Currently planned to be divided into three parts, "Map and Territory", "How To Actually Change Your Mind", and "Mysterious Answers to Mysterious Questions" - that should give you an idea of the intended content. No ETA, still struggling to find a writing methodology that gets up to an acceptable writing speed.

One of the expected effects of the karma change is to make people more cautious about what they put in a top level post. Perhaps this is only evidence of that effect.

Is this a behavior we want?

No. It is not difficult to create a top level post that is approved of or at least kept at '0'. I want undesirable top level posts to hurt.

If not, what can we do about it?

Replace all '-ve' karma value displays of top level posts with '- points' or '<0 points'. We don't necessarily need to know just how disapproved of a particular post is.

I've called before for median-based karma: you set a score you think a post should have and the median is used for display purposes, with "fake votes" reducing the influence of individual votes until there are enough to gain a true picture.

There is a limited downvote budget for each voter (in some ratio to the voter's budget). Downvoting a post now uses 10 points from that budget rather than 1, so perhaps low-karma downvoters (or downvoters who have exhausted their downvote budgets) are now having less of an impact.

It could be sympathy, or a judgment that the poster shouldn't be excessively discouraged from posting in the future.

Is this a behavior we want?

Sure, why not? We can always change things later if we start getting overrun by bad posts, and people still aren't willing to vote them down into negative territory.

I like seeing the negative number on my posts. But I have also noticed a voting trend that seems to be much more forgiving than the posts of old.

The first wave of readers seem to vote up; the second wave votes down; over time it stabilizes somewhere near where the first wave peaked. This doesn't seem to happen on posts that are really superb.

I think showing the full number of up and down votes would be helpful to authors and also let people know why a post is at the number it is. Seeing +5 -7 is different than seeing -2.

That being said, karma inflation seems to be hitting. I am rarely getting downvoted on comments anymore. I don't think I have improved that much as a commentator. I am not convinced that the effects you are seeing are only happening to Posts.

Is this a behavior we want? If not, what can we do about it?

I think a great way to handle the Post karma is to hide the actual number for a week. Let it show + or - for positive or negative but no numbers. By the time one week has passed most people will have moved on.

Another solution may be to keep actual voting history available and let people see votes by people who said their history is public. As far as I can tell, that preference doesn't do anything yet.

ETA: Another solution would be to set karma rewards to only happen after a certain threshold. Between 0 and +5 you don't get any karma. After that, you get 10 karma per point. Everything under 0 still penalizes you 10 karma per point.

Or the above but only getting rewards after a certain percentage votes up. +5 -1 nets 40 karma, +20 -16 nets nothing, but each have a score of +4.

This is a reasonable justification for using a sockpuppet, and I'll try to keep it in mind the next time I have something to write that would not be instantaneously identifiable as me.

It has seemed to me that some of Eliezer's recent post scores have been inflated by around 5-10 points due to his being Eliezer; it would be interesting to test this hypothesis.

My karma management techniques:

1) If I'm in a thread and someone's comment is rated equally with mine, and therefore potentially displaying atop my comment, I downvote theirs until it'll pass mine despite my downvote, to give my comment more exposure. I remove the downvote later, usually upvoting (their comment is getting voted better than mine because it's good).

2) If I'm debating someone and I want to downvote their comment, I upvote it for a day or so, then later return to downvote it. This gives the impression that two objective observers who read the thread later agreed with me. This works best on long debate threads, because a) if my partner's comments are getting immediately upvoted, they tend to be encouraged and will continue the debate, further exposing themselves to downvotes and b) they get fewer reads, so a single vote up or down makes a much bigger impression when almost all the comments in the thread are rarely upvoted/downvoted past +/- 2.

3) Karma is really about rewarding or punishing an author for content, to encourage certain types of content. Comments that are too aggressive will not be upvoted even if people agree with the point, because they don't want to reward aggressive behavior. Likewise, comments that are not aggressive enough are given extra karma - the reader's first instinct is to help promote this message because the timid author won't promote it enough on his own. This is nonsensical in this format, but the instinct is preserved.

