alkexr

• alkexr 12 Jul 2023 1:12 UTC

  1 point

  0

  in reply to: ChristianKl’s comment on: Some ba­sics of the hy­per­com­pe­tence the­ory of government

  I’m not an expert either, and I won’t try to end the F-35 debate in a few sentences. I maintain my position that the original argument was sloppy. “F-35 isn’t the best for specific wars X, Y and Z, therefore it wasn’t a competent military decision” is non sequitur. “Experts X, Y and Z believe that the F-35 wasn’t a competent decision” would be better in this case, because that seems to be the real reason why you believe what you believe.

• alkexr 11 Jul 2023 17:36 UTC

  1 point

  0

  in reply to: ChristianKl’s comment on: Some ba­sics of the hy­per­com­pe­tence the­ory of government

  F-35 aren’t the crucial component to winning the kind of wars in Iraq or Afghanistan. They also aren’t the kind of weapon that are important to defend Taiwan. They are just what the airforce culture wants instead of being a choice made by a hypercompetent military.

  I mostly agree with your perception of state (or something) competence, but this seems to me like a sloppy argument? True, the US does have to prepare for the most likely wars, but they also have to be prepared for all other wars that don’t happen because they were prepared, aka. deterrence. The F-35 may not be the most efficient asset when it comes to e.g. Taiwan, but it’s useful in a wide range of scenarios, and its difficult to predict what exactly one will need as these platforms have to be planned decades in advance.

  Not sure how to put this in a way that isn’t overly combative, but since the only point you made where I have domain specific understanding seems to be sloppy, it makes me wonder how much I should trust the rest? At a glance it doesn’t look like artight reasoning.

  EDIT: As a sidenote, what the airforce culture wants is in itself a military consideration. It’s often better to have the gear that works well with established doctrine than some other technology that outperforms it on paper.

• alkexr 11 Jul 2023 17:05 UTC

  1 point

  0

  in reply to: trevor’s comment on: Some ba­sics of the hy­per­com­pe­tence the­ory of government

  • You believe that there is a strong evolutionary pressure to create powerful networks of individuals that are very good at protecting their interests and surviving in competition with other similar networks

  • You believe that these networks utilize information warfare to such an extent that they have adapted by cutting themselves off from most information channels, and are extremely skeptical of what anyone else believes

  • You believe that this policy is a better adaptation to this environment than what anyone else could come up with

  • These networks have adapted by being so extremely secretive that it’s virtually impossible to know anything about them

  • You happen to know that these networks have certain (self-perceived) interests related to AI

  • You happen to believe that these networks are dangerous forces and it makes sense to be scared

  • This image that you have of these networks leads to anxiety

  • Anxiety leads to you choosing and promoting a strategy of self-deterrence

  • Self deterrence leads to these networks having their (self-perceived) interests protected at no cost on their behalf

  Given the above premises (which, for the record, I don’t share), you have to conclude that there’s a reasonable chance that your own theory is an active information battleground.

• alkexr 10 Jul 2023 10:33 UTC

  16 points

  0

  on: Some ba­sics of the hy­per­com­pe­tence the­ory of government

  1. You’re saying that these hypothetical elites are hypercompetent to such a hollywoodical degree that normal human constraints that apply to everyone else don’t apply to them, because of “out of distribution” reasons. It seems to me that “out of distribution” here stands as a synonym of magic.

  2. You’re saying that these hypothetical elites are controlling the world thanks to their hypercompetence, but are completely oblivious to the idea that they themselves could lose control to an AI that they know to be hypercompetent relative to them.

  3. It seems to me that lie detection technology makes the scenario you’re worried about even less likely? It would be enough for just a single individual from the hypothetical hypercompetent elites to testify under lie detection that they indeed honestly believe that AI poses a risk to them.

  4. It’s worth pointing out I suppose that the military industrial complex is still strongly interested in the world not being destroyed, no matter how cartoonishly evil you may believe they are otherwise, unless they are a Cthulhu cult or something. They could still stumble into such a terrible outcome via arms race dynamics, but if hypercompetence means anything, it’s something that makes such accidents less, not more, likely.

