TimFreeman

Karma: 736

TimFreeman 12 Apr 2011 23:14 UTC
5 points
on: We are not living in a simulation
The original post takes the trouble to define “simulation” but not “qualia”. The argument would make much more sense to me if it offered a definition of “qualia” precise enough to determine whether a simulated being does or does not have qualia, since that’s the crux of the argument. I’m not aware of a commonly accepted definition that is that precise.

As it stands, I had to make sure as I was reading it to keep in mind that I didn’t know what the author meant by “qualia”, and after discarding all the statements using that undefined term the remainder didn’t make much sense.

TimFreeman 13 Apr 2011 0:41 UTC
1 point
in reply to: sgr’s comment on: Is masochism necessary?

Arousal: The Secret Logic of Sexual Fantasies

Thanks, I’ll check that out.

pathological beliefs … “I might hurt my partner unless I control myself”

I’m not sure that’s pathological. I’ve read a few independent reports of broken penises from an overenthusiastic woman on top. I’ve also been warned that some people clench their teeth during orgasm, which can make some types of oral sex a problem.

I agree that the proposed examples solve the problems posed by the beliefs, whether true or not.

a person worried about the intensity of his desire hurting his partner might normally self-inhibit; but when tied up, he can see he has no scope to hurt his partner, and thus let his sexuality run free at full intensity—because it’s safe to do so.

That’s bondage, not masochism.

a person exceedingly focused on pleasing his partner might be less able to focus on his own body’s sensations; but when presented with a demonstrably strong, happy partner taking what he/she wants, it’s safe to set that worry aside.

That’s domination, not masochism.

Your examples are interesting, but they aren’t helping to understand masochism. Perhaps there’s some other example from the book you cited that pertains to masochism?

TimFreeman 13 Apr 2011 1:35 UTC
0 points
in reply to: Vladimir_M’s comment on: Human errors, human values

Individual vN-M utility functions therefore cannot be compared, aggregated, or averaged across individuals, which is what any flavor of utilitarianism requires one way or another (and which invariably leads into nonsense, in my opinion).

Estimates of individual utility functions can be averaged, if you do it right, so far as I can tell. A possible estimate of everybody’s utility is a computable function that given a person id and the person’s circumstances, returns a rational number in the interval [0,1]. Discard the computable functions inconsistent with observed behavior of people. Average over all remaining possibilities weighing by the universal prior, thus giving you an estimated utility for each person in the range [0, 1]. We’re estimating utilities for humans, not arbitrary hypothetical creatures, so there’s an approximate universal minimum utility (torturing you and everyone you care about to death) and an approximate maximum utility (you get everything you want). We’re estimating everybody’s utility with one function, so an estimate that says that I don’t like to be tortured will be simpler than one that doesn’t even if I have never been tortured, because other people have attempted to avoid torture.

Does that proposal make sense? (I’m concerned that I may have been too brief.)

Does anything obvious break if you average these across humans?
What links here?
- TimFreeman's comment on What is Eliezer Yudkowsky’s meta-ethical theory? by lukeprog (18 Apr 2011 11:18 UTC; 0 points)

TimFreeman 13 Apr 2011 2:53 UTC
2 points
on: Torture vs. Dust Specks
These ethical questions become relevant if we’re implementing a Friendly AI, and they are only of academic interest if I interpret them literally as a question about me.

If it’s a question about me, I’d probably go with the dust specs. A small fraction of those people will have time to get to me, and of those, none of those people are likely to bother me if it’s just a dust speck. If I were to advocate the torture, the victim or someone who knows him might find me and try to get revenge. I just gave you a data point about the psychology of one unmodified human, which is relatively useless, so I don’t think that’s the question you really wanted answered.

Perhaps the question is really what a non-buggy omnipotent Friendly AI would do. If it has been constructed to care equally about that absurd number of people, IMO it should choose torture. If it’s not omnipotent, then it has to consider revenge of the victim, so the correct answer depends on the details of how omnipotent it isn’t.

TimFreeman 13 Apr 2011 4:12 UTC
7 points
in reply to: David_Gerard’s comment on: Meta: A 5 karma requirement to post in discussion
Hi, I’m new.

Yes, that’s Stross’s Spamularity proposal:

http://www.antipope.org/charlie/blog-static/2010/12/its-made-out-of-meat.html

TimFreeman 13 Apr 2011 19:24 UTC
4 points
in reply to: Joshua_W._Burton’s comment on: Bayesian Judo
The link broke sometime between 2007 and 2011. Do you have another pointer or some summary of what it said?

TimFreeman 14 Apr 2011 13:45 UTC
1 point
in reply to: Vladimir_M’s comment on: Human errors, human values
Analysis paralysis is one path to defeat.

