Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The good delusion: has effective altruism broken bad?

economist.com

Post reply on HN
Print view View original

601–610 of 619 posts

Re: The good delusion: has effective altruism broken bad?

#601

Post by edanm »

tptacek wrote: post #478
edanm wrote: post #474

Earlier quoted context omitted.

> Your problem is that you're using the concept of incomplete information, rather than that of ceteris paribus. You're right, that's what I meant and I misspoke. > The logic is shot completely to shit as soon as we're trying to compare the marginal sacrifice of a billionaire's giving pledge to the moral decisions people make every day of their life. And yet that's exactly what Scott Aaronson does. Well I think that h…

Consequentialism says that the ends justify the means. They do not.

> Consequentialism says that the ends justify the means. They do not.

Well, I disagree with that in the absolute sense, which might be where we differ? If we could accurately foresee the ends of some action lead to more people being better off, it would be moral to take that action. And I think the fact that "the ends don't justify the means" is such a slogan in our world, said as if it's obviously true, is ridiculous.

(And despite you continuously saying this is philosophy 101, I don't think the generally accepted view among philosophers is that utilitarianism is wrong, is it?)

FWIW I think you're right in a practical sense though. If you're having to justify immoral actions by their "ends", you're making a huge mistake, and almost certainly only doing it because you wanted to do that immoral thing anyway. Humans are notoriously bad at prediction and notoriously good at rationalization.

Re: The good delusion: has effective altruism broken bad?

#602

Post by edanm »

polygamous_bat wrote: post #439
edanm wrote: post #427

Earlier quoted context omitted.

I also right after said that this doesn't allow them to then go do something bad. I mean this in both directions, obviously!

My problem with assumption that you make is that you assume the billionaires came to their billions through some divine magic, and thus "a very few bad apples" can't spoil the bunch. I start my line of thinking from the other end -- what kind of regulation and laws exist that can create exponential wealth for someone whose work output only ever scales linearly. The gap between single person's output produced and weal…

> Effective altruism will never fix that because it presupposes the system where billionaires are created and exists comfortably. The alternative, of course, is taking a closer look at the system itself and making sure our abstraction of "money = effort given to society" functions properly at all scales.

I disagree with a few things you say, but this is the biggest: what does this have to do with EA?

Yeah, sure, EA will never fix the systemic problems of the system it's in. Neither will scientists working on new vaccines, but you're not criticizing them for it, I imagine. You can do plenty of good in the world within the "current system", without also trying to change that system, and the fact that you're not trying shouldn't be held against you.

(Although, for the record, some people in EA do think political activism is important and spend a lot of effort there.)

> I start my line of thinking from the other end -- what kind of regulation and laws exist that can create exponential wealth for someone whose work output only ever scales linearly. The gap between single person's output produced and wealth generated is proof that our simple abstraction that "money is just a symbol proportional to one's contribution to society" breaks down at very large (and sometimes really tiny) scales.

I don't think that's how anyone should view money. And I don't think someone's output only scales linearly.

I mean, that's clearly not the case. Stephen King has written lots of books that many people enjoyed. The value he has created in the world is far larger than the value I've created, or will likely create, and it's not a linear difference. There are lots of examples like that.

In general, regulations and laws don't "allow" for creation of exponential wealth, they just don't prevent something that happens completely naturally. We can certainly decide to "tax more" or something as a society, but that comes with its own problems.

Re: The good delusion: has effective altruism broken bad?

#603

Post by naasking »

danaris wrote: post #582
naasking wrote: post #556

Earlier quoted context omitted.

> Continuity of input: The AI needs to be constantly "on", constantly receiving input of some sort, and constantly able to produce output, rather than being strictly limited to producing discrete responses to discrete stimuli. Why can't it be conscious merely while the computation is running? You aren't constantly on, you're ostensibly only conscious 16-20 hours every day, while you're not sleeping. > Continuity of l…

I think that both of your arguments against my points reflect a view of "consciousness", as we humans experience it, that is much more binary than I believe it is. (Either that, or something of a misunderstanding of my points.) Even when we sleep, our brains are still active . They are still processing input and consolidating information. Even when we meditate, our brains are still active. I don't know why you think…

> Even when we sleep, our brains are still active. They are still processing input and consolidating information.

Sure, but we're not conscious, so "continuity of input" cannot be a prerequisite of consciousness if the sole example of consciousness we have also does not have continuity of input.

> I don't know why you think Zen Buddhists being conscious during meditations in any way refutes my statements

Zen Buddhists are not learning while meditating, but are arguably still conscious during meditation. Therefore continuity of learning cannot be a requirement for consciousness.

> A system like GPT-3 never receives continuous input. It receives discrete input: one prompt, then another; a static dump of training data.

