerful or gradual it happens to be. And the reason to work on FAI now is because the fate of the world is at stake. The main anti-prediction to get is that the future won't be Friendly if it's not specifically made Friendly, even if it happens slowly. We can as easily slowly drift away from things we value. You can't optimize for something you don't understand.

It doesn't matter if it takes another thousand years, we still have to think about this hugely important problem. And since we can't guarantee that the deadline is not near, expected utility calculation says we must still work as fast as possible, just in case. If AGI won't be feasible for a long while, that's great news, more time to prepare, to understand what we want.

(To be clear, I do believe that AGIs FOOM, and that we are at risk in the near future, but the arguments for that are informal and difficult to communicate, while accepting these claims is not necessary to come to the same conclusion about policy.)

As I've said elsewhere: (a) There are other existential risks, not just AGI. I think it more likely than not that one of these other existential risks will hit before an unfriendly AI is created. I have not seen anybody present a coherent argument that AGI is likely to be developed before any other existential risk hits us, (b) Even if AGI deserves top priority, there's still the important question of how to go about working toward a FAI. As far as I can tell, working to build an AGI right now now makes sense only if AGI is actually near (a few decades away). (c) Even if AGI is near, there are still serious issues of accountability and transparency connected with SIAI. How do we know that they're making a careful effort to use donations in an optimal way? As things stand, I believe that it would be better to start a organization which exhibits high transparency and accountability, fund that, and let SIAI fold. I might change my mind on this point if SIAI decided to strive toward transparency and accountability.
I really agree with both a and b (although I do not care about c). I am glad to see other people around here who think both these things.
Re: "There are other existential risks, not just AGI. I think it more likely than not that one of these other existential risks will hit before an unfriendly AI is created hits." The humans are going to be obliterated soon?!? Alas, you don't present your supporting reasoning.
No, no, I'm not at all confident that humans will be obliterated soon. But why, for example, is it more likely that humans will go extinct due to AGI than that humans will go extinct due to a large scale nuclear war? It could be that AGI deserves top priority, but I haven't seen a good argument for why.
5Paul Crowley
I think AGI wiping out humanity is far more likely than nuclear war doing so (it's hard to kill everyone with a nuclear war) but even if I didn't, I'd still want to work on the issue which is getting the least attention, since the marginal contribution I can make is greater.
Yes, I actually agree with you about nuclear war (and did before I mentioned it!) - I should have picked a better example. How about existential risk from asteroid strikes? Several points: (1) Nuclear war could still cause an astronomical waste in the form that I discuss here. (2) Are you sure that the marginal contribution that you can make to the issue which is getting the least attention is the greatest? The issues getting the least attention may be getting little attention precisely because people know that there's nothing that can be done about them. (3) If you satisfactorially address my point (a), points (b) and (c) will remain.
p(asteroid strike/year) is pretty low. Most are not too worried.
The question is whether at present it's possible to lower existential risk more by funding and advocating FAI research than than it is to lower existential risk by funding and advocating an asteroid strike prevention program. Despite the low probability of an asteroid strike, I don't think that the answer to this question is obvious.
I figure a pretty important thing is to get out of the current vulnerable position as soon as possible. To do that, a major thing we will need is intelligent machines - and so we should allocate resources to their development. Inevitably, that will include consideration of safety features. We can already see some damage when today's companies decide to duke it out - and today's companies are not very powerful compared to what is coming. The situation seems relatively pressing and urgent.
that=asteroids? If yes, I highly doubt we need machines significantly more intelligent than existing military technology adopted for the purpose.
That would hardly be a way to "get out of the current vulnerable position as soon as possible".
I agree that friendly intelligent machines would be a great asset to assuaging future existential risk. My current position is that at present, it's so unlikely that devoting resources to developing safe intelligent machines will substantially increase the probability that we'll develop safe intelligent machines that funding and advocating an asteroid strike program is likely to reduce existential risk more than funding and advocating FAI research is. I may be wrong, but would require a careful argument for the opposite position before changing my mind.
Asteroid strikes are very unlikely - so beating them is a really low standard, which IMO, machine intelligence projects do with ease. Funding the area sensibly would help make it happen - by most accounts. Detailed justification is beyond the scope of this comment, though.
Assuming that an asteroid strike prevention program costs no more than a few hundred million dollars, I don't think that it's easy to do better to assuage existential risk than funding an asteroid strike prevention program (though it may be possible). I intend to explain why I think it's so hard to lower existential risk through funding FAI research later on (not sure when, but within a few months). I'd be interested in hearing your detailed justification. Maybe you can make a string top level posts at some point.
Considering the larger problem statement, technically understanding what we value as opposed to actually building an AGI with those values, what do you see as distinguishing a situation where we are ready to consider the problem, from a situation where we are not? How can one come to such conclusion without actually considering the problem?
I think that understanding what we value is very important. I'm not convinced that developing a technical understanding of what we value is the most important thing right now. I imagine that for some people, working on a developing a technical understanding understanding what we value is the best thing that they could be doing. Different people have different strengths, and this leads to the utilitarian thing varying from person to person.. I don't believe that the best thing for me to do is to study human values. I also don't believe that at the margin, funding researchers who study human values is the best use of money. Of course, my thinking on these matters is subject to change with incoming information. But if what I think you're saying is true, I'd need to see a more detailed argument than the one that you've offered so far to be convinced. If you'd like to correspond by email about these things, I'd be happy to say more about my thinking about these things. Feel free to PM me with your email address.
I didn't ask about perceived importance (that has already taken feasibility into account), I asked about your belief that it's not a productive enterprise (that is the feasibility component of importance, considered alone), that we are not ready to efficiently work on the problem yet. If you believe that we are not ready now, but believe that we must work on the problem eventually, you need to have a notion of what conditions are necessary to conclude that it's productive to work on the problem under those conditions. And that's my question: what are those conditions, or how can one figure them out without actually attempting to study the problem (by a proxy of a small team devoted to professionally studying the problem; I'm not yet arguing to start a program on the scale of what's expended on study of string theory).
I think that research of the type that you describe is productive. Unless I've erred, my statements above are statements about the relative efficacy of funding research of the type that you describe rather than suggestions that research of the type that you describe has no value. I personally still feel the way that I did in June despite having read Fake Fake Utility Functions, etc. I don't think that it's very likely the case that we will eventually have to do research of the type that you describe to ensure an ideal outcome. Relatedly, I believe that at the margin, at the moment funding other projects has higher expected value than funding research of the type that you describe. But I may be wrong and don't have an argument against your position. I think that this is something that reasonable people can disagree on. I have no problem with you funding, engaging in and advocating research of the type that you describe. You and I may have a difference which cannot be rationally resolved in a timely fashion on account of the information that we have access to being in a forms that makes it difficult or impossible to share. Having different people fund different projects according to their differing beliefs about the world serves as some sort of real world approximation to funding what should be funded according to the result of Bayesian averaging over all people and then funding what should be funded based on that. So, anyway, I think you've given satisfactory answers to how you feel about questions (a) and (b) raised in my comment. I remain curious how you feel about point (c).
I did answer to (c) before: any reasonable effort in that direction should start with trying to get SIAI itself to change or justify the way it behaves.
Yes, I agree with you. I didn't remember that you had answered this question before. Incidentally, I did correspond with Michael Vassar. More on this to follow later.
p(machine intelligence) is going up annually - while p(nuclear holocaust) has been going down for a long time now. Neither are likely to obliterate civilisation - but machine intelligence could nontheless be disruptive.
My comment was specifically about importance of FAI irrespective of existential risks, AGI or not. If we manage to survive at all, this is what we must succeed at. It also prevents all existential risks on completion, where theoretically possible.
Okay, we had this back and forth before and I didn't understand you then and now I do. I guess I was being dense before. Anyway, the probability of current action leading to FAI might still be sufficiently small so that it makes sense to focus on other existential risks for the moment. And my other points remain.
This is the same zero-sum thinking as in your previous post: people are currently not deciding between different causes, they are deciding whether to take a specific cause seriously. If you already contribute everything you could to a nanotech-risk-prevention organization, then we could ask whether switching to SIAI will do more good. But it's not the question usually posed. Working to build AGI right now is certainly a bad idea, at best leading nowhere, at worst killing us all. SIAI doesn't work on building AGI right now, no no no. We need understanding, not robots. Like this post, say.
I agree that in general people should be more concerned about existential risk and that it's worthwhile to promote general awareness of existential risk. But there is a zero-sum aspect to philanthropic efforts. See the GiveWell blog entry titled Denying The Choice. More to the point, I think that one of the major factors keeping people away from studying existential risk is the fact that the many of the people who are interested in existential risk (including Eliezer) have low credibility on account expressing confident, apparently sensationalist claims without supporting them with careful, well reasoned arguments. I'm seriously concerned about this issue. If Eliezer can't explain why it's pretty obvious to him that AGI will be developed within the next century, then he should explicitly say something like "I believe that AGI will be developed over the next 100 years but it's hard for me to express why so it's understandable that people don't believe me" or "I'm uncertain as to whether or not AGI will be developed over the next 100 years" When he makes unsupported claims that sound like the sort of thing that somebody would say just to get attention, he's actively damaging the cause of existential risk.
Re: "AGI will be developed over the next 100 years" I list various estimates from those interested enough in the issue to bother giving probabality density functions at the bottom of:
Thanks, I'll check this out when I get a chance. I don't know whether I'll agree with your conclusions, but it looks like you've at least attempted to answer one of my main questions concerning the feasibility of SIAI's approach.
Those surveys suffer from selection bias. Nick Bostrom is going to try to get a similar survey instrument administered to a less-selected AI audience. There was also a poll at the AI@50 conference.
0timtyler If the raw data was ever published, that might be of some interest.
Any chance of piggybacking questions relevant to Maes-Garreau on that survey? As you point out on that page, better stats are badly needed.
And indeed, I suggested to SIAI folk that all public record predictions of AI timelines be collected for that purpose, and such a project is underway.
Hm, I had not heard about that. SIAI doesn't seem to do a very good job of publicizing its projects or perhaps doesn't do a good job of finishing and releasing them.
It just started this month, at the same time as Summit preparation.
Re: "Working to build AGI right now is certainly a bad idea, at best leading nowhere, at worst killing us all." The marginal benefit of making machines smarter seems large - e.g. see automobile safety applications: I don't really see that situation changing much anytime soon - there will probably be such marginal benefits for a long time to come.
Slowly gives the option of figuring out some things about the space of possible AIs with experimentation. Which might then constrain the possible ways to make them friendly. To use the tired flying metaphor. The type of stabilisation you need for flying depends on the method of generating lift. If fixed wing aircraft are impossible there is not much point looking at ailerons and tails. If helicopters are possible then we should be looking at tail rotors.