I've noticed that the comments that get voted up the most are those that do probability calculations, those whose authors' names pop out of the page, and those which are cynical on the surface, possibly with a wry humor, while revealing a deep earnestness. If you have something unpopular to say, or are just plain losing an argument, that's the best tone to take, because people will avoid downvoting if they disagree, but will usually upvote if they do agree.

EDIT: I agree with Alicorn that votes shouldn't be anonymous, as it would remove the dirtiest of these variably dirty techniques, but in the meantime, play to win.

An automatic block against downvoting any comment that's a direct response to one of yours would be good.

or [vote the comment] down to -2, otherwise your opponent might infer that you are the one who downvoted it. [...] [T]he counterintuitive, but strategically correct response is to downvote more.

(Downvoted. EDIT: Vote cancelled; see below.) "Opponent"? "Strategically correct response"? Are you sure we're playing the same game?

I've noticed this too. It is one of several annoying problems that would evaporate if votes weren't anonymous.

I find myself once again missing Usenet.

Perhaps if LW had an API we could get back to writing specially-designed clients, which could do all the aggregation magic we might hope for?

Apparently even specific users have their own rss feeds, so I've settled with a feed aggregated from the feeds of a few people. It'd be better if the "friend" functionality worked (maybe it even does, but I don't know it!), so that the same could be done within the site, with voting and parent/context links.

If I've gained 500 karma, I should be able to upvote or downvote F(500) comments, where F would probably be a log function of some sort.

Um, that math doesn't work out unless the number of new users expands exponentially fast. You need F(n) to be at least n, and probably significantly greater, in order to avoid a massive bottleneck.

A community is only as good as its constituents. I would hope that there are enough people around who like majority-wisdom-challenging insights, to offset this problem. "Insights" being the key word.

Non-Born rules give us anthropic superpowers. It is plausibly the case that the laws of reality are such that no anthropic superpowers are ever possible, and that this is a quickie explanation for why the laws of reality give rise to the Born rules. One would still like to know what, exactly, these laws are.

To put it another way, the universe runs on causality, not modus tollens. Causality is rules like "and then, gravity accelerates the bowling ball downward". Saying, "Well, if the bowling ball stayed up, we could have too much fun by hanging off it, and the universe won't let us have that much fun, so modus tollens makes the ball fall downward" isn't very causal.

<.05

I am no cognitive scientist, but I believe some of my "thinking" takes place outside of brain (elsewhere in my body) and I am almost certain some of it takes place on my paper and computer.

operate solely on a substrate of neurons in the physical brain.

As opposed to ...? Ion channels? Quantum phenomena? Multiple interacting brains? Non-neuronal tissue? Neuronal-but-extracranial cells? Soul? Beings outside the observable universe, running the simulator?

What is this belief supposed to be distinguished from?

Like others, I see some ambiguity here. Let me assume that the substrate includes not just the neurons, but the glial and other support cells and structures; and that there needs to be blood or equivalent to supply fuel, energy and other stuff. Then the question is whether this brain as a physical entity can function as the substrate, by itself, for high level mental functions.

I would give this 95%.

That is low for me, a year ago I would probably have said 98 or 99%. But I have been learning more about the nervous system these past few months. The brain's workings seem sufficiently mysterious and counter-intuitive that I wonder if maybe there is something fundamental we are missing. And I don't mean consciousness at all, I just mean the brain's extraordinary speed and robustness.

I'm at least +70 decibans ("99.99999%") confident that mental states supervene on to physical states. Whether your exact description to do with neurons in the brain completely captures all the physical states I'm less confident of.

EDIT: updated from 30 to 70 decibans: I would more easily be convinced that I had won the lottery than that this wasn't so.

Still curious... How about giving us an ETA?

To get nitpicky, the brain is made of both neurons and glial cells - and the glial cells also seem to play a role in cognition.