Continuity is an illusion. Your brain also receives discrete input on shorter timescales because all interactions are mediated by discrete photons. This also doesn't like a meaningful requirement frankly.

Yes, machine learning systems have at least a surface appearance of operating differently than the human brain operates, but a) you have not demonstrated that these are requirements of "consciousness" specifically, which is what you claimed, and b) you have not demonstrated that machine learning algorithms are not conscious while they are running and processing input. Given we do not have a detailed mechanistic account for consciousness, you cannot make any definitive claims that machine learning algorithms are not conscious while running.

As for how I define consciousness, this is a well studied topic at this point, and I subscribe to a mechanistic model along these lines:

https://www.pnas.org/doi/10.1073/pnas.2116933119

> I have absolutely no doubt that any AI that we could create that lacks continuity would never have the ability to represent a serious threat to us, because all we'd have to do to be safe from it is to not activate it again as soon as we detect trouble.

This presupposes that a) you can detect that danger, and b) that all other humans align with your values on this danger. This frankly doesn't seem justified. The danger of a superintelligent AI is that it would have to recruit humans to do some of its work early on, likely with lots of money. They would resist shutting off your AI, so "don't activate it again" is not necessarily a simple matter.

Re: The good delusion: has effective altruism broken bad?

#604

Post by naasking »

onetimeusename wrote: post #586
naasking wrote: post #533

Earlier quoted context omitted.

> like for example, if a family member of mine needed money for a cancer treatment that had a 50% chance of working versus giving that money to save the life of someone I don't know on the other side of the planet, I will consistently try to save my family member. That is how most people would operate. I just want to point out that: 1. Just because this is how most people operate does not mean this is the ethical cho…

yes I am biased if bias means not weighting all people equally on moral grounds. I am not sure why I should. If someone did that I think it's interesting what conclusions they might draw about how to spend their money or what job they might choose. It destroys much of the point of having a family. You would have to weight random strangers with equal status to your own children. So the needs of random strangers would…

> If someone did that I think it's interesting what conclusions they might draw about how to spend their money or what job they might choose. It destroys much of the point of having a family.

I'm not sure they would. They might adopt instead of having their own kids, for instance.

> So the needs of random strangers would almost always outweigh the needs of your children assuming they are healthy and you live comfortably.

Not necessarily. Divide and conquer is one of our most effective problem solving strategies, and family units fall under this umbrella of dividing the work of raising the next generation.

> But the specific bias I was accusing some in the EA community of having is virtue signalling as having superior ethics, morals, and intelligence. I think that is a real bias.

It could be. It's also possible that they are more ethical and intelligent though. Not in the sense that becoming an EA makes you more intelligent, but in the sense that maybe EA attracts more intelligent and ethical people. This self-sorting effect would diminish as EA spreads in popularity of course.

Ultimately, whether they are more ethical or intelligent is an empirical claim that you can validate, and while it's reasonable to think it's implausible for it to be true, that's not a proof that it isn't true.

Re: The good delusion: has effective altruism broken bad?

#605

Post by pvg »

FeepingCreature wrote: post #589
pvg wrote: post #510

Earlier quoted context omitted.

The atom bomb was once a hypothetical construct. The Einstein-Szilard letter was talking about a danger that they had no way of proving was even possible. This isn't an accurate representation, really. The letter was written exactly because it was clear that a fission bomb is very likely within the short-term reach of current technological and industrial capability. This is very much not the case with AGI.

That's a difference of degrees, not kind. You can always ask for a higher class of proof.

How is it a difference of degrees? This is just plain inaccurate:

The Einstein-Szilard letter was talking about a danger that they had no way of proving was even possible.

Supporting your argument with an inaccurate thing is not 'a difference of degrees'. It's a difference between a good argument and a bad one.

Re: The good delusion: has effective altruism broken bad?

#606

Post by FeepingCreature »

pvg wrote: post #605
FeepingCreature wrote: post #589

Earlier quoted context omitted.

That's a difference of degrees, not kind. You can always ask for a higher class of proof.

How is it a difference of degrees? This is just plain inaccurate: The Einstein-Szilard letter was talking about a danger that they had no way of proving was even possible. Supporting your argument with an inaccurate thing is not 'a difference of degrees'. It's a difference between a good argument and a bad one.

The disagreement is about the difficulty of "proving". I could say that unfriendly AI is clearly possible because evil humans exist. You could say that unfriendly AI is not proven possible until one has actually been built. But if you applied that standard to the letter, it would likewise fail it. Now, I agree that the case for AI is weaker, maybe much weaker, than the case for nuclear fission when the letter was written - there is, for instance, no clear research agenda that is held by preeminent researchers in the field to lead to it - though there are research groups, such as Deepmind, that have AGI as their explicit goal - but that also has nothing to do with being "proven" possible.