I agree with you that my proposed scheme is computationally intractable, and that it has other issues too. IMO the other issues can be fixed and I hope to get feedback on a completed version at some point. Assuming the fixes are good, we’d then have an unimplementable specification of a way to fairly balance the interests of different people, and a next step would be to look for some implementable approximation to it. That would be an improvement over not having a specification, right?

...this is still only one possible way to aggregate and compare utilities interpersonally, with no clear reason why you would use it instead of some other one that would favor and disfavor different groups and individuals.

The implied principle here seems to be that if we can’t find a unique way to balance the interests of different people, we shouldn’t do it at all. I believe there are multiple plausible schemes, so we will be paralyzed as long as we refuse to pick one and continue. There is precedent for this—many cultural norms are arbitrary, for example.

I wish I actually had multiple plausible schemes to consider. I can think of some with obvious bugs, but it doesn’t seem worthwhile to list them here. I could also make a trivial change by proposing unfair weights (maybe my utility gets a weight of 1.1 in the average and everyone else gets a weight of 1, for example). If anybody can propose an interestingly different alternative, I’d love to hear it.

Also, if I incorrectly extracted the principle behind the parent post, I’d like to be corrected.

TimFreeman 15 Apr 2011 21:48 UTC
2 points
in reply to: handoflixue’s comment on: The Bias You Didn’t Expect
Google to the rescue: http://www.guenergy.com/

TimFreeman 16 Apr 2011 8:54 UTC
1 point
in reply to: syllogism’s comment on: What is Eliezer Yudkowsky’s meta-ethical theory?
Your argument against nihilism is fundamentally “I feel the need to care about other people, and it doesn’t help me to pretend I don’t”.

(I’ll accept for the purpose of this conversation that the empty ethical system deserves to be called “nihilism”. I would have guessed the word had a different meaning, but let’s not quibble over definitions.)

That’s not an argument against nihilism. If I want to eat three meals a day, and I want other people not to starve, and I want my wife and kids to have a good life, that’s all stuff I want. Caring for other people is entirely consistent with nihilism, it’s just another thing you want.

Utiliarianism doesn’t solve the problem of having a bunch of contradictory desires. It just leaves you trying to satisfy other people’s contradictory desires instead of your own. However, I am unfamiliar with Peter Singer’s version. Does it solve this problem?

Friendly to who?

TimFreeman16 Apr 2011 11:43 UTC

4 points

38 comments3 min readLW link

TimFreeman 16 Apr 2011 13:00 UTC
1 point
in reply to: timtyler’s comment on: Friendly to who?
Ah. I interpreted the challenge “Oh, so you get to say what ‘Friendly’ means” to mean that the speaker is objecting that the listener seems to feel that he’s entitled to say what the AI is supposed to do. You seem to be having a different question of what’s the intended general meaning of the term.

I had taken as given that a “Friendly” AI is, by definition, one that you trust to do what you want, even if it’s much smarter and more powerful than you. If your desires are contradictory, it should still do something reasonable.

This leaves “reasonable” and “what you want” undefined. Filling in those definitions with something technically well-defined and getting to the point where you can reasonably expect it to be stable when it’s much smarter than it’s designers is the crux of the problem.

Wikipedia’s definition at http://en.wikipedia.org/wiki/Friendly_artificial_intelligence is: “A Friendly Artificial Intelligence or FAI is an artificial intelligence (AI) that has a positive rather than negative effect on humanity.” Of course, this leaves “positive” undefined, which is not any better than leaving “what you want” or “reasonable” undefined.

At the end of the day, you don’t need the concept. The question the creators of the AI will ask themselves will be “Do we want to run this AI?”

TimFreeman 16 Apr 2011 13:13 UTC
4 points
in reply to: timtyler’s comment on: Friendly to who?
I agree, that’s the “whom” in CEV.

I have an issue with CEV, which is that I don’t think we should extrapolate. We should give the current crop of humans what they want, not what we imagine they might want if they were extrapolated into some finished version of themselves. In the example where Fred wants to kill Steve, CEV says the FAI shouldn’t help because Fred aspires to give up hatred someday and the extrapolated Fred wouldn’t want to kill Steve. On the contrary, I say the FAI shouldn’t help because Steve wants to live more than Fred wants to kill him.

For example, in the original post, if our AI cares about an extrapolated version of speaker A, then it’s possible that that extrapolated A will want different things from the actual present A, so the actual A would be wise to withhold any assistance until he clearly understood the extrapolation process.