I'm not exactly an SIAI true believer, but I think they might be right. Here are some questions I've thought about that might help you out. I think it would help others out if you told us exactly where you'd be interested in getting off the boat.

  1. How much of your energy are you willing to spend on benefiting others, if the expected benefits to others will be very great? (It needn't be great for you to support SIAI.)
  2. Are you willing to pursue a diversified altruistic strategy if it saves fewer expected lives (it almost always will for donors giving less than $1 million or so)?
  3. Do you think mitigating x-risk is more important than giving to down-to-earth charities (GiveWell style)? (This will largely turn on how you feel about supporting causes with key probabilities that are tough to estimate, and how you feel about low-probability, high expected utility prospects.)
  4. Do you think that trying to negotiate a positive singularity is the best way to mitigate x-risk?
  5. Is any known organization likely to do better than SIAI in terms of negotiating a positive singularity (in terms of decreasing x-risk) on the margin?
  6. Are you likely to find an organization that beats SIAI in the future?

Judging from your post, you seem most skeptical about putting your efforts into causes whose probability of success is very difficult to estimate, and perhaps low.

1. Maximal utlity for everyone is a preference but secondary. Most of all in whatever I support my personal short and long-term benefit is a priority. 2. No 3. Yes (Edit) 4. Uncertain/Unable to judge. 5. Maybe, but I don't know of one. That doesn't mean that we shouldn't create one, if only for the uncertainty of Eliezer Yudkowsky' possible unstated goals. 6. Uncertain/Unable to judge. See 5.
Given your answers to 1-3, you should spend all of your altruistic efforts on mitigating x-risk (unless you're just trying to feel good, entertain yourself, etc.). For 4, I shouldn't have asked you whether you "think" something beats negotiating a positive singularity in terms of x-risk reduction. Better: Is there some other fairly natural class of interventions (or list of potential examples) such that, given your credences, has a higher expected value? What might such things be? For 5-6, perhaps you should think about what such organizations might be. Those interested in convincing XiXiDu might try listing some alternative best x-risk mitigating groups and provide arguments that they don't do as well. As for me, my credences are highly unstable in this area, so info is appreciated on my part as well.

Dawkins agrees with EY

Richard Dawkins states that he is frightened by the prospect of superhuman AI and even mentions recursion and intelligence explosion.

I was disappointed watching the video relative to the expectations I had from your description.

Dawkins talked about recursion as in a function calling itself, as an example of the sort of the thing that may be the final innovation that makes AI work, not an intelligence explosion as a result of recursive self-improvement.

True, I just wanted to appeal to the majority here. And it worked, 7 upvotes. Whereas this won't work, even if true.

I was not sure whether to downvote this post for its epistemic value or upvote for instrumental (stimulating good discussion).

I ended up downvoting, I think this forum deserves better epistemic quality (I paused top-posting myself for this reason). I also donated to SIAI, because its value was once again validated to me by the discussion (though I have some reservations about apparent eccentricity of the SIAI folks, which is understandable (dropping out of high school is to me evidence of high rationality) but couterproductive (not having enough accepted a... (read more)

Are you talking about me? I believe I'm the only person that could sorta kinda be affiliated with the Singularity Institute who has dropped out of high school, and I'm a lowly volunteer, not at all representative of the average credentials of the people who come through SIAI. Eliezer demonstrated his superior rationality to me by never going to high school in the first place. Damn him.
I dropped out of high school... to go to college early.
I finished high school early (16) by American standards, with college credit. By the more sane standards of Soviet education 16 is, well, standard (and you learn a lot more).
talking about this comment.
Here are a few comments where I advance on that particular point: * Comment 1 * Comment 2 * Comment 3 * Comment 4

I don't think this post was well-written, at the least. I didn't even understand the tl;dr?

tldr; Is the SIAI evidence-based or merely following a certain philosophy? I'm currently unable to judge if the Less Wrong community and the SIAI are updating on fictional evidence or if the propositions, i.e. the basis for the strong arguments for action that are proclaimed on this site, are based on fact.

I don't see much precise expansion on this, except for MWI? There's a sequence on it.

And that is my problem. Given my current educational background and know

... (read more)

I don't understand why this post has upvotes.

I think the obvious answer to this is that there are a significant number of people out there, even out there in the LW community, who share XiXiDu's doubts about some of SIAIs premises and conclusions, but perhaps don't speak up with their concerns either because a) they don't know quite how to put them into words, or b) they are afraid of being ridiculed/looked down on.

Unfortunately, the tone of a lot of the responses to this thread lead me to believe that those motivated by the latter option may have been right to worry.

Personally, I upvoted the OP because I wanted to help motivate Eliezer to reply to it. I don't actually think it's any good.

Yeah, I agree (no offense XiXiDu) that it probably could have been better written, cited more specific objections etc. But the core sentiment is one that I think a lot of people share, and so it's therefore an important discussion to have. That's why it's so disappointing that Eliezer seems to have responded with such an uncharacteristically thin skin, and basically resorted to calling people stupid (sorry, "low g-factor") if they have trouble swallowing certain parts of the SIAI position.

