"Unfortunately the universe doesn't agree with me. We'll see which one of us is still standing when this is over."
-- Eliezer Yudkowsky
"I am by heritage a Jew, by citizenship a Swiss, and by makeup a human being, and only a human being, without any special attachment to any state or national entity whatsoever."
Nationalism is a not far behind religion in the 'reasons to kill people' stakes, but receives a tiny fraction of the airtime.
I like the 'experience' and 'damage' one above. Implications for a future where suffering is all but eliminated?
Hope the move goes well Eliezer - keep up the great work.
Prof. Jaynes would doubtless be surprised by the power of algorithms such as Markov Chain Monte Carlo, importance sampling, and particle filtering. The latter method is turning out to be one of the most fundamental and powerful tools in AI and robotics. A particle filter-like process has also been proposed to lie at the root of cognition, see Lee and Mumford "Hierarchical Bayesian Inference in the Visual Cortex".
The central difficulty with Bayesian reasoning is its deep, deep intractability. Some probability distributions just can't be modeled, other than by random sampling.
Implications for a future where suffering is all but eliminated?
I don't think you've given those implications enough thought. Consider for a moment what it would be like living in an environment in which it is virtually impossible for you to be physically injured in an accident. Think about the level of control that would require, and the level of predictability. And the sorts of things that would have to be excluded.
I suspect you'd be demanding to be let out within an hour, and your sanity would begin to collapse within a week.
Prof. Jaynes would doubtless be surprised by the power of algorithms such as Markov Chain Monte Carlo
Do you have any idea who you're talking about... no, obviously not. Jaynes knew quite well what MCMC was, dear. Though particle filtering was after Jaynes's time.
Derandomization requires some thought. One doesn't always have the time to put in the thought. And then, after you've put in some thought, it often requires some computing power. Often it isn't worth the computing power. But wherever it is possible to predictably do worse than random, you have prior knowledge about the problem that could let you do predictably better than random. That's one reason derandomization is a minor cottage industry in algorithmics.
Eliezer,
Can you point to some place where Jaynes criticizes MCMC specifically? IIRC he took a swipe at bare bones MC integration in PT:LOS, but I never saw him give a mathematically backed argument against MCMC in particular.
"That which can be destroyed by the truth should be". I've seen this attributed to P.C. Hodgell, but without enough detail to check on it.
Eliezer: Am a bit confused on what you mean. Do you claim that there cannot be randomized algorithms that can solve problems in less expected computing time than any deterministic algorithm would take for that problem?
As I understand it (and computational complexity people please please please correct me if I'm way off on this) there's significant theoretical reason to suspect that probabalistic algorithms are strictly more powerful. That is, can be expected to solve some problems faster than the best possible deterministic algorithms.
Proof that this is not the case:
For any probabilistic algorithm A and problem P, there exists some sequence of random actions, among the many of which A is capable, that solves the problem the problem the fastest. (It doesn't take the same amount of time each time, so there has to be some shortest possible time it can take.) Then, there is a deterministic algorithm A' that always takes that series of actions on that problem.
But... I mean, when you combine the running time with the effort it takes to find the better, deterministic solution, it might not be worth it. It’s only worth it if you can generalize the solution to other problems. So I guess randomized algorithms are better when the time you spend thinking about how to do it better pays for itself in time saved, summed across all the problems you apply that thinking to.
EDIT: Oops that last sentence is basically the opposite of what I was trying to say
Memo to Jaynes: please don't generalize beyond statistics. Cough... mixed strategy equilibria in game theory.
Also, what is Harris's quote supposed to mean? (About the moral duty to save children, that is. Not the god one, which is wholly unobjectionable.) I want to interpret it as some kind of skepticism about normative statements, but if that's what he means, it's very oddly expressed. Perhaps it's supposed to be some conceptual analysis about "duty?"
I mean, one ought to understand a syllogism, just as one ought to save the drowning child... no?
It appears to be a quite general principle that, whenever there is a randomized way of doing something, then there is a nonrandomized way that delivers better performance but requires more thought.
If I was a casino owner, I would not purchase a non-randomized slot-machine or a non-randomized roulette wheel. (I might if I was running an underground gaming room.)
Two uses of randomness:
Have to express a sequence, and need that sequence to have minimal information content about inner state.
Don't want to be doomed by history, always want maintain a tiny chance of success.
What other uses am I missing?
Paul, I'm equally confused, especially because Harris is a moral realist. In context, it doesn't make any more sense.
I love the Steve Jobs one.
I love quotes!