TimFreeman 17 Apr 2011 0:34 UTC
1 point
in reply to: Perplexed’s comment on: Friendly to who?
Thanks, I agree. It’s good to see that this multiplayer game notion of morality leads to a new insight that I didn’t build into it.

TimFreeman 17 Apr 2011 0:56 UTC
−1 points
in reply to: David_Gerard’s comment on: Friendly to who?
If you accept morality being a statement about strategies in the multiplayer game, then I agree. However, if you take the usual stand that “action X is moral” is a statement that is either true or false, no matter what anyone else thinks, then whether your moral system leads other people to trust you is irrelevant.

Here’s an environment where it makes a practical difference. At one point my dad tried (and failed) to get me to have “racial consciousness”. I didn’t pay much attention, but I gather he meant that I should not be color-blind in my social interactions so racist white people would trust me. He’s not stupid, so I assume there really were enough whites with that flavor of racism, somewhere, to form an in-group of meaningful size. Thus, if you accept that morality is about getting a specific in-group to trust you, racism is moral for the purposes of signalling membership in that specific in-group.

That conclusion just seems too repugnant to me. I’d rather stop using the word “moral” than use it with that definition. I won’t argue definitions with people, but I do want to point out that that definition leads to some odd-looking patterns of words in true statements.

TimFreeman 17 Apr 2011 1:08 UTC
1 point
in reply to: fubarobfusco’s comment on: Friendly to who?
I agree that you have to renorm everyone’s wants for this to work. I also agree that if you can construct broken minds for the purpose of manipulating the FAI, we need provisions to guard against that. My preferred alternative at the moment follows:
- Before people become able to construct broken minds, the FAI cares about everything that’s genetically human.
- After we find the first genetically human mind deliberately broken for the purpose of manipulating the FAI, we guess when the FAI started to be influenced by that, and retroactive to just before that time we introduce a new policy: new individuals start out with a weight of 0, and can receive weight transferred from their parents, so the total weight is conserved. I don’t want an economy of weight-transfer to arise, so it would be a one-way irreversible transfer.
This might lead to a few people running around with a weight of 0 because their parents never made the transfer. This would be suboptimal, but it would not have horrible conclusions because the AI would care for the parents who probably care for the new child, so the AI would in effect care some for the new child.

Death of the parents doesn’t break this. Caring about the preference of dead people is not a special case.

I encourage people to reply to this post with bugs in this alternative or with other plausible alternatives.
What links here?
- TimFreeman's comment on What is Eliezer Yudkowsky’s meta-ethical theory? by lukeprog (18 Apr 2011 11:18 UTC; 0 points)

TimFreeman 17 Apr 2011 1:55 UTC
−1 points
in reply to: jimrandomh’s comment on: Friendly to who?
There is surprisingly little incentive for selfish AI writers to tilt the friendliness towards themselves.

For normal humans and at the end of the game, I agree with you. However, there are two situations where people may want tilt:
- Narcissists seem to have an unlimited appetite for adoration from others. That might translate to a desire to get the AI tilted as much as possible in their favor. They are 1% of the population according to the abnormal psych literature, but in my experience I see a much larger fraction of the population being subclinically narcissistic enough to be a problem.
- If there’s a slow takeoff, the AI will be weak for some period of time. During this time the argument that it controls enough resources to satisfy everyone doesn’t hold. If the organization building it has no other available currency to pay people for help, it might pay in tilt. If the tilt decays toward zero at some rate we could end up with something that is fair. I don’t know how to reconcile that with the scheme described in another comment for dealing with utility monsters by tilting away from them.
It also creates a time window during which, if the creator dies or suffers brain damage, the AI ends up unfriendly to everyone (including the creator).

I agree that there will be windows like that. To avoid that, we would need a committee taking the lead with some well-defined procedures that allow a member of the committee to be replaced if the others judge him to be insane or deceptive. Given how poorly committee decision making works, I don’t know if that presents more or less risk than simply having one leader and taking the risk of him going insane. The size of the window depends on whether there’s a hard or soft takeoff, and I don’t know which of those to expect.

TimFreeman 17 Apr 2011 2:04 UTC
0 points
in reply to: Perplexed’s comment on: Friendly to who?
I agree that that’s a valid worry, but I intended to look at things from the point of view of the people making the AI so I was expressing a different worry. I edited my post to clarify.

TimFreeman 17 Apr 2011 2:29 UTC
1 point
in reply to: syllogism’s comment on: What is Eliezer Yudkowsky’s meta-ethical theory?
I can’t interpret your post as a reply to my post. Did you perhaps mean to post it somewhere else?

My fundamental question was, how is a desire to help others fundamentally different from a desire to eat pizza?

You seem to be defining a broken version of the zero ethical system that arbitrarily disregards the former. That’s a strawman.