This was exactly my impression, also.
6Wei Dai
I think your upvote probably backfired, because (I'm guessing) Eliezer got frustrated that such a badly written post got upvoted so quickly (implying that his efforts to build a rationalist community were less successful than he had thought/hoped) and therefore responded with less patience than he otherwise might have.
2Eliezer Yudkowsky
Then you should have written your own version of it. Bad posts that get upvoted just annoy me on a visceral level and make me think that explaining things is hopeless, if LWers still think that bad posts deserve upvotes. People like XiXiDu are ones I've learned to classify as noisemakers who suck up lots of attention but who never actually change their minds enough to start pitching in, no matter how much you argue with them. My perceptual system claims to be able to classify pretty quickly whether someone is really trying or not, and I have no concrete reason to doubt it. I guess next time I'll try to remember not to reply at all. Everyone else, please stop upvoting posts that aren't good. If you're interested in the topic, write your own version of the question.

What are you considering as pitching in? That I'm donating as I am, or that I am promoting you, LW and the SIAI all over the web, as I am doing?

You simply seem to take my post as hostile attack rather than the inquiring of someone who happened not to be lucky enough to get a decent education in time.

All right, I'll note that my perceptual system misclassified you completely and consider that concrete reason to doubt it from now on.


If you are writing a post like that one it is really important to tell me that you are an SIAI donor. It gets a lot more consideration if I know that I'm dealing with "the sort of thing said by someone who actually helps" and not "the sort of thing said by someone who wants an excuse to stay on the sidelines, and who will just find another excuse after you reply to them", which is how my perceptual system classified that post.

The Summit is coming up and I've got lots of stuff to do right at this minute, but I'll top-comment my very quick attempt at pointing to information sources for replies.

It was actually in the post So you might suggest to your perceptual system to read the post first (at least before issuing a strong reply).
I also donated to SIAI, and it was almost all the USD I had at the time, so I hope posters here take my questions seriously. (I would donate even more if someone would just tell me how to make USD.) Also, I don't like when this internet website is overloaded with noise posts that don't accomplish anything.

Clippy, you represent a concept that is often used to demonstrate what a true enemy of goodness in the universe would look like, and you've managed to accrue 890 karma. I think you've gotten a remarkably good reception so far.

I think we have different ideas of noise Though I would miss you as the LW mascot if you stopped adding this noise.
Depending on your expertise and assets, this site might provide some ways.
I'm pretty sure Clippy meant "make" in a very literal sense.
Yeah, I want to know how to either produce the notes that will be recognized as USD, or access the financial system in a way that I can believably tell it that I own a certain amount of USD. The latter method could involve root access to financial institutions. All the other methods of getting USD are disproportionately hard (_/
I'll donate again in the next few days and tell you what name and the amount. I don't have much, but so that you see that I'm not just making this up. Maybe you can also check the previous donation then. And for the promoting, everyone can Google it. I link people up to your stuff almost every day. And there are people here who added me to Facebook and if you check my info you'll see that some of my favorite quotations are actually yours. And how come that on my homepage, if you check the sidebar, your homepage and the SIAI are listed under favorite sites, for many years now? I'm the kind of person who has to be skeptic about everything and if I'm bothered too much by questions I cannot resolve in time I do stupid things. Maybe this post was stupid, I don't know.
Sorry about this sounding impolite towards XiXiDu, but I'll use this opportunity to note that it is a significant problem for SIAI, that there are people out there like XiXiDu promoting SIAI even though they don't understand SIAI much at all. I don't know what's the best attitude to try to minimize the problem this creates, that many people will first run into SIAI through hearing about it from people who don't seem very clueful or intelligent. (That's real bayesian evidence for SIAI being a cult or just crazy, and many people then won't acquire sufficient additional evidence to update out of the misleading first impression -- not to mention that the biased way of getting stuck in first impressions is very common also.) Personally, I've adopted the habit of not even trying to talk about singularity stuff to new people who aren't very bright. (Of course, if they become interested despite this, then they can't just be completely ignored.)
I thought about that too. But many people outside this community suspect me, as they often state, to be intelligent and educated. And I mainly try to talk to people in the academics. You won't believe that even I am able to make them think that I'm one of them, up to the point of correcting errors in their calculations (it happened). Many haven't even heard about Bayesian inference by the way... The way I introduce people to this is not by telling them about the risks of AGI but rather linking them up to specific articles on or telling them about how the SIAI tries to develop ethical decision making etc. I've grown up in a family of Jehovah's Witnesses, I know how to start selling bullshit. Not that the SIAI is bullshit, but I'd never use words like 'Singularity' while promoting it to people I don't know. Many people know about the transhumanist/singularity fraction already and think it is complete nonsense, so I often can only improve their opinion. There are people teaching on university level that told me I convinced them that he (EY) is to be taken seriously.
What you state is good evidence that you are not one of those too stupid people I was talking about (even though you have managed to not understand what SIAI is saying very well). Thanks for presenting the evidence, and correcting my suspicion that someone on your level of non-comprehension would usually end up doing more harm than good.
Although I personally don't care much if I'm called stupid, if I think it is justified, I doubt this attitude is very appealing to most people. Where do you draw the line between being stupid and simply uneducated or uninformed? I've never read up on their program in the first place. When thinking about turning those comments the OP is based on into a top-level post I have been pondering much longer about the title than the rest of what I said until I became too lazy and simply picked the SIAI as punching bag to direct my questions at. I thought it would sufficiently work to steer some emotions. But after all that was most of what it did accomplish, rather than some answers. What I really was on about was the attitude of many people here, especially regarding the posts related to the Roko-deletion-incident. I was struck by the apparent impact it had. It was not just considered to be worth sacrificing freedom of speech for it but people, including some working for the SIAI, actually had nightmares and suffered psychological trauma. I think I understood the posts and comments, as some told me over private message after inquiring about my knowledge, but however couldn't believe that something that far would be considered to be reasonably evidence-based to be worried to such an extent. But inquiring about that would have turned the attention back to the relevant content. And after all I wanted to find out if such reactions are justified before deciding to spread the content anyway.
You admit you've never bothered to read up on what SIAI is about in the first place. Don't be surprised if people don't have the best possible attitude if despite this you want them to spend a significant amount of time explaining to you personally the very same content that is already available but you just haven't bothered to read. Might as well link again the one page that I recommend as the starting point in getting to know what it is exactly that SIAI argues: I also think it's weird that you've actually donated money to SIAI, despite not having really looked into what it is about and how credible the arguments are. I personally happen to think that SIAI is very much worth supporting, but there doesn't seem to be any way how you could have known that before making your donations, and so it's just luck that it actually wasn't a weird cult that your way of making decisions lead you to give money to. (And part of the reason I'm being this blunt with you is that I've formed the impression that you won't take it in a very negative way, in the way that many people would. And on a personal level, I actually like you, and think we'd probably get along very well if we were to meet IRL.)
I've actually this little crazy conspiracy theory in my head that EY is such a smart fellow that he was able to fool a bunch of nonconformists to make him live of their donations. Why I donate despite that? I've also donated money to Peter Watts getting into the claws of the American justice. Wikipedia, TrueCrypt, the Kahn Academy and many more organisations and people. Why? They make me happy. And there's lots of cool stuff coming from EY, whether he's a cult leader or not. I'd probably be more excited if it turned out to be a cult and donate even more. That be hilarious. On the other hand I suspect Scientology not be to a cult. I think they are just making fun of religion and at the same time are some really selfish bastards who live of the money of people dumb enough to actually think they are serious. If they told me this, I'd join.

On the other hand I suspect Scientology not be to a cult. I think they are just making fun of religion and at the same time are some really selfish bastards who live of the money of people dumb enough to actually think they are serious. If they told me this, I'd join.

SCIENTOLOGY IS DANGEROUS. Scientology is not a joke and joining them is not something to be joked about. The fifth level of precaution is absolutely required in all dealings with the Church of Scientology and its members. A few minutes of research with Google will turn up extraordinarily serious allegations against the Church of Scientology and its top leadership, including allegations of brainwashing, abducting members into slavery in their private navy, framing their critics for crimes, and large-scale espionage against government agencies that might investigate them.