I am quite comfortable with the idea that I am my brain, that my brain is made of ordinary living matter (atoms making up molecules making up proteins making up cells), that this matter forms specialized structures responsible for cognition, and I would be hugely surprised if given proof that the highest mental functions cannot be explained adequately in terms of that ontology. The strangest alternative I can think of is Penrose's ENM incomputable-quantum-coherence hypothesis and I'd assign less than 5% probability to his thesis being correct.

How does "operate solely on" regard distributed cognition arguments, like "creative thought is created via interaction with the remaining human culture" and "we constantly offload cognitive processes (such as memory) to external substrates (like computers and books)"?

Also, the "highest" human mental functions operate via a number of lower-level processes. Does "solely on human neurons" include e.g. possible quantum phenomena on a low level?

With a straightforward interpretation of your question, I'd answer "95%".

But since you made special mention of being "sneaky", I'll assume you've attempted to trick me into misunderstanding the question, and so I'll lower my probability estimate to 75%, with the missing twenty points accounting for you tricking me by your phrasing of the question.

Commenting before reading other replies---I'm going to give the boring, sneaky reply that the question isn't well-specified enough to have an answer; I'd need to know more about what you mean by something to operate solely on a substrate. I mean, clearly there are a lot of cognitive tasks that most people can only do given a pencil and paper, or a computer ... is that the sneaky part, that we store information in the environment, and therefore we're not solely neurons?

Could you clarify what you mean by operate on? Or is that part of the point?

Using the definition of 'operate on' that I think is most natural, I'd say there is a .05% chance that these functions only operate on (effect) the physical brain. Unless you mean directly, and then I would assign an 80% chance.

Using the definition of 'operating on' meaning 'requiring', I'd say that there is a 90% chance (probability) that only the brain is required for 90% (fraction) of its functioning. The probabilities I assign would fall down dramatically as you try to raise the 2nd 90% (the fraction). So that I would probably only assign a 1% chance that 100% of higher functions require only the brain.

SIAI: Utopia or hundred times your money back!

Eliezer, would you accept a bet $100 against $10000?

You definitely win. If I say "you'll get killed doing that" and you are, I shan't expect to pay back my winnings when you're reanimated.

Perhaps you've already defined "superintelligent" as meaning "self-directed, motivated, and recursively self-improving" rather than merely "able to provide answers to general questions faster and better than human beings"... but if you haven't, it seems to me that the latter definition of "superintelligent" would have a much higher probability of you losing the bet. (For example, a Hansonian "em" running on faster hardware and perhaps a few software upgrades might fit the latter definition.)

So long as you make your Newcomb's choice for what seem like good reasons rather than by flipping a quantum coin, it is likely that very many of you will pick the same good reasons, and that Omega can easily achieve 99% or higher accuracy. I would expect almost no Eliezer Yudkowskys to two-box - if Robin Hanson is right about mangled worlds and there's a cutoff for worlds of very small amplitude, possibly none of me. Remember, quantum branching does not correspond to high-level decisionmaking.

I think Omega's capabilities serve a LCPW function in thought experiments; it makes the possibilities simpler to consider than a more physically plausible setup might.

Also, I'd say that our wetware brains probably aren't close to deterministic in how we decide (though it would take knowledge far beyond what we currently have to be sure of this), but e.g. an uploaded brain running on a classical computer would be perfectly (in principle) predictable.

Thank to everyone who replied. So I see that we don't really believe that the universe is deterministic in the way implied by the problem. OK, that's consistent then.

If your choice to one-box or two-box is dependent upon some random factors, then Omega can't predict what will happen because when he makes the prediction, he is up-branch of you. He doesn't know which branch you'll be in.

What Omega can do instead is simulate every branch and count the number of branches in which you two-box, to get a probability, and treat you as a two-boxer if this probability is greater than some threshold. This covers both the cases where you roll a die, and the cases where your decision depends on events in your brain that don't always go the same way. In fact, Omega doesn't even need to simulate every branch; a moderate sized sample would be good enough for the rules of Newcomb's problem to work as they're supposed to.

But the real reason for treating Omega as a perfect predictor is that one of the more natural ways of modeling an imperfect predictor is to decompose it into some probability of being a perfect predictor and some probability of its prediction being completely independent of your choice, the probabilities depending on how good a predictor you think it really is. In that context, denying the possibility that a perfect predictor could exist is decidedly unhelpful.