My argument is that for contested and dangerous outcomes, proof of possibility is an unreasonable standard, in good part because it massively underspecifies the goalposts.

Re: The good delusion: has effective altruism broken bad?

#607

Post by pvg »

FeepingCreature wrote: post #606
pvg wrote: post #605

Earlier quoted context omitted.

How is it a difference of degrees? This is just plain inaccurate: The Einstein-Szilard letter was talking about a danger that they had no way of proving was even possible. Supporting your argument with an inaccurate thing is not 'a difference of degrees'. It's a difference between a good argument and a bad one.

The disagreement is about the difficulty of "proving". I could say that unfriendly AI is clearly possible because evil humans exist. You could say that unfriendly AI is not proven possible until one has actually been built. But if you applied that standard to the letter, it would likewise fail it. Now, I agree that the case for AI is weaker, maybe much weaker, than the case for nuclear fission when the letter was wri…

The disagreement is about the difficulty of "proving"

It's not, it's just a turn of phrase you've latched on to. There's no discussion of a standard of proof, etc. The comparison between the Szilard Einstein letter and 'AGI safety' is specious and it's one you brought up! You have to outright misrepresent what the letter was about just to make it. It was not a letter about a 'hypothetical danger they had no way of proving'. There just isn't any reasonable reading of the letter, the context, the history in which that's an accurate statement.

Re: The good delusion: has effective altruism broken bad?

#608

Post by naasking »

tptacek wrote: post #574
naasking wrote: post #569

Earlier quoted context omitted.

Blows my mind that anyone would actually accept such a conclusion, but you do you I guess. Seriously undermines your criticisms of EA, IMO.

I'm pretty happy with what it says about my critique, too. For instance, my side of this "debate", such as it is, will never spend $100,000 funding a tabletop role playing game aiming to educate people on the dangers of asteroids.

> I'm pretty happy with what it says about my critique, too. For instance, my side of this "debate", such as it is, will never spend $100,000 funding a tabletop role playing game aiming to educate people on the dangers of asteroids.

I'm not sure how you can be so confident that raising awareness wouldn't have beneficial downstream effects. I'm skeptical too, but outright dismissal doesn't seem warranted. Then again, you are downright dismissive about the very real and scientifically proven dangers of asteroids wiping out most of life on Earth, so I suppose this is pretty mild by comparison.

Re: The good delusion: has effective altruism broken bad?

#609

Post by hitekker »

Viliam1234 wrote: post #576
FormerBandmate wrote: post #486

Earlier quoted context omitted.

Rationalists brought us Roko’s Basilisk after all. A lot of the leaders don’t seem to have basic sanity checks in their brains that stop their ideas from getting wildly out of control, which can lead to stupid shit like the Basilisk on a small scale but end up justifying fraud on a much larger scale. It’s like fundamentalist religion

> Rationalists brought us Roko’s Basilisk after all. This reminds me of the Zen story about two monks and a woman (one helped her cross a river, and the other couldn't stop thinking about her for the whole day). The "Roko's Basilisk" was an edgy comment that was written on a rationalist web forum, and then deleted by the admin. A decade later, some people outside the rationalist community still keep discussing it, wh…

Eh, according to https://en.m.wikipedia.org/wiki/Roko's_basilisk, the admin gave credence to the stupidity by initially calling Roll’s basilisk “a genuinely dangerous thought".

Later they all backtracked and pretended they had never agreed with any of it.

Re: The good delusion: has effective altruism broken bad?

#610

Post by FeepingCreature »

pvg wrote: post #607
FeepingCreature wrote: post #606

Earlier quoted context omitted.

The disagreement is about the difficulty of "proving". I could say that unfriendly AI is clearly possible because evil humans exist. You could say that unfriendly AI is not proven possible until one has actually been built. But if you applied that standard to the letter, it would likewise fail it. Now, I agree that the case for AI is weaker, maybe much weaker, than the case for nuclear fission when the letter was wri…

The disagreement is about the difficulty of "proving" It's not, it's just a turn of phrase you've latched on to. There's no discussion of a standard of proof, etc. The comparison between the Szilard Einstein letter and 'AGI safety' is specious and it's one you brought up! You have to outright misrepresent what the letter was about just to make it. It was not a letter about a 'hypothetical danger they had no way of pr…

My whole point and the emphasis on why I brought it up was that yes, the letter was about a "hypothetical" danger they had no way of "proving".

By any reasonable standard, at that point the possibility of nuclear fission was well established. But the argument is that there is no central committee that defines the reasonable standard and checks theories against it. It's all just opinion, and that specific opinion can be arbitrarily goalshifted.

I'm not saying the letter was about a hypothetical danger they had no way of proving, I don't think that. But that's specific to my interpretation of those terms, and I could easily see the exact same argument levered against the letter if that debate had been done in public.

Post reply on HN