  My arguments end here. From this point on, I just want to talk about… smell. Because I smell anxiety.

  Your framing isn’t “here is what I think is most likely the truth”. Your framing is “here is something potentially very dangerous that we don’t know and can’t possibly ever really know”.

  Also, you explicitly, “secretly” ask for downvotes. Why? Is something terrible going to happen if people read this? It’s just a blogpost. No, it’s not going to accidentally push all of history off course down into a chasm.

  Asking for downvotes also happens to be a good preemptive explanation of negative reception. Just to be clear, I downvoted not because I was asked to. I downvoted because of poor epistemic standards.

  Do note that I’m aware that very limited information is available to me. I don’t know anything about you. I’m just trying to make sense of the little I see, and the little I see strongly pattern matches with anxiety. This is not any sort of an argument, of course, and there isn’t necessarily anything wrong with that, but I feel it’s still worth bringing up.

A Tele­pathic Exam about AI and Consequentialism

• alkexr 22 Nov 2022 15:06 UTC

  4 points

  2

  on: Here’s the exit.

  I immediately recognize the pattern that’s being playing out in this post and in the comments. I’ve seen it so many times, in so many forms.

  Some people know the “game” and the “not-game”, because they learned the lesson the hard way. They nod along, because to them it’s obvious.

  Some people only know the “game”. They think the argument is about “game” vs “game-but-with-some-quirks”, and object because those quirks don’t seem important.

  Some people only know the “not-game”. They think the argument is about “not-game” vs “not-game-but-with-some-quirks”, and object because those quirks don’t seem important.

  And these latter two groups find each other, and the “gamers” assume that everyone is a “gamer”, the “non-gamers” assume that everyone is a “non-gamer”, and they mostly agree in their objections to the original argument, even though in reality they are completely talking past each other. Worse, they don’t even know what the original argument is about.

  Other. People. Are. Different.

  Modeling them as mostly-you-but-with-a-few-quirks is going to lead you to wrong conclusions.

• alkexr 1 Oct 2022 3:57 UTC

  15 points

  3

  on: A Few Ter­rify­ing Facts About The Russo-Ukrainian War

  (Meta: writing this in separate comment to enable voting /​ agreement /​ discussion separately)

  If you want to make the case for tactical nuclear deployment not happening (which I hope is the likely outcome), I want to see a model of how you see things developing differently

  I’ll list a few possible timelines. I don’t think any of these is particularly likely, but they are plausible, and together with many other similar courses of events they account for significant chunks of probability mass.

  1. Discontinuity in power in Russia.

  2. Internal turmoil or collapse in Russia (e.g. regions start declaring independence). It becomes clear that nuclear weapons won’t save Russia.

  3. Abrupt cut in western support to Ukraine, including ammunition (e.g. due to another big war). Putin thinks he can win without nuclear weapons.

  4. Russian army starts being competent. Putin thinks he can win without nuclear weapons.

  5. Conflict freezes over winter, then turns into boiling-the-frog: events happening too slowly to trigger nuclear response. Over the years defeat slowly becomes an accepted fact in Russia.

• alkexr 1 Oct 2022 2:58 UTC

  10 points

  2

  on: A Few Ter­rify­ing Facts About The Russo-Ukrainian War

  On Nord Stream sabotage:

  1. Looks like sabotage. Accidents very rarely look like this. (Very high confidence.)

  2. Probably by state actors. It seems like a task that requires significant resources and planning. Also there was plenty of military presence in the area, it’s just hard to believe that non-state actors could perform something like this unnoticed. (Medium confidence.)

  3. It wasn’t an ally of Germany. There is always a chance that you get caught /​ leave evidence, and after attacking the critical infrastructure of an ally no one will have a reason to trust you. None of the allies of Germany stand to gain anything that is anywhere near comparable to that risk (that I can think of). (High confidence.)

  4. Given geographic reach, the countries that could perform something like this are probably those around the Baltic sea, plus USA, UK, France. (Medium confidence.)