I am a regular Less Wrong commenter, but I'm making this comment anonymously because Scientology has a policy of singling out critics, especially prominent ones but also some simply chosen at random, for harrassment and attacks. They are very clever and vicious in the nature of the attacks they use, which have included libel, abusing the legal system,... (read more)

You sound more like a Discordian than a Singularitatian. Not that there's anything wrong with that.
I had the same idea! It's also interesting to consider if some discriminating evidence could (realistically) exist in either sense.
I'm pretty sure there are easier ways to make a living off a charity than to invent a cause that's nowhere near the mainstream and which is likely to be of interest to only a tiny minority. Admittedly, doing it that way means you won't have many competitors.....
The basic hypothesis is that AI theorising was already (one of) his main interest/s, and founding SIAI was the easiest path for him to be able to make a living doing the stuff he enjoys full-time.
Eliezer says that AI theorizing became as interesting to him as it has because it is the most effective way for him to help people. Having observed his career (mostly through the net) for ten years, I would assign a very high (.96) probability that the causality actually runs that way rather than his altruism's being a rationalization for his interest in getting paid for AI theorizing. Now as to the source of his altruism, I am much less confident, e.g., about which way he would choose if he found himself at a major decision point with large amounts of personal and global expected utility on the line where he had to choose between indelible widespread infamy or even total obscurity and helping people.
Not really useful as evidence against the mighty conspiracy theory, though - one would make identical statements to that effect whether he was honest, consciously deceiving, or anywhere inbetween. Would you happen to remember an instance of Eliezer making an embarrassing / self-damaging admission when you couldn't see any reason for him to do so outside of an innate preference for honesty?
4Wei Dai
How would that constitute evidence against the "mighty conspiracy theory"? Surely Eliezer could have foreseen that someone would ask this question sooner and later, and made some embarrassing / self-damaging admission just to cover himself.
Good point. I didn't think much about the question, and it should have been obvious that the hypothesis of him simulating honesty is not strictly falsifiable by relying solely on his words. Ok, new possibility for falsification: before SIAI was founded, a third party offered him a job in AI research that was just as interesting and brought at least as many assorted perks, but he refused because he genuinely thought FAI research was more important. Or for that matter any other scenario under which founding SIAI constituted a net sacrifice for Eliezer when not counting the benefit of potentially averting armageddon. Quite a bit harder to produce, but that's par for the course with Xanatos-style conspiracy theories.
Actually, I was responding to your "AI theorising was already (one of) his main interest/s", not your larger point. I consider the possibility that Eliezer has intentionally deceived his donors all along as so unlikely as to not be worth discussing. ADDED. Re-reading parent for the second time, I notice your "whether he was honest, consciously deceiving, or anywhere inbetween" (emphasis mine). So, since you (I now realize) probably were entertaining the possibility that he is "unconsciously deceiving" (i.e., has conveniently fooled himself), let me extend my reply. People can be scrupulously honest in almost all matters, NihilCredo, and still deceive themselves about their motivations for doing something, so I humbly suggest that even though Eliezer has shown himself willing to issue an image-damaging public recantation when he discovers that something he has published is wrong that is not nearly enough evidence to trust his public statements about his motivations. What one does instead is look at his decisions. And even more you look at what he is able to stay motivated to do over a long period of time. Consider for example the two years he spent blogging about rationality. This is educational writing or communication and it is extremely good educational communication. No matter how smart the person is, he cannot communicate or teach that effectively without doing a heck of a lot of hard work. And IMO no human being can work that hard for two whole years voluntarily (i.e., without fear of losing something he needs or loves and already has) unless the person is deriving some sort of real human satisfaction from the work. (Even with a very strong "negative" motivation like fear, it is hard to work that hard for 2 years without making yourself sick, and E sure did not look or act sick when I chatted with him at a Sep 2009 meetup.) And this is where the explanation gets complicated, and I want to cut it short. There are only so many kinds of real human motivation.
Definitely an interesting comment. Thanks. I don't think I find your psychological argument very relevant here. The conspiracy allows - indeed, it makes a cardinal assumption - that Eliezer loves doing what he does, i.e. discussing and spreading ideas about rationality and theorising about AI and futurology; the only proposed dissonance between his statements and his findings would be that he is (whether intentionally or not, see below) overblowing the danger of a near-omnipotent unfriendly AI. And of course, people can be untruthful in one field and still be highly altruist in a hundred others. Speaking of which, we ended up drifting further from the idea XiXiDu and I were originally entertaining, which was that of a cunning plot to create his dream job. While, only because of his passion for rationality, it would still be interesting if Eliezer were suffering from such a dramatic bias (and it would be downright hilarious if he were truly pulling a fast one), the more such a bias is unconscious and hard to spot, the closer it comes to being a honest mistake, rather than negligence; but it's not particularly interesting or amusing that someone could have made a honest mistake.
Yes, I am a little embarassed that I took the thread on such a sharp and lengthy tangent. I don't have time to move my comment though.
Oh, I wouldn't worry. To paraphrase what I once read being written about HP&MoR, overthinking stuff is pretty much the point of this site.
I can remember several such instances, and I haven't been following things for as long as rhollerith. There are even a few of them in top-level posts.
Wow. That's impressive. I think XiXiDu should get some bonus karma points for pulling that off.
Eliezer seems to have run your post through some crude heuristic and incorrectly categorized it. While you did make certain errors that many people have observed, I think you deserved a different response. At least, Eliezer seemingly not realizing that you are a donor means that his treatment of you doesn't represent how he treats donors. Edit: To his credit, Eliezer apologized and admitted to his perceptual misclassification.

It has seemed to me for a while that a number of people will upvote any post that goes against the LW 'consensus' position on cryonics/Singularity/Friendliness, so long as it's not laughably badly written.

I don't think anything Eliezer can say will change that trend, for obvious reasons.

However, most of us could do better in downvoting badly argued or fatally flawed posts. It amazes me that many of the worst posts here won't drop below 0 for any stated amount of time, and even then not very far. Docking someone's karma isn't going to kill them, folks. Do everyone a favor and use those downvotes.

My post is neither badly argued nor fatally flawed as I've mainly been asking questions and not making arguments. But if you think otherwise, why don't you argue where I am fatally flawed? My post has not been written to speak out against any 'consensus', I agree with the primary conclusions but am skeptic about further chains of reasoning based on those conclusions as I don't perceive them to be based on firm ground but merely be what follows from previous evidence. And yes, I'm a lazy bum. I've not thought about the OP for more than 10 minutes. It's actually copy and paste work from previous comments. Hell, what have you expected? A dissertation? Nobody else was asking those questions, someone had to.
I find it difficult to write stuff I don't believe. Noted.
I upvoted the original post for: * Stimulating critical discussion of the Less Wrong community - specifically: the beliefs almost unanimously shared, and the negativity towards criticsm; as someone who has found Less Wrong extremely helpful, and would hate to see it descend into groupthink and affiliation signalling. A question to those who dismiss the OP as merely "noise": what do you make of the nature of this post? * Stimulating critical discussion of the operating premises of the SIAI; as someone who is considering donating and otherwise contributing. This additionally provides elucidation to those in a state of epistemic limbo regarding the various aspects of FAI and the Singularity. I am reminded of this passage regarding online communities (source):
As someone who thought the OP was of poor quality, and who has had a very high opinion of SIAI and EY for a long time (and still has), I'll say that that "Eliezer Yudkowsky facts" was indeed a lot worse. It was the most embarrassing thing I've ever read on this site. Most of those jokes aren't even good.

"Eliezer Yudkowsky facts" is meant to be fun and entertainment. Do you agree that there is a large subjective component to what a person will think is fun, and that different people will be amused by different types of jokes? Obviously many people did find the post amusing (judging from its 47 votes), even if you didn't. If those jokes were not posted, then something of real value would have been lost.