If you want to say that the zero ethical system is broken, you have to say that something breaks when people try to enact their desires, including the desires to help others.

What matters is that the people are helped.

Sorry, that’s incoherent. Someone is helped if they get things they desire. If your entire set of desires is to help others, then the solution is that your desires (such as eating pizza) don’t matter and theirs do. I don’t think you can really do that. If you can do that, then I hope that few people do that, since somebody has to actually want something for themselves in order for this concept of helping others to make any sense.

(I do believe that this morality-is-selfless statement probably lets you get positive regard from some in-group you desire. Apparently I don’t desire to have that in-group.)

TimFreeman 17 Apr 2011 16:15 UTC
−1 points
in reply to: David_Gerard’s comment on: Friendly to who?

Er, your second paragraph appears to say “morality is part of signaling therefore signaling is part of morality therefore the repugnance of a given use of signaling disproves your thesis.”

If the phrase “your thesis” refers to your claim:

Well, of course [statements about morality are] subject to experimentation, or at least real-world testing: do other agents consider you sufficiently trustworthy to deal with? In the indefinitely-iterated prisoner’s dilemma we call society, are you worth the effort of even trying to deal with?

then we’re failing to communicate. For me to agree or disagree with your statement requires me to guess what you mean by “morality”, and for my best guess, I agree with your statement. I said what my best guess was, and pointed out that it leads to some odd-looking true statements, such as “racism is moral” with all those qualifications I said.

Of course, if “your thesis” meant something else, then we’re failing to communicate in a less interesting way because I have no clue WTF you mean.

(Please correct me if I’ve misparsed it.)

I think you have. I was guessing that you’re saying morality is a useful concept and you’re defining it to be is a type of signaling. If you meant something else please clarify. That’s a fine definition, and we can use it if you want, but it leads to the odd conclusion that racism is moral if you’re trying to signal to a group of racists. If you accept that conclusion, that’s great, we have a definition of morality that has odd consequences but it has the advantage of being empirically testable.

If you don’t like to admit the assertion “racism is moral” with the abovementioned qualifications, we need a different definition of morality. Ideally that different definition would still let us empirically test whether “X is moral”. I don’t know what that definition would be.

TimFreeman 17 Apr 2011 16:55 UTC
0 points
in reply to: syllogism’s comment on: What is Eliezer Yudkowsky’s meta-ethical theory?
Thanks, I think I understand better. We have some progress here:
- We agree that the naive model of a selfish person who doesn’t have any interest in helping others hardly ever describes real people .
- We seem to agree that guilt-aversion as a desire doesn’t make sense, but maybe for different reasons.
I think it doesn’t make sense because when I say someone desires X, I mean that they prefer worlds with property X over worlds lacking that property, and I’m only interested in X’s that describe the part of the world outside of their own thought process. For the purposes of figuring out what someone desires, I don’t care if they want it because of guilt aversion or because they’re hungry or some other motive; all I care is that I expect them to make some effort to make it happen, given the opportunity, and taking into account their (perhaps false) model of how the world works.

Maybe I do agree with you enough on this that the difference is unimportant. You said:

If the aim is actually guilt-aversion, this collapses back to position 1), because the person must admit to themselves that other people’s desires are only a correlate of what they want (which is to not feel guilty).

I think you’re assuming here that people who claim a desire to help people and are really motivated by guilt-aversion are ineffective. I’m not sure that’s always true. Certainly, if they’re ineffective at helping people due to their own internal process, in practice they don’t really want to help people.

Has a desire to help others, and pursues it in good faith, using some definition of which universes are preferable that does not weight their own desires over the desires of others.

I don’t know what it means to “weight their own desires over the desires of others”. If I’m willing to donate a kidney but not donate my only liver, and the potential liver recipient desires to have a better liver, have I weighted my own desires over the desires of others? Maybe you meant “weight their own desires to the exclusion of the desires of others”.

We might disagree about what it means to help others. Personally, I don’t care much about what people want. For example, I have a friend who is alcoholic. He desires alcohol. I care about him and have provided him with room and board in the past when he needed it, but I don’t want him to get alcohol. So my compassion for him is about me wanting to move the world (including him) to the place I want it to go, not some compromise between my desires and his desires.

So I want what I want, and my actions are based on what I want. Some of the things I want give other people some of the things they want. Should it be some other way?

Now a Friendly AI is different. When we’re setting up its utility function, it has no built-in desires of its own, so the only reasonable thing for it to desire is some average of the desires of whoever it’s being Friendly toward. But you and I are human, so we’re not like that—we come into this with our own desires. Let’s not confuse the two and try to act like a machine.