"Reality is that which, when you stop believing in it, doesn't go away." - Philip K. Dick
Memo to Jaynes: please don't generalize beyond statistics. Cough... mixed strategy equilibria in game theory.
Game theory changes things because in many cases you are trying to reduce the effectiveness of an opponent, rather than increasing your own cleverness. Thus a "genuinely random" algorithm can be superior to a pseudo-random one, if you presume the opponent is clever enough to figure it out - and the math in game theory does. Certainly noise can act as an antidote to intelligence.
A good litmus test for whether your math is doing this, is to ask whether a fixed string of ones and zeroes is considered defective compared to a "random" generator that happens to produce exactly the same string.
Nancy, the quote is supposed to be from the book Seeker's Mask.
Cyan, there's numerous improvements on MCMC that are less random and more efficient (see e.g. the Wikipedia entry). I don't know if Jaynes would agree with this statement, but once you've used all of your prior knowledge about the structure of the problem, you should be able to use any generator you like for your remaining decision bits, including a "random" one. I think, though, that the main attractiveness of randomness is not knowing whether the resulting run will be flawed, a state of mind that is confused with a definite knowledge of the absence of flaws.
Oh, I'm on top of the MCMC literature -- I just wondered if you'd read something by Jaynes that I hadn't.
"In theory, there's no difference between theory and practice. In practice, there is." - attributed to numerous people
Paul, I was just as confused as you, but in the context of the paragraph, it makes sense. The preceding sentence added, it reads:
"The fact that our ethical intuitions have their roots in biology reveals that our efforts to ground ethics in religious conceptions of "moral duty" are misguided. Saving a drowning child is no more a moral duty than understanding a syllogism is a logical one."
The point appears to be that using the word duty adds too much conscious thought where there is none. Our selfish genes make us lust to save the child, regardless of how we justify it later that day while wearing a smoking jacket and scratching our beards. Similarly, it makes no sense to talk about syllogisms as "logical duties". You have to have understood them before you are even capable of having that discussion.
Leo, hmm... I see the point, but it's gotta be an error. It's a straightforward instance of the genetic fallacy to reason from "our moral intuitions have biological origins" to "therefore, it makes no sense to speak of 'moral duties.'" It might make no sense to speak of religious moral duties -- but surely that's because there's no god, and not because the source of our moral intuitions is otherwise. The quoted sentence seems to equivocate between religious claims of moral duty -- which was the topic of the rest of the surrounding paragraphs -- and [deontological?] claims about moral duty generally.
I like the idea that this is Eliezer's level of input when he's taking a break!
Caledonian: I raised a question, I didn't elicit any thoughts. But yes, what you said is roughly what I'm driving at. An interesting topic.
I'd never heard the theory/practice one before. Toes the line between tautology and prescience brilliantly!
Paul and Leo, perhaps the original quotation would be better interpreted with some choice emphasis:
"Saving a drowning child is no more a moral duty than understanding a syllogism is a logical one."
Re: the Harris quote
The quote may be true. But if you passed up the opportunity to save a drowning child, then the ratio at which you value the happiness of yourself to the happiness of others is probably pretty damn high. So you're still an asshole.
Ben, I quite agree. Paul, I think that's what I was saying, in fact. That to start talking about it as a matter of morals is to oversell how hard of a question it is. It's the proverbial red button that keeps the world from blowing up. Sometimes the answer is obvious.
Ben, I quite agree. Paul, I think that's what I was saying, in fact. That to start talking about it as a matter of morals is to oversell how hard of a question it is. It's the proverbial red button that keeps the world from blowing up. Sometimes the answer is obvious.
RE:"It's possible to describe anything in mathematical notation" I would be interested in seeing the mathematical notation describing love.
Paul,
That's a very good question, I'll do my best to give you an answer.
[Puts on Devil's Advocate hat]
Semi-rhetorical question: if one person says 'I saved the child because my genes directed me', and another says 'I saved the child because my morals directed me,' who is giving the more accurate account of their motive?
Sitting down to mull the decision over in a smoking jacket lets the child drown - this is as much reaction as decision. Hence, I think the first person's probably closer to the mark. You do have a moral duty to save the child, but your ethics-centre probably doesn't even get a word in before you're in the water. Why can you not call the part of the mind that takes that decision in a split second your 'genetic ethics'?
Caveat: genetic ethics, such as they are (or aren't), aren't rational, they are often the source of bias. But since when is morality rational? As long as we're talking about split-second judgment calls, our genes usually do pretty well by us.
I'll be moving to Redwood City, CA in a week, so forgive me if I don't get a regular post out every day between now and then. As a substitute offering, some items from my (offline) quotesfile: