= 27d567bc2d48d9f40e9ccc20ea370b15)
DaFranker comments on Rationality Quotes November 2012 - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (898)
I thought the whole point of CEV was to extrapolate forwards in time towards the ultimate reflectively-consistent set of values to formulate one single coherent utility function (with multiple parameters and variables, of course) that represents the optimal equilibrium of all that humans would want if they were exactly as they would want to be and would want exactly that which they would wish to want.
This reminds me more of CAV (Coherent Aggregated Volition) than CEV. CEV is, IIRC, intended as a bootstrap towards "Whatever humans would collectively find the best possible optimization after infinite re-evaluations", if any such meta-ethics exists.
The Coherent Extrapolated Volition of one group of humans is not the same thing as the Coherent Extrapolated Volition of another group of humans. Humans populations change and even evolve over time due to forces that are not carefully constructed to move the population in the same direction as the CEV of their ancestors and so later generations will not have the same CEV as previous ones.
Eliezer has a lot to answer for when it comes to encouraging magical thinking along the lines of "all (subsets of) humans have the same Coherent Extrapolated Volition". He may not be confused himself but his document certainly encourages it.
It depends on how you define "humans", but considering how old some of the references to the Golden Rule are at least some of our utility function is older than most civilizations. Do you have any proof that previous generations were fundamentally different to us, and not, like most (all?) humans today, confused about how to implement their utility function (if we give the poor healthcare, they wont have an incentive to work!)
Well... IMO, not counting psychopaths as human amounts to a no-true-Scotsman fallacy.
I was referring to extinct species and subspecies of human. Of course psychopaths are human, but AFAIK they have always been a small minority.
The existence of blind people is not usually taken to disprove "human beings have sight".
Indeed. Imagine someone arguing that past civilizations saw colour differently to modern humans; it makes a pretty god analogy for this discussion.
The no-true-Scotsman fallacy applies to an argument when it excludes particular cases by rhetoric rather than for objective reasons. It does not apply to any particular drawing of category boundaries on its own.
I've always interpreted no-true-Scotsman as warning about the dangers of arguing by definition. At the very least, saying psychopaths are not human runs the risk of being argument by definition.
Well, I'd say it depends on the complexity of those objective reasons. “The way to carve reality at its joints, is to draw simple boundaries around concentrations of unusually high probability density in Thingspace. Otherwise you would just gerrymander Thingspace.”
(OTOH I think language should also depend on what you value: if your utility function is the number of inwardly-thrice-bent metal wires capable of nondestructively fastening several standard sheets of paper together at an edge in the universe, it's handy to have a single word for ‘inwardly-thrice-bent metal wire capable of nondestructively fastening several standard sheets of paper together at an edge’, whether that's a natural category or not. But you shouldn't pretend it's a natural category.)
"No true Scotsman":
Not "No true Scotsman":
The second is just using a nonstandard definition, not redefining the word to fit the line of argument, so does not fall under the No True Scotsman fallacy. Even if you're gerrymandering reality ahead of time, it doesn't count as No True Scotsman (At the very least, that isn't even an argument yet, so can't be a fallacious argument!)
"Everybody likes to watch a beautiful sunset"
"Fred doens't. Mind you, he's blind".
"Then he doesn't count"
True Scotsman or not?
Clearly not. As I noted upthread, True Scotsman requires that the redefinition is happening for rhetorical, rather than objective, reasons. The relevant reference class could have been picked out ahead of time, and I wouldn't predict those two folks are going to continue that dispute.
It is trivially true that restricting the definition of 'human' can reduce the possible differences between the CEVs of subsets of humans. This is just a matter of shifting the workload into the 'human' definition. Unless you plan to restrict the definition of human to one individual, however, there are still going to be differences between the CEV of subsets (except by coincidence).
Having a weak-to-moderate norm in favour of doing things that you would consider helpful or at least not harmful to others in your social group does seem to be popular (not as consistent or as strong as norms against excreting waste products in public but right up there!). That CEVs of various combinations of humans are similar isn't the point. Of course they will be. In fact, on average I'd expected them to be more similar than the groups of humans themselves are. But they are not identical (except by coincidence).
No!
That isn't a dichotomy. Clearly both past humans and current humans aren't effectively optimising toward their respective CEVs. But those CEVs are also going to be different because there isn't any magic (or focused expenditure of optimisation power) holding the CEV constant!
(I'm not sure what "fundamental" means exactly so I'll just note that I've never proposed any kind of difference beyond "not the same").
It would be great if you wrote up a short discussion level post to clear up what seems to be a common misconception. Please consider doing so.
I'm not sure how useful that would be, or rather whether I'm the right person to be doing it. I thought I said everything that needed to be said in this thread already but it wasn't necessarily successful at reaching the target audience. Perhaps someone more in tune with the idealism behind the disagreement could explain better.
I meant that, say, Neanderthals have a good chance of a serious CEV difference. However, your statement that all humans have different CEVs is unsupported by any evidence. For example:
Historically, dumping waste products was considered relatively harmless; sure it smells a little but hey, what doesn't? These people lacked the germ theory of disease, remember. No-one thought deliberately spreading disease was OK.
That is not a fully general counterargument against your lack of any evidence at all.
But there's no magic changing it! If you assume human morality evolved, why would our ethics have changed much more than, say, our diet?
Nobody said that they would have.
You are arguing against a straw man. Please read some of the message you replied to or the ones preceding it. Even, say, 1/3 of the sentences is likely to be sufficient---I've been repeating myself to make this clear.
You are claiming that the CEV of any group of humans - including all humanity - changes over time, yes? You seem to think this is a self-evident truth, but I have yet to see any examples of such a change. You removed the first half of that sentence - as I pointed out, if human morality evolved (which I assume you believe) then there is no reason to think that it would change any more than human dietary preferences - a child may discover sweets taste better than cabbage, and henceforth refuse cabbage in favor of sweets, but this is true for all children. What you are suggesting the same as if I claimed our taste buds had rearranged themselves, and that is why the Romans ate roast dormouse and we don't.
Both sides of this debate are hamstrung by failing to distinguish between basic values and extrapolated volition. There have been major shifts in ethics within living memory, regarding race, gender, the environment and sexuality. Whether they are shifts in basic valiues or in the way basica values are extrapolated is not obvious.
Except that it is. You don't dissuade a racist or a misogynist or whatever with brain surgery. You just show them that their model of minorities/women/homosexuals/whatever is flawed. You don't alter their brain to terminally value, say, preventing slavery, you just show them that to satisfy their existing terminal value of avoiding human suffering they should prevent slavery. That has no effect on Coherent Extrapolated Volition.
Except that it does, because having been persuaded they extrapolate differently. But maybe by CEV you mean some idealised version.
Your impression that it is obvious that changes in de facto ethics are not changes in basic values rests heavilty on the assumption than basic values are in hardware, not software. That is not obvious , either.
Very idealistic. What if putting down other humans is an actual terminal value for some people?
It is, but my prior that two logically different things turn out to be exactly identical is pretty small. EDIT: OTOH, I think that almost all humans' CEVs would be so similar that a world with a FAI optimizing for CEV<Group A> would be very unlikely to feel like a dystopia to Group B, unless the membership criteria to Group A are deliberately gerrymandered to achieve that.
Of course there will be some variation between individuals, yes. But, as you say, probably not enough to matter; unless you're actively filtering it should average out the same for most large groups.
No, his argument is that CEVs of any (subset of) humans is a tiny cluster in value space.
He has, in fact, made that argument (as well). I repeat the claim:
Who knows? It's possible EY thinks it will be. There doens't seem to be any authoritative answer to that.
Poll here
Thank you. I had slightly misunderstood what you were saying, but I also hadn't looked at all the variables and you pointed right at what I was missing.