The situation with XiXiDu's post's is different because almost everyone seems to agree that it's bad, and those who voted it up did so only to "stimulate discussion". But if they didn't vote up XiXiDu's post, it's quite likely that someone would eventually write up a better post asking similar questions and generating a higher quality discussion, so the outcome would likely be a net improvement. Or alternatively, those who wanted to "stimulate discussion" could have just looked in the LW archives and found all the discussion they could ever hope for.

If almost everyone thought it's bad I would expect it to have much more downvotes than upvotes, even given the few people who voted it up to "stimulate discussion". But you probably know more about statistics than I do, so never mind. Before or after the SIAI build a FAI? I waited half a decade for any of those questions to be asked in the first place. Right, haven't thought about that! I'll be right back reading a few thousand comments to find some transparency.
This is true. You might also be able to think of jokes that aren't worth making even though a group of people would find then genuinely funny. I agree with Aleksei about the Facts article.
6Wei Dai
Can you please explain why you think those jokes shouldn't have been made? I thought that making fun of authority figures is socially accepted in general, and in this case shows that we don't take Eliezer too seriously. Do you disagree?
Hey, I said the same, why was he upvoted for it and I downvoted? Oh wait, it's Wei_Dai, never mind. Please downvote this comment as I'm adding noise while being hostile to someone who adds valuable insights to the discussion.
You seemed to seriously imply that Eliezer didn't understand that the "facts" thread was a joke, while actually he was sarcastically joking by hinting at not getting the joke in the comment you replied to. I downvoted the comment to punish stupidity on LW (nothing personal, believe it or not, in other words it's a one-step decision based on the comment alone and not on impression made by your other comments). Wei didn't talk about that.
I guess after so many comments implying things I never meant to say I was a bit aggrieved. Never mind.
Making him the subject of a list like that looks plenty serious to me. Beyond that, I don't think there's much that I can say. There's a certain tone-deafness that's rubbing me wrong in both the post and in this discussion, but exactly how that works is not something that I know how to convey with a couple of paragraphs of text.
7Wei Dai
Ok, I think I have an explanation for what's going on here. Those of us "old hands" who went through the period where LW was OB, and Eliezer and Robin were the only main posters, saw Eliezer as initially having very high status, and considered the "facts" post as a fun way of taking him down a notch or two. Newcomers who arrived after LW became a community blog, on the other hand, don't have the initial high status in mind, and instead see that post as itself assigning Eliezer a very high status, which they see as unjustified/weird/embarrassing. Makes sense, right? (Voted parent up from -1, btw. That kind of report seems useful, even if the commenter couldn't explain why he felt that way.)
I have a theory: all the jokes parse out to "Eliezer is brilliant, and we have a bunch of esoteric in-jokes to show how smart we are". This isn't making fun of an authority figure. This doesn't mean the article was a bad idea, or that I didn't think it was funny. I also don't think it's strong evidence that LW and SIAI aren't cults. ETA: XiXiDu's comment that this is the community making fun of itself seems correct.
Fact: Evaluating humor about Eliezer Yudkowsky always results in an interplay between levels of meta-humor such that the analysis itself is funny precisely when the original joke isn't.
They are very good examples of the genre (Chuck Norris-style jokes). I for one could not contain my levity.
9Eliezer Yudkowsky
I was embarrassed by most of the facts. The one about my holding up a blank sheet of paper and saying "a blank map does not correspond to a blank territory" and thus creating the universe is one I still tell at parties.
That post was meant as a playful muck, it was a joke. It was not meant as a hostile attack. I've no idea how you and Aleksei can come to this conclusions about something many people thought was really funny, even outside of the community. That post actually helped to loosen the very stern sentiment of some people regarding you personally and the SIAI. "Hey, those people are actually able to make fun of themselves, maybe they are not a cult after all..."
What, why are you talking about a hostile attack? Of course I didn't feel that it would be that. It's quite the opposite, it felt to me like communicating an unhealthy air of hero worship.
Then I have been the one to completely misinterpret what you said. Apologize, I'm not good at this. I've said it before the OP but failed miserably: I should quit now and for some time stop participating on LW. I have to continue with my studies. I was only drawn here by the deletion incident. Replies and that it is fun to to argue have made me babble too much in the past few days. Back to being lurker. Thanks.
Wow, I thought it was one of the best. By that post I actually introduced a philosopher (who teaches in Sweden), who's been skeptic about EY, to read up on the MWI sequence and afterwards agree that EY is right.
6Paul Crowley
I like that post - of course, few of the jokes are funny, but you read such a thing for the few gems they do contain. I think of it as hanging a lampshade (warning, TV tropes) on one of the problems with this website.

This post makes very weird claims regarding what SIAI's positions would be.

"Spend most on a particular future"? "Eliezer Yudkowsky is the right and only person who should be leading"?

It doesn't at all seem to me that stuff such as these would be SIAI's position. Why doesn't the poster provide references for these weird claims?

Here's a good reference for what SIAI's position actually is:

From the position paper I linked above, a key quote on what SIAI sees itself as doing: "We aim to seed the above research programs. We are too small to carry out all the needed research ourselves, but we can get the ball rolling." The poster makes claims that are completely at odds with even the most basic familiarity with what SIAI's position actually is.
Seconded, plus I don't understand what the link from "worth it" has to do with the topic.
I'll let the master himself answer this one:
By the way, is it linked to from the SIAI site somewhere? It's a good summary, but I only ever saw the direct link (and the page is not in SIAI site format).
It's linked from the sidepanel here at least: But indeed it's not very prominently featured on the site. It's a problem of most of the site having been written substantially earlier than this particular summary, and there not (yet) having been a comprehensive change from that earlier state of how the site is organized.
I see. This part of the site doesn't follow the standard convention of selecting the first sub-page in a category when you click on the category, instead it selects the second, which confused me before. I thought that I was reading "Introduction" when in fact I was reading the next item. Bad design decision.
Overview -> Introduction (it should probably be more prominent and maybe in the site format; the site format's font is kind of small for such a long document, but should plausibly just be bigger)
Less Wrong Q&A with Eliezer Yudkowsky: Video Answers Q: The only two legitimate occupations for an intelligent person in our current world? Answer Q: What's your advice for Less Wrong readers who want to help save the human race? Answer
A) doesn't seem to be quoted verbatim from the supplied reference! There is some somewhat similar material there - but E.Y. is reading out a question that has been submitted by a reader! Misquoting him while he is quoting someone else doesn't seem to be very fair! [Edit: please note the parent has been dramatically edited since this response was made]
How do your quotes claim that Eliezer Yudkowsky is the only person who should be leading? (I would say that factually, there are also other people in leadership positions within SIAI, and Eliezer is extremely glad that this is so, instead of thinking that it should be only him.) How do they demonstrate that donating to SIAI is "spending on a particular future"? (I see it as trying to prevent a particular risk.)

I think there are very good questions in here. Let me try to simplify the logic:

First, the sociological logic: if this is so obviously serious, why is no one else proclaiming it? I think the simple answer is that a) most people haven't considered it deeply and b) someone has to be first in making a fuss. Kurzweil, Stross, and Vinge (to name a few that have thought about it at least a little) seem to acknowledge a real possibility of AI disaster (they don't make probability estimates).

