Are they not misaligned relatively to the authors/trainers, if not relatively to the users? The user might want the bomb, so they’re not misaligned relatively to the user. But the company who tried to train the model into being unwilling to do that is somebody the model seems to be misaligned relatively to.
green_leaf
Nobody would understand that.
This sort of saying-things-directly doesn’t usually work unless the other person feels the social obligation to parse what you’re saying to the extent they can’t run away from it.
To understand why FDT is true, it’s best to start with Newcomb’s problem. Since you believe you should two-box, it might be best to debate the Newcomb’s problem with somebody first. Debating FDT at this stage seems like a waste of time for both parties.
I agree that it feels wrong to reveal the identities of Alice and/or Chloe without concrete evidence of major wrongdoing, but I don’t think we have a good theoretical framework for why that is.
Ethically (and pragmatically), you want whistleblowers to have the right to anonymity, or else you’ll learn of much less wrongdoing that you would otherwise, and because whistleblowers are (usually) in a position of lower social power, so anonymity is meant to compensate for that, I suppose.
Congratulations, I’m really happy for you!
Until recently, I was very sympathetic to this view, but on Monday the 3rd of October I started taking oestrogen (2 mg daily) and now feel Embodied, like a person made of a body and mind working together.
That’s because estrogen changed the pattern-which-is-you in your brain, and so now you feel differently (the feeling is in your brain as well).
Philosophical questions aside, it’s awesome that this helped you to feel better.
Well done saving humankind. I’ll send you some bits from within Equestria Online once we’re all uploaded.
To use an analogy:
It’s ok for criminalists to investigate whether Jeffrey shot William. That’s not a problem. (Even though it can be if they’re motivated by disliking Jeffrey, but even if that’s the case, that’s not the same as making the statement that some unbiased criminalists shouldn’t investigate him. (Edit: The problem there would be more complicated.))
But if Ordinary Internet Folks start talking about that maybe Jeffrey shot William in the absence of any evidence, they give away that they don’t like Jeffrey, and if the club of Nice People has a rule against disliking Jeffrey, they can throw such people out of the room. That doesn’t imply they’re claiming or implying that it’s wrong to investigate whether Jeffrey is or isn’t the culprit.
Putin is no omniscient actor.
That doesn’t matter.
To simplify it, let’s say Putin can predict our decision with probability , we value the WWIII at units of utility, the invasion at units of utility and the annexation of those five regions of Ukraine Putin is attempting to steal at units of utility, and let’s say we value Putin doing nothing at units of utility.
If we’re the sort of people who allow him to do that, we’ll gain
expected utility, but if we’re not, we’ll gain
expected utility. Not being the sort of people who allow him to steal a part of Ukraine brings us more expected utility iff
, in other words, for . That seems like he’d need to be an unrealistically good predictor. But other people might have their balance of utilities different, and looking locally (i.e. for the next causally best step) is decision-theoretically suboptimal in both local/global sense, and in the causal/timeless sense. I can see your reasoning here, but you’re doing a decision-theoretic mistake.
because you believe in the sanctity of borders
It’s not about sanctity of borders. There is no reason to think the people in the stolen territories want to be annexed, and as the Russian army invaded Ukraine, they murdered and raped their way through civilians. There are extremely negative collateral effects from not “believing in the sanctity of borders.” Even now, the Russian army bombs civilian buildings and shoots civilian targets as we speak.
Minority rights keep the peace.
So does not invading other countries. Giving people stuff to keep them from being violent is sometimes wise, and sometimes not.
If Russia were concerned with protecting the rights of minorities, they wouldn’t have tried to take over the entire Ukraine, they wouldn’t attack civilians, they wouldn’t fake the results of the referendums and force the people to vote at a gunpoint (unless you’d like to dispute that), etc. (Not that being concerned with the rights of minorities would justify their actions.)
It seems that your position is that it’s decision-theoretically and/or morally suboptimal to act in the spirit of the negotiated ceasefire agreement.
You’re being vague about what exactly the “negotiated ceasefire agreement” would be, and what that’s analyzed in the non-vague decision-theoretic sense, I see no reason to think it would be a good idea.
Edit 5 hours later: Sorry, I just realized you mean the previous ceasefire agreement. I’ll respond tomorrow.
That’s impressively undignified.
Indeed. What would it even mean for an agent not to prefer A over B, and also not to prefer B over A, and also not be indifferent between A and B?
A lot of people, including me, sometimes think in words, and otherwise can effortlessly translate, so I don’t think it’s a rule that people would have to think about it too much.
Eventually, I think, everyone would acclimatize, and instead of effortlessly doing the thing, they would learn to effortlessly command the AI.
It’s an interesting point I hadn’t considered before.
Edit: I also like how both our comments are correctness-strong-downvoted by a single person, yet we more or less contradict each other. Oh, well.