I'm sufficiently uninformed on how quantum mechanics would interact with determinism that so far I've been operating under the assumption that it doesn't. Maybe someone here can enlighten me? Does the behavior of things-that-behave-quantumly typically affect macro-level events, or is this restricted to when you look at them and record experimental data as a direct causal result of the behavior? Is there some way to prove that quantum events are random, as opposed to caused deterministically by something we just haven't found? (I'm not sure even in principle how you could prove that something is random. It'd be proving the negative on the existence of causation for a possibly-hidden cause.)

Perfection is impossible, but a very, very accurate prediction might be possible.

I'm not sure I understand the connection between truth-seeking and polling, unless the specific truth you seek is simply the percentage of people who give a particular answer to a poll. Are you simply talking about a more accurate polling company or using polling to find other truths?

Here's an idea for how a LW-based commercial polling website could operate. Basically it is a variation on PredictionBook with a business model similar to TopCoder.

The website has business clients, and a large number of "forecasters" who have accounts on the website. Clients pay to have their questions added to the website, and forecasters give their probability estimates for whichever questions they like. Once the answer to a question has been verified, each forecaster is financially rewarded using some proper scoring rule. The more money assigned to a question, the higher the incentive for a forecaster to have good discrimination and calibration. Some clever software would also be needed to combine and summarize data in a way that is useful to clients.

The main advantage of this over other prediction markets is that the scoring rule encourages forecasters to give accurate probability estimates.

For this specific case, you could try asking the analogous question with a higher probability value. E.g. "if you’ve got a one-in-two DNA match on a suspect, does that mean it’s one-in-two that you’ve got that dude’s DNA?". Maybe you can have some graphic that's meant to represent a several million people, with half of the folks colored as positive matches. When they say "no, it's not one-in-two", you can work your way up to the three million case by showing pictures displaying what the estimated amount of hits would be for a 1 to 3, 1 to 5, 1 to 10, 1 to 100, 1 to 1000 etc. case.

In general, try to use examples that are familiar from everyday life (and thus don't feel like math). For the Bayes' theorem introduction, you could try "a man comes to a doctor complaining about a headache. The doctor knows that both the flu and brain cancer can cause headaches. If you knew nothing else about the case, which one would you think was more likely?" Then, after they've (hopefully) said that the man is more likely to be suffering of a flu, you can mention that brain cancer is much more likely to cause a headache than a flu is, but because flu is so much more common, their answer was nevertheless the correct one.

If you're using Powerpoint, you might want to make a slide that says something like:

2,999 negatives -> 1% test positive -> 30 false positives

1 positive -> 99% test positive -> 1 true positive

So out of 31 positive tests, only 1 person has SDS.

Do it with pictures

I take it you've already looked at Eliezer's "Intuitive Explanation"?

I think it's really important to get the idea of a sliding scale of evidentiary strength across to people. (This is something that has occurred to me from some of my recent attempts to explain the Knox case to people without training in Bayesianism.) One's level of confidence that something is true varies continuously with the strength of the evidence. It's like a "score" that you're keeping, with information you hear about moving the score up and down.

The abstract structure of the prosecutor's fallacy is misjudging the prior probability. People forget that you start with a handicap -- and that handicap may be quite substantial. Thus, if a piece of evidence (like a test result) is worth, say "10 points" toward guilt, hearing about that piece of evidence doesn't necessarily make the score +10 in favor of guilt; if the handicap was, say, -7, then the score is only +3. If, say, a score of +15 is needed for conviction, the prosecution still has a long way to go.

(By the way, did you see my reply to your comment about psychological evidence?)

If you actually want to debate this, we could do so in the comments section of my post, or alternatively over in the Richard Dawkins forum.

(Though since you say "my intent is merely to debunk komponisto's post rather than establish Amanda's guilt", I'm suspicious. See Against Devil's Advocacy.)

Make sure you've read my comments here in addition to my post itself.