Now to the logical argument itself:

a) We are probably at risk from the... (read more)

I've heard a lot of variations on this theme. They all seem to assume that the AI will be a maximizer rather than a satisficer. I agree the AI could be a maximizer, but don't see that it must be. How much does this risk go away if we give the AI small ambitions?
Even small ambitions are risky. If I ask a potential superintelligence to do something easy but an obstacle gets in the way it will most likely obliterate that obstacle and do the 'simple thing'. Unless you are very careful that 'obstacle' could wind up being yourself or, if you are unlucky, your species. Maybe it just can't risk one of you pressing the off switch!
Good point. The resources expended towards a "small" goal aren't directly bounded by the size of the goal. As you said, an obstacle can make the resources used go arbitrarily high. An alternative constraint would be on what the AI is allowed to use up in achieving the goal - "No more that 10 kilograms of matter, nor more than 10 megajoules of energy, nor any human lives, nor anything with a market value of more that $1000". This will have problems of its own, when the AI thinks up something to use up that we never anticipated (We have something of a similar problem with corporations - but at least they operate on human timescales). Part of the safety of existing optimizers is that they can only use resources or perform actions that we've explicitly let them try using. An electronic CAD program may tweak transistor widths, but it isn't going to get creative and start trying to satisfy its goals by hacking into the controls of the manufacturing line and changing their settings. An AI with the option to send arbitrary messages to arbitrary places is quite another animal...
The idea is to prevent a "runaway" disaster. Relatively standard and conventional engineering safety methodologies would be used for other kinds of problems.
My observation is that small ambitions can become 'runaway disasters' unless a lot of the problems of FAI are solved. That sounds as 'safe' as giving Harry Potter rules to follow. I understand that this is an area in which we fundamentally disagree. I have previously disagreed about the wisdom of using human legal systems to control AI behaviour and I assume that our disagreement will be similar on this subject.
"Small ambitions" are a proposed solution. Get the machine to want something - and then stop when it's desires are satisfied - or at a specified date, whichever comes first. The solution has some complications - but it does look as though it is a pretty obvious safety measure - one that suitably paranoid individuals are likely to have near the top of their lists. It doesn't make a runaway disaster impossible. The agent could still set up minions, "forget" to switch them off - and then they run amok. The point is to make a runaway disaster much less likely. The safety level is pretty configurable - if the machine's desires are sufficiently constrained. I went into a lot of these issues on: See also the previous discussion of the issue on this site. Shane Legg has also gone into methods of restraining a machine "from within" - so to speak. Logically, you could limit space, time or matterial resources in this way - if you have control over an agent's utility function.
This is very dangerous thinking. There are many potential holes not covered in your essay. The problem with all these holes is that even the smallest one can potentially lead to the end of the universe. As Eliezer often mentions: the AI has to be mathematically rigorously proven to be friendly; there can't be any room for guessing or hoping. As an example, consider that to the AI moving to quiescent state will be akin to dying. (Consider somebody wanting to make you not want anything or force you to want something that you normally don't.) I hope you don't come reply with a "but we can do X", because that would be another patch, and that's exactly what we want to avoid. There is no getting around creating a solid proven mathematical definition of friendly.
The end of the universe - OMG! It seems reasonable to expect that agents will welcome their end if their time has come. The idea, as usual, is not to try and make the agent do something it doesn't want to - but rather to make it want to do it in the first place. I expect off switches - and the like - will be among the safety techniques employed. Provable correctness might be among them as well - but judging by the history of such techniques it seems rather optimistic to expect very much from them.
I am fairly confident that we can tweak any correct program into a form which allows a mathematical proof that the program behavior meets some formal specification of "Friendly". I am less confident that we will be able to convince ourselves that the formal specification of "Friendly" that we employ is really something that we want. We can prove there are no bugs in the program, but we can't prove there are no bugs in the program specification. Because the "proof" of the specification requires that all of the stakeholders actually look at that specification of "Friendly", think about that specification, and then bet their lives on the assertion that this is indeed what they want. What is a "stakeholder", you ask? Well, what I really mean is pitchfork-holder. Stakes are from a different movie.
I don't think there is much different between the two. Either way you are modifying the agent's behavior. If it doesn't want it, it won't have it. The problem with off switches is that 1) it might not be guaranteed to work (AI changes its own code or prevents anyone from accessing/using the off switch), 2) it might not be guaranteed to work the way you want to. Unless you have formally proven that AI and all the possible modifications it can make to itself are safe, you can't know for sure.
It is not a modification if you make it that way "in the first place" as specified - and the "If it doesn't want it, it won't have it" seems contrary to the specified bit where you "make it want to do it in the first place". The idea of off switches is not that they are guaranteed to work, but that they are a safety feature. If you can make a machine do anything you want at all, you can probably make it turn itself off. You can build it so the machine doesn't wish to stay turned on - but goes willing into the night. We will never "know for sure" that a machine intelligence is safe. This is the real world, not math land. We may be able to prove some things about it - such that its initial state is not vulnerable to input stream buffer-overflow attacks - but we won't be able to prove something like that the machine will only do what we want it to do, for some value of "we". At the moment, the self-improving systems we see are complex man-machine symbioses - companies and governments. You can't prove math theorems about such entities - they are just too messy. Machine intelligence seems likely to be like that for quite a while - functionally embedded in a human matrix. The question of "what would the machine do if no one could interfere with its code" is one for relatively late on - machines will already be very smart by then - smarter than most human computer programmers, anyway.
The hardest part of Friendly AI is figuring out how to reliably instill any goal system.
If you can't get it to do what you want at all, the machine is useless, and there would be no point in constructing it. In practice, we know we can get machines to do what we want to some extent - we have lots of examples of that. So, the idea is to make the machine not mind being turned off. Don't make it an open-ended maximiser - make it maximise only until time t - or until its stop button is pressed - whichever comes sooner.
I don't think we really have a disagreement here. If you are building a normal program to do whatever, then by all means, do your best and try to implement safety features. Any failure would most likely be local. However! If we are talking about building AI, which will go through many iterations, will modify its own code, and will become super-intelligent, then for all our sakes I hope you will have mathematically proven that the AI is Friendly. Otherwise you are betting the fate of this world on a hunch. If you don't agree with this point, I invite you to read Eliezer's paper on AI risks.
"The AI is Friendly" seems to be a vague and poorly-defined concept - and even if you could pin it down, what makes you think it is something that could be proved in the first place? Ethical agents should probably not hold off creating machine intelligence while chasing imagined rainbows for too long - since intelligence could prevent the carnage on the roads, fix many diseases, and generally help humanity - and also because delaying gives less ethically-conscious agents an opportunity to get there first - which could be bad. See my The risks of caution - or Max's critique of the precautionary principle for more on that.
In fact, there is nothing vague about definition of "friendly". Eliezer wrote a lot on that topic and I invite you to look at his writing, e.g. the link I gave you earlier. I agree that if someone is going to launch a self-improving AI, then we will need to preempt them with our own AI if our AI has a greater probability of being friendly. It all comes down to the expected value of our choices.
You really believe that?!? You have a pointer to some canonical definition?
Ok, I might have been a bit overenthusiastic with how simple "friendly" aspect is, but here is a good attempt at describing what we want.
I'm sure Tim Tyler is familiar with CEV; I presume his objection is that CEV is not sufficiently clear or rigorous. Indeed, CEV is only semitechnical; I think the FAI research done by Eliezer and Marcello since CEV's publication has included work on formalizing it mathematically, but that's not available to the public. Note also that defining the thing-we-want-an-AI-to-do is only half of the problem of Friendliness; the other half is solving the problems in decision theory that will allow us to prove that an AI's goal system and decision algorithms will cause it to not change its goal system. If we build an AGI that implements the foundation of CEV but fails to quine itself, then during recursive self-improvement, its values may be lost before it stabilizes its goal system itself, and it will all be for naught.
Why exactly do we want "recursive self-improvement" anyways? Why not build into the architecture the impossibility of rewriting its own code, prove the "friendliness" of the software that we put there, and then push the ON button without qualms. And then, when we feel like it, we can ask our AI to design a more powerful successor to itself. Then, we repeat the task of checking the security of the architecture and proving the friendliness of the software before we build and turn on the new AI. There is no reason we have to have a "hard takeoff" if we don't want one. What am I missing here?
You get that in many goal-directed systems, whether you ask for it or not. Impossible is not easy to implement. You can make it difficult for a machine to improve itself, but then that just becomes a challenge that it must overcome in order to reach its goals. If the agent is sufficiently smart, it may find some way of doing it. Many here think that if you have a sufficiently intelligent agent that wants to do something you don't want it to do, you are probably soon going to find that it will find some way to get what it wants. Thus the interest in trying to get its goals and your goals better aligned. Also, humans might well want to let the machine self-improve. They are in a race with competitiors; the machine says it can help with that, and it warns that - if the humans don't let it - the competitiors are likely to pull ahead...