  That leaves us with Russia and Germany. I don’t see what Germany could gain from this. I don’t see what Russia could gain from this either, but then Russia has developed a habit of doing things despite having nothing to gain from them. Also, I see some reasons why Russia could think this is a good idea (implicitly threatening the West by demonstrating willingness to use grey-zone warfare against their critical infrastructure, to try to get them to back down).

  So possibly Russia. (Low confidence.)

  Epistemic status: proof by lack of imagination.

You can tell a draw­ing from a painting

• alkexr 2 Feb 2022 16:54 UTC

  3 points

  in reply to: Slider’s comment on: Be­fore Colour TV, Peo­ple Dreamed in Black and White

  Thus I claim we don’t know whether people see dreams.

  That’s a pretty bold claim just a few sentences after claiming to have aphantasia.

  Some of my dreams have no visuals at all, just a vague awareness of the setting and plot points. Others are as vivid and detailed as waking experience (or even more, honestly), at least as far as vision is concerned. Dreams can fall anywhere on a spectrum between these extremes, and sometimes they can even be a mixture (e.g. a visual experience of the place and an awareness of characters in that place that don’t appear visually).

  Yes, people do see dreams. I’m fairly certain I can tell the difference.

• alkexr 27 Jan 2022 18:24 UTC

  5 points

  in reply to: Martin Sustrik’s comment on: Why do we need a NEW philos­o­phy of progress?

  Yes, I’m aware of all that, and I agree with your premises, but your argument doesn’t prove what you think it does. Let’s try to reductio it ad absurdum, and turn the same argument against the possibility of fast technological or scientific feedback cycles.

  If you live in a technologically backwards society (think bronze age), you can’t become more advanced technologically yourself, because you’ll starve spending your time trying to do science. The technology of society (including agriculture, communication, tools, etc.) needs to progress as a whole. If you live in a scientifically backwards society, you can’t have more accurate beliefs, because you’ll be burned at the stake by all the people believing in nonsense. Therefore, science and technology can only progress as fast as the majority can adopt it.

  And all of the above is true, actually, up to a certain point in history. But once the scientific understanding of society advances to the point where it understands that science is a thing and has a basic understanding of how science works, it can basically create a mesa-feedback-loop. Similarly, once you have technologies like writing and free market capitalism, suddenly it’s possible to set up a tech company, sell something worthwhile and in exchange not starve.

  And that’s the frame for my original comment. I didn’t mean to imply that a fast moral feedback loop would involve a single person going on some meditation retreat that is somehow a clever feedback loop in disguise and then come back more moral or whatnot. I think it is possible that there is some innovation, moral or social or otherwise (e.g. a common understanding of common knowledge), that would enable the creation of fast moral and social feedback loops.

  So the question, again: what are the necessary conditions for such a feedback loop? Are they present? What would it look like? How would you recognize it if it was happening right in front of you?

  (EDIT: spelling)

• alkexr 26 Jan 2022 12:33 UTC

  3 points

  in reply to: Wei Dai’s comment on: Why do we need a NEW philos­o­phy of progress?

  It seems pretty likely that moral and social progress are just inherently harder problems, given that you can’t [...] have fast feedback cycles from reality (like you do when trying to make scientific, technological and industrial progress).

  We can’t? Have we tried? Have you tried? Is there some law of physics I’m missing? What would a real, genuine attempt to do just that even look like? Would you recognize it if it was done right in front of you?

• alkexr 26 Jan 2022 12:16 UTC

  2 points

  in reply to: Dumbledore's Army’s comment on: Why do we need a NEW philos­o­phy of progress?

  There are multiple meanings of “progress” afoot here. Tabooing the word, my reading of your point is “moving toward any specific imagined future state of the world we all agree is good is good, therefore moving forward is good”.

• alkexr 11 Jan 2022 21:19 UTC

  20 points

  0

  on: Cal­ibra­tion proverbs

  (Another non-native having a go at it...)

  When your advice both ways seems fine,
  Calibrate, then make it rhyme.

• alkexr 13 Dec 2021 21:19 UTC

  2 points

  1

  on: Is “gears-level” just a syn­onym for “mechanis­tic”?

  more transparent to outsiders

  There is the danger of it being more transparency-illuding instead. (Yeah, I just invented that term, but what did I mean by it?)