There is one thing I agree with you about, and that is that this statement of mine

these two things constituting so far as I know the entirety of the physical "evidence" against the couple

is misleading. The misleading part is the phrase "so far as I know", which has been interpreted by people who evidently did not read my preceding survey post to mean that I had not heard about all the other alleged physical evidence. I didn't consider this interpretation because I was assuming that my readers had read both True Justice and Friends of Amanda, knew from my previous post that I had obviously read them both myself, and would understand my statement for what it was -- a dismissal of the rest of the so-called "evidence". However, in retrospect, I should have foreseen this misunderstanding, so I've now edited the sentence to read:

these two things constituting pretty much the entirety of the physical "evidence" against the couple.

ETA: At least one person has upvoted the parent without also upvoting this comment, which I interpret as an endorsement of Rolf Nelson's essay. I find this baffling. Almost every one Nelson's points (autopsy report, luminol prints, staged break-in, alleged cleanup...) was extensively discussed in comments at the time. The only one that wasn't (a supposed handprint of Knox's on a pillow in Kercher's room) is an outright falsehood -- as you will see from following Nelson's link, it's not even (close to) what that article claims. Furthermore, Nelson criticizes me for "accept[ing] propaganda from the Friends of Amanda (FoA) at face value" while citing True Justice for an "Introduction to Logic 101".

I challenge anyone who thinks that this represents a serious challenge to my post to come out and identify themselves.

I've created a rebuttal to komponisto's misleading Amanda Knox post, but don't have enough karma to create my own top-level post. For now, I've just put it here:

I don't understand how this was worked around. It looks like (rolf's karma + karma lost by this being posted at the top level) would still have been insufficient.

The karma limit was serving the purpose for which it was intended. If, for some reason, an exception was granted I would like to see this announced.

Criticizing komponisto for citing "Friends of Amanda Knox" while you yourself cite "True Justice" causes those criticisms to fall flat.

Unfortunately, I find that your essay is wading into Dark Arts territory, since its intent is to show that komponisto's original essay was "misleading", and that that would somehow give veracity to arguments of Amanda Knox's guilt. Using that same logic, one would have to consider the implications of the chief prosecutor in Amanda Knox's case being convicted of abuse of office in another murder trial.

However, I would be interested in seeing komponisto and rolf nelson discuss the actual details of the case; in particular, the points that rolf nelson brought up in the essay.

These are excellent questions/ideas. I want a mentor too!

I thought about contacting you to see if you wanted to start a little study group reading through the sequences. (For example, I started reading through the metaethics sequence and it was useless. My kinds of questions are like, 'What do any of these words mean? What's the implied context? Etc., etc.) But I'm not very good at details, and couldn't imagine any way of doing so. Except maybe meeting somewhere like Second Life so we can chat...

The remarkable and depressing thing to me is that most people are not able to see it at a glance. To me it just seems like a string of obvious bluffs and non-sequiturs. Do you remember what was going on in your head when you didn't see it at a glance?

Static? Unchanging? Complete (as far as definitions of the situation go)? Simple (as far as equations go - it lacks the dynamic variable representing the need to update)?

In a world with an FAI Singleton, actions that would violate another individual's rights might be simply unavailable, making the concept of a legal/justice system obsolete.

In other scenarios, uploading/splitting would still take resources, which might be better used than in absorbing a criminal punishment. A legal/justice system could apply punishments to multiple instances of the criminal, and could be powerful enough to likely track them down.

If I had criminal intentions, I would upload myself just before the criminal act, so that the upload would be innocent

I am not convinced that the upload would be innocent. Maybe, if the upload was rolled back to before the criminal intentions. Any attempt by the upload to profit from the crime would definitely make it complicit.

Criminal punishment could also take the form of torture, effective if the would be criminal fears any of its instances being tortured, even if some are not.

For posts, this might work.

For comments, these are loaded without most readers reading them. Furthermore, the likelihood that any single comment will be read decreases with the number of all comments. It seems like this would work much less well for comments.