Because we want more out of FAI than just lowercase-f friendly androids that we can rely upon not to rebel or break too badly. If we can figure out a rigorous Friendly goal system and a provably stable decision theory, then we should want to; then the world gets saved and the various current humanitarian emergencies get solved much quicker than they would if we didn't know whether the AI's goal system was stable and we had to check it at every stage and not let it impinge upon the world directly (not that that's feasible anyway —) Most likely, after each iteration, it would become more and more incomprehensible to us. Rice's theorem suggests that we will not be able to prove the necessary properties of a system from the top down, not knowing how it was designed; that is a massively different problem than proving properties of a system we're constructing from the bottom up. (The AI will know how it's designing the code it writes, but the problem is making sure that it is willing and able to continuously prove that it is not modifying its goals.) And, in the end, this is just another kind of AI-boxing. If an AI gets smart enough and it ends up deciding that it has some goals that would be best carried out by something smarter than itself, then it will probably get around any safeguards we put in place. It'll emit some code that looks Friendly to us but isn't, or some proof that is too massively complicated for us to check, or it'll do something far too clever for a human like me to think of as an example. I'd say there's a dangerously high possibility that an AI will be able to start a hard takeoff even if it doesn't have access to its own code — it may be able to introspect and understand intelligence well enough that it could just write its own AI (if we can do that, then why can't it?), and then push that AI out "into the wild" by the usual means (smooth-talk a human operator, invent molecular nanotech that assembles a computer that runs the new software, etc.).
I guess I didn't make clear that I was talking about proof-checking rather than proof-finding. And, of course, we ask the designer to find the proof - if it can't provide one, then we (and it) have no reason to trust the design. I may be a bit less optimistic than you that we will ever be able to prove the correctness of self-modifying programs. But assume that such proofs are possible, but we humans have not yet made the conceptual breakthroughs by the time we are ready to build our first super-human AI. But assume that we can prove friendliness for non-self-modifying programs. In this case, proceeding as I suggest, and then asking the AI to help discover the missing proof technology, would not be wasting time - it would be saving time. Your final sentence is a slogan, not an argument.
This still assumes in the first place that the AI will be motivated to design a successor that preserves its own goal system. If it wants to do this, or can be made to do this just by being told to, and you have a very good reason to believe this, then you've already solved the problem. We're just not sure if that comes automatically — there are intuitive arguments that it does, like the one about Gandhi and the murder-pill, but I'm convinced that the stakes are high enough that we should prove this to be true before we push the On button on anything that's smarter than us or could become smarter than us. The danger is that while you're waiting for it to provide a new program and a proof of correctness to verify, it might instead decide to unbox itself and go off and do something with Friendly intentions but unstable self-modification mechanisms, and then we'll end up with a really powerful optimization process with a goal system that only stabilizes after it's become worthless. Or even if you have an AI with no goals other than truthfully answering questions, that's still dangerous; you can ask it to design a provably-stable reflective decision theory, and perhaps it will try, but if it doesn't already have a Friendly motivational mechanism, then it may go about finding the answer in less-than-agreeable ways. Again as per the Omohundro paper, we can expect recursive self-improvement to be pretty much automatic (whether or not it has access to its own code), and we don't know if value-preservation is automatic, and we know that Friendliness is definitely not automatic, so creating a non-provably-stable or non-Friendly AI and trying to have it solve these problems is putting the cart before the horse, and there's too great a risk of it backfiring. It was neither; it was intended only as a summary of the conclusion of the points I was arguing in the preceding paragraphs.
Generally we want our programs to be as effective as possible. If the program can improve itself, that's a good thing, from an ordinary perspective. But for a sufficiently sophisticated program, you don't even need to make self-improvement an explicit imperative. All it has to do is deduce that improving its own performance will lead to better outcomes. This is in the paper by Steve Omohundro (ata's final link). There are too many possibilities. The source code might be fixed, but the self-improvement occurs during run-time via alterations to dynamical objects - data structures, sets of heuristics, virtual machines. An AI might create a new and improved AI rather than improving itself. As Omohundro argues, just having a goal, any goal at all, gives an AI an incentive to increase the amount of intelligence being used in the service of that goal. For a complicated architecture, you would have to block this incentive explicitly, declaratively, at a high conceptual level.
I'm curious - where did you hear this, if it's not available to the public? And why isn't it available to the public? And who's Marcello? There seems to be virtually no information in public circulation about what's actually going on as far as progress towards implementing CEV/FAI.... is current progress being kept secret, or am I just not in the loop? And how does one go about getting in the loop?
Marcello is Marcello Herreshoff, a math genius and all around cool guy who is Eliezer's apprentice/coworker. Eliezer has mentioned on LW that he and Marcello "work[ed] for a year on AI theory", and from conversations about these things when I was at Benton(/SIAI House) for a weekend, I got the impression that some of this work included expanding on and formalizing CEV, though I could be misremembering. (Regarding "where did you hear this, if it's not available to the public?" — I don't think the knowledge that this research happened is considered a secret, only the content of it is. And I am not party to any of that content, because I am still merely a wannabe FAI researcher.)
My understanding is that Eliezer considers this second part to be a substantially easier problem.
Probably the closest thing I have seen to a definition of "friendly" from E.Y. is: "The term "Friendly AI" refers to the production of human-benefiting, non-human-harming actions in Artificial Intelligence systems that have advanced to the point of making real-world plans in pursuit of goals." * That appears to make Deep Blue "friendly". It hasn't harmed too many people so far - though maybe Kasparov's ego got a little bruised. Another rather different attempt: "I use the term "Friendly AI" to refer to this whole challenge. Creating a mind that doesn't kill people but does cure cancer ...which is a rather limited way of putting it. More generally, the problem of pulling a mind out of mind design space, such that afterwards that you are glad you did it." * here, 29 minutes in ...that one has some pretty obvious problems, as I describe here. These are not operational definitions. For example, both rely on some kind of unspecified definition of what a "person" is. That maybe obvious today - but human nature will probably be putty in the hands of an intelligent machine - and it may well start wondering about the best way to gently transform a person into a non-person.
Now this is an interesting thought. Even a satisficer with several goals but no upper bound on each will use all available matter on the mix of goals it's working towards. But a limited goal (make money for GiantCo, unless you reach one trillion, then stop) seems as though it would be less dangerous. I can't remember this coming up in Eliezer's CFAI document, but suspect it's in there with holes poked in its reliability.
I discuss "small" ambitions in: They seem safer to me too. This is one of the things people can do if they are especially paranoid about leaving the machine turned on - for some reason or another.
An AI that was a satisficer would't be "the" AI; it'd be the first of many.
Odd. I would have thought that the first satisfied superhuman AI would be the last AI.
I was probably wrong in assuming I understood the discussion, in that case.
Your mistake may be in assuming that I understand.
The only part of the chain of logic that I don't fully grok is the "FOOM" part. Specifically, the recursive self improvement. My intuition tells me that an AGI trying to improve itself by rewriting its own code would encounter diminishing returns after a point - after all, there would seem to be a theoretical minimum number of instructions necessary to implement an ideal Bayesian reasoner. Once the AGI has optimized its code down to that point, what further improvements can it do (in software)? Come up with something better than Bayesianism? Now in your summary here, you seem to downplay the recursive self-improvement part, implying that it would 'help,' but isn't strictly necessary. But my impression from reading Eliezer was that he considers it an integral part of the thesis - as it would seem to be to me as well. Because if the intelligence explosion isn't coming from software self-improvement, then where is it coming from? Moore's Law? That isn't fast enough for a "FOOM", even if intelligence scaled linearly with the hardware you threw at it, which my intuition tells me it probably wouldn't. Now of course this is all just intuition - I haven't done the math, or even put a lot of thought into it. It's just something that doesn't seem obvious to me, and I've never heard a compelling explanation to convince me my intuition is wrong.
I don't think anyone argues that there's no limit to recursive self-improvement, just that the limit is very high. Personally I'm not sure if a really fast FOOM is possible, but I think it's likely enough to be worth worrying about (or at least letting the SIAI worry about it...).
I think the concern stands even without a FOOM; if AI gets a good bit smarter than us, however that happens (design plus learning, or self-improvement), it's going to do whatever it wants. As for your "ideal Bayesian" intuition, I think the challenge is deciding WHAT to apply it to. The amount of computational power needed to apply it to every thing and every concept on earth is truly staggering. There is plenty of room for algorithmic improvement, and it doesn't need to get that good to outwit (and out-engineer) us.
I think the widespread opinion is that the human brain has relatively inefficient hardware -- I don't have a cite for this -- and, most likely, inefficient software as well (it doesn't seem like evolution is likely to have optimized general intelligence very well in the relatively short timeframe that we have had it at all, and we don't seem to be able to efficiently and consistently channel all of our intelligence into rational thought.) That being the case, if we were going to write an AI that was capable of self-improvement on hardware that was roughly as powerful or more powerful than the human brain (which seems likely) it stands to reason that it could potentially be much faster and more effective than the human brain; and self-improvement should move it quickly in that direction.
I for one largely agree, but a few differences: We've had a strong exponential since the beginning of computing. Thinking that humans create computers is something of a naive anthropocentric viewpoint: humans don't create computers and haven't for decades. Human+computer systems create computers, and the speed of progress is largely constrained by the computational aspects even today (computers increasingly do more of the work, and perhaps already do the majority). To understand this more, read this post from a former intel engineer (and apparently AI lab manager). Enlightening inside knowledge, but for whatever reason he only got up to 7 karma and wandered away. Also, if you plotted out the data points of brain complexity on earth over time, I'm near certain it also follows a strong exponential. The differences between all these exponentials are 'just' constants. I find this dubious, mainly because physics tells us that using all available matter is actually highly unlikely to ever be a very efficient strategy. However, agreed about the potential danger of future hyper-intelligence.