• alkexr 15 Nov 2021 19:35 UTC

  9 points

  in reply to: Shmi’s comment on: Im­prov­ing on the Karma System

  My gut feeling is that attracting more attention to a metric, no matter how good, will inevitably Goodhart it.

  That is a good gut feeling to have, and Goodhart certainly does need to be invoked in the discussion. But the proposal is about using a different metric with a (perhaps) higher level of attention directed towards it, not just directing more attention to the same metric. Different metrics create different incentive landscapes to optimizers (LessWrongers, in this case), and not all incentive landscapes are equal relative to the goal of a Good LessWrong Community (whatever that means).

  I am not sure what problem you are trying to solve, and whether your cure will not be worse than the disease.

  This last sentence comes across as particularly low-effort, given that the post lists 10 dimensions along which it claims karma has problems, and then evaluates the proposed system relative to karma along those same dimensions.

Si­mu­lacrum Levels of Trust

• alkexr 15 Jun 2021 20:16 UTC

  8 points

  on: The value of low-con­scien­tious­ness peo­ple on teams

  The way this topic appears to me is that there are different tasks or considerations that require different levels of conscientiousness for the optimal solution. In this frame, one should just always apply the appropriate level of conscientiousness in every context, and the trait conscientiousness is just a bias people have in one direction or the other that one should try to eliminate.

  This frame is useful, because it opens up the possibility to do things like “assess required conscientiousness for task”, “become aware of bias”, “reduce bias”, etc. But it may also be wrong in an important way. It’s somewhere between difficult to impossible to tell how much conscientiousness is required in any particular case, what’s more, even what constitutes an optimal solution may not be obvious. In this frame, trait conscientiousness is not bias, but diversity that nature threw against the problem to do selection with.

  I have trouble understanding why, in this case, everyone would need to have a consistently high or consistently low level of it across a wide range of contexts; and why, for example, one can’t just try a range of approaches of varying consientiousness levels and learn to optimize from the experience. It isn’t necessary in any of the examples above to have a person involved with consistently low levels of it, just a person who in that particular case takes the low-conscientiousness approach. This way we could still fall back on the interpretation as a bias and blame nature for just being inefficient doing the evolution biologically instead of memetically.

• alkexr 9 Jun 2021 13:04 UTC

  1 point

  in reply to: Ericf’s comment on: The re­verse Good­hart problem

  I think it’s empirical observation.

  The world doesn’t just happen to behave in a certain way. The probability that all examples point in a single direction without some actual mechanism causing it is negligible.

• alkexr 9 Jun 2021 12:53 UTC

  2 points

  in reply to: Stuart_Armstrong’s comment on: The re­verse Good­hart problem

  I ended up using mathematical language because I found it really difficult to articulate my intuitions. My intuition told me that something like this had to be true mathematically, but the fact that you don’t seem to know about it makes me consider this significantly less likely.

  If we have a collection of variables $\left\{v\right\}$, and $V=max\left(v\right)$, then $V$ is positively correlated in practice with most $U$ expressed simply in terms of the variables.

  Yes, but $V$ also happens to be very strongly correlated with most $U$ that are equal to $V$. That’s where you do the cheating. Goodhart’s law, as I understand it, isn’t a claim about any single proxy-goal pair. That would be equivalent to claiming that “there are no statistical regularities, period”. Rather, it’s a claim about the nature of the set of all potential proxies.

  In a Bayesian language, Goodhart’s law sets the prior probability of any seemingly good proxy being a good proxy, which is virtually 0. If you have additional evidence, like knowing that your proxy can be expressed in a simple way using your goal, then obviously the probabilities are going to shift.

  And that’s how your $V$ and $V^{\prime }$ are different. In the case of $V$, the selection of $U$ is arbitrary. In the case of $V^{\prime }$, the selection of $U$ isn’t arbitrary, because it was already fixed when you selected $V^{\prime }$. But again, if you select a seemingly good proxy $U^{\prime }$ at random, it won’t be an actually good proxy.