Not directly applicable, but perhaps relevant: I was told this advice and found it useful (in that I used it to make important life decisions). "Don't do your passion for a job," she said. "Everyone wakes up one day and hates their job. Don't wake up one day and hate what you love. Do something you like that you're at."

Also, I don't remember who told me this or if I made it up, but I've relayed it to people: Don't look for fulfillment from your job. Don't go for the highest peaks; just try to avoid the lowest valleys.

p(A|X) = p(X|A)*p(A) / ( p(X|A)*p(A) + p(X|~A)*p(~A) )

A = box has diamond
X = box beeped

p(A) = .01
p(X|A) = .88
p(X|~A) = .08

p(A|X) = .88 * .01 / ( .88 * .01 + .08 * .99)
p(A|X) = .0088 / (.0088 + .0792)
p(A|X) = .0088 / .088
p(A|X) = .1

This is different than the likelihood ratio:

LR = p(X|A) / p(X|~A)
LR = .88 / .08
LR = 11

The likelihood ratio can be worded as, "It is 11 times more likely to be a diamond when it beeps." The original formula answers the question, "What is the probability that this beep means a diamond?"

In other words, the likelihood ratio is starting with the contents of a box and asking whether that box is going to beep. p(A|X) is starting with a beep and trying to figure out what that beep means about the contents of the box.

If you haven't read Bayes Theorem yet, it's definitely the place to start.

The likelihood ratio is Pr(beep | diamond) / Pr(beep | empty) = 0.88/0.08 = 11. I was going to say you ought to read the link for "likelihood ratio", but there's nothing there, so you should try the other wiki.

Also, don't think of running the detector over every box; think of testing one box at random.

A quick glance seems to indicate that they are achieving these linear time algorithms through massive parallelization. This is "cheating" because to do a linear-time sort of size n, you need O(n) processing units. While they seem to be arguing that this is acceptable because processing is becoming more and more parallel, this breaks down for large n. One can easily use traditional algorithms to sort a billion elements in O(n * log n); however for their algorithm to sort such a list in O(n) time, they need a billion (times some constant factor) times more processing units than to sort a list of size n.

I'm also vaguely perplexed by their basic argument. They want to have programming tools and computational models which are closer to the metal to take advantage of the features of new machines. This ignores the fact that the current abstractions exist, not just for historical reasons, but because they are easy to reason about.

This is all from a fairly cursory read of their paper, however, so take it with a grain of salt.

You should definitely start a blog. I for one look forward to reading and commenting.

I only have a very general feel for where that line is, so I can't help with that, but I would personally be interested in following such a diary. Perhaps you could start a blog?

Given what you've said so far about your personal situation, I think it's appropriate to discuss your personal progress and introspection regarding rationality on this site. I think a lot of us would find it helpful and interesting to see how your thought processes and beliefs change as you reexamine them.

I'm especially curious about more details regarding your personal situation, your past history of religious beliefs, and "Event X".

I recall EY commenting at some point that the way to make political progress is to convert intractable political problems into tractable technical problems. I think this kind of discussion would be more interesting and more profitable than a "traditional" mind-killing political debate.

It might be interesting, for example, to develop formal rationalist political methods. Some principles might include:

• Always conduct a comprehensive fact-gathering phase before beginning any policy discussion.
• Develop techniques to prevent people from becoming emotionally committed or status-linked to positions.
• Subject every statement to formal logical analysis; if the statement fails any obvious rule of logical inference, the statement is deleted and its author censured.
• Rigorously untangle webs of inference. A statement arguing against the death penalty should involve probability estimates of the number of crimes the penalty does (or does not) deter, the cost of administering it, etc, and connect these estimates to a global utility function. The statement must include analyses of how the argument changes in response to changes in the underlying probability estimates.

I disagree; discovering that someone holds political views opposed to yours can inhibit your ability to rational consider arguments; arguments become soldiers, etc.,

Besides, I think the survey from ages ago showed the general spread of political views, and I doubt much has changed since. For discussing particular issues, there are other places available, and it may be that only by not discussing hot topics can we keep the barriers to entry up that keep the LW membership productive.