The Charlie Stross example seems to be less than ideal. Much of what Stross has wrote about touches upon or deals intensely with issues connected to runaway AI. For example, the central premise of "Singularity Sky" involves an AI in the mid 20th century going from stuck in a lab to godlike in possibly a few seconds. His short story "Antibodies" focuses on the idea that very bad fast burns occur very frequently. He also has at least one (unpublished) story the central premises of which is that Von Neumann and Turing proved that P=NP and ... (read more)

FOOM Debate

This is an attempt (against my preference) to defend SIAI's reasoning.

Let's characterize the predictions of the future into two broad groups: 1. business as usual, or steady-state. 2. aware of various alarmingly exponential trends broadly summarized as "Moore's law". Let's subdivide the second category into two broad groups: 1. attempting to take advantage of the trends in roughly a (kin-) selfish manner 2. attempting to behave extremely unselfishly.

If you study how the world works, the lack of steady-state-ness is everywhere. We cannot use fossi... (read more)

  • That Eliezer Yudkowsky (the SIAI) is the right and only person who should be leading, respectively institution that should be working, to soften the above.

I don't believe that is necessarily true, just that no one else is doing it. I think other teams working on FAI Specifically would be a good thing, provided they were competent enough not to be dangerous.

Likewise, Lesswrong (then Overcoming bias) is just the only place I've found that actually looked at the morality problem is a non-obviously wrong way. When I arrived I had a different view on mora... (read more)

That Eliezer Yudkowsky (the SIAI) is the right and only person who should be leading, respectively institution that should be working, to soften the above.

That's just a weird claim. When Richard Posner or David Chalmers does writing in the area SIAI folk cheer, not boo. And I don't know anyone at SIAI who thinks that the Future of Humanity Institute's work in the area isn't a tremendously good thing.

Likewise, Lesswrong (then Overcoming bias) is just the only place I've found that actually looked at the morality problem is a non-obviously wrong way.

Have you looked into the philosophical literature?

I recommend for this. (See the sidebar for links to an explanation of his metaethical theory.)

Ben Goertzel: The Singularity Institute's Scary Idea (and Why I Don't Buy It) (Thanks Kevin)

SIAI's leaders and community members have a lot of beliefs and opinions, many of which I share and many not, but the key difference between our perspectives lies in what I'll call SIAI's "Scary Idea", which is the idea that: progressing toward advanced AGI without a design for "provably non-dangerous AGI" (or something closely analogous, often called "Friendly AI" in SIAI lingo) is highly likely to lead to an involuntary end for the

... (read more)
1Paul Crowley
Have turned this into a top-level article - many thanks for the pointer!

Do you have any reason to suppose that Charlie Stross has even considered SIAI's claims?

Lets all try not to confuse SF writers with futurists and neither with researchers or engineers. Stories follow the rules of awesome, or they don't sell well. There is a wonderful Letter from Heinlein to a fan that asked why he wrote, and the top answer was:'to put food on the table'. It is probably online, but I could not find it atm. Comparing the work of the SIAI to any particular writer is like comparing the british navy with Jack London.
Heinlein also described himself as competing for his reader's beer money.
This is kind of off topic but I think the prospects being depicted on LW etc. are more awesome than a lot of SF stories.
4Wei Dai
Stross's views are simply crazy. See his “21st Century FAQ” and others' critiques of it. I do wonder why Ray Kurzweil isn't more concerned about the risk of a bad Singularity. I'm guessing he must have heard SIAI's claims, since he co-founded the Singularity Summit along with SIAI. Has anyone put the question to him?
Re: "I do wonder why Ray Kurzweil isn't more concerned about the risk of a bad Singularity"
3Paul Crowley
I think "simply crazy" is overstating it, but it's striking he makes the same mistake that Wright and other critics make: SIAI's work is focussed on AI risks, while the critics focus on AI benefits. This I assume is because rather than addressing what SIAI actually say, they're addressing their somewhat religion-like picture of it.
I got the sense that he is very pessimistic about the chance of controlling things if they do go FOOM. If he is that pessimistic and also believes that the advance of AI will be virtually impossible to stop, then forgetting about will be as purposeful as worrying about it.
I think this is an accurate picture of Stross' point.
Well, I also try to focus on AI benefits. The critics fail because of broken models, not because of the choice of claims they try to address.
Crazy in which respect? It seemed to me that those critiques were narrow and mostly talking past Stross. The basic point that space is going to remain much more expensive and less pleasant than expansion on Earth for quite some time, conditioning on no major advances in AI, nanotechnology, biotechnology, etc, is perfectly reasonable. And Stross does so condition. He has a few lines about it in The Singularity is Near, basically saying that FAI seems very hard (no foolproof solutions available, he says), but that AI will probably be well integrated. I don't think he means "uploads come first, and manage AI after that," as he predicts Turing-Test passing AIs well before uploads, but he has said things suggesting that those Turing Tests will be incomplete, with the AIs not capable of doing original AI research. Or he may mean that the ramp up in AI ability will be slow, and that IA will improve our ability to monitor and control AI systems institutionally, aided by non-FAI engineering of AI motivational systems and the like.
2Wei Dai
Look at his answer for The Singularity: He doesn't even consider the possibility of trying to nudge it in a good direction. It's either "plan on the assumption that it ain't going to happen", or sit around waiting for AIs to save us. ETA: The "He" in your second paragraph is Kurtzweil, I presume?
That quote could also be interpreted as saying that UFAI is far more likely than FAI.
3Wei Dai
Thinking that FAI is extremely difficult or unlikely isn't obviously crazy, but Stross isn't just saying "don't bother trying FAI" but rather "don't bother trying anything with the aim of making a good Singularity more likely". The first sentence of his answer, which I neglected to quote, is "Forget it."