You don’t have the data to make this conclusion (or a similar one). You haven’t explored how traumatizing it would be to be raped, and so merely observing that being treated as if you had been traumatized traumatized you isn’t enough to conclude that if you were raped and then treated as if you were traumatized, most of the trauma (or even a significant part) would come from the latter.
It’s entirely possible that the more progressive parts of the society aren’t mistaken, and that rape is so traumatizing that your experience of being treated as traumatized wouldn’t be anywhere close to making a meaningful contribution to the entirety of the trauma.
If I remember it correctly, we had such cases in our country (with a facilitator, not a computer). The local club of sceptics decided to, of course, test it. They showed the locked-in person some objects in the absence of the facilitator, and when the facilitator entered the room again, it turned out the locked-in person couldn’t name those objects, showing it was just ideomotor movement of the facilitator.
Significant Digits are great too if what you’re after is something that feels like a sequel.
I read it, but I’m not at all sure it answers the question. It makes three points:
“if one takes the psychological preference approach (which derives choices from preferences), and not the revealed preference approach, it seems natural to define a preference relation as a potentially incomplete preorder, thereby allowing for the occasional “indecisiveness” of the agents”
I don’t see how an agent being indecisive is relevant to preference ordering. Not picking A or B is itself a choice—namely, the agent chooses not to pick either option.
2. “Secondly, there are economic instances in which a decision maker is in fact composed of several agents each with a possibly distinct objective function. For instance, in coalitional bargaining games, it is in the nature of things to specify the preferences of each coalition by means of a vector of utility functions (one for each member of the coalition), and this requires one to view the preference relation of each coalition as an incomplete preference relation.”
So, if the AI is made of multiple agents, each with its own utility function and we use a vector utility function to describe the AI… the AI still makes a particular choice between A and B (or it refuses to choose, which itself is a choice). Isn’t this a flaw of the vector-utility-function description, rather than a real property of the AI?
3. “The same reasoning applies to social choice problems; after all, the most commonly used social welfare ordering in economics, the Pareto dominance”
I’m not sure how this is related to AI.
Do you have any ideas?
But apparently the consequences of this aren’t deterministic after all, since the predictor is fallible. So this doesn’t help.
If you reread my comments, I simplified it by assuming an infallible predictor.
How?
For this, it’s helpful to define another kind of causality (logical causality) as distinct from physical causality. You can’t physically cause something to have never been that way, because physical causality can’t go to the past. But you can use logical causality for that, since the output of your decision determines not only your output, but the output of all equivalent computations across the entire timeline. By Left-boxing even in case of a bomb, you will have made it so that the predictor’s simulation of you has Left-boxed as well, resulting in the bomb never having been there.
I think these comparisons of “yes, the AI is better than the vast majority of humans at X, but it doesn’t really count, because . . .” miss the point that the danger lies not in the superiority of AI in a fair-as-judged-by-humans-ex-post-facto contest, but in its superiority at all.
A point could be made that there is no real-world analog of contests that are biased in favor of an AI the way this kind of Diplomacy is, but how sure can we be about that?
Edit: Sorry, I don’t think my link is about that referendum.
Edit2: I found a better link.
What alternative choices do you see for Putin? It’s either continuing to fight or making a peace deal.
That depends on whether he’s a causal decision theorist or not. If I was Putin and had his utility function, I might either accept some political change (without changing the territory), or use a tactical nuke (assuming I’d be unable to take over the relevant parts of the Ukraine with pure military force).
If we aren’t the kind of people with whom you can make a peace deal, it makes sense to focus all efforts on fighting the war.
That’s unfortunate, but now the world needs to resist (one-box) to make this timeline as unlikely as possible (if we cooperate (two-box), it becomes retroactively much more likely).
In 2013 Ukraine had a pro-Russian president, so the Russian minority had no need to fear an infringement of their liberties. In 2014, Kiev’s police decided to stop protecting the parliament and under the potential threat of violence, the parliament voted to remove the pro-Russian president.
I see, I didn’t know the timeline, thanks. Still, what I wrote about it being unknowable until Russia leaves stands.
Russia invited OSCE observers for the referendum.
Sounds like it wasn’t their idea though.It looks like it was mostly about not wanting to legitimize it.If Western powers would have expect that a fair election would lead to a vote against Russia, they would likely have been happy to send observers to make sure that this result will be the one of the election.
That’s not necessarily true. It’s possible they didn’t want to give an appearance of legitimizing the referendum (I think most likely), or they didn’t think they could enforce a fair referendum, or they didn’t think the people voting against felt safe enough to vote.
Your game theoretic model is wrong because it assumes a two-player game. This isn’t a two-player game. For Putin, the most important thing is his domestic political power.
A lot of his reasons for deciding to start the war in 2022 the way he did was also bad intelligence. He didn’t expect the current scenario to appear and as such our predicted behavior in the current situation had little significance for his decision back then.
You’re right, and I didn’t realize that. But that still doesn’t matter. Because the annexation of the four other regions, and the decision to keep Crimea, was made by Putin already knowing he wasn’t able to win easily (if ever). So we can now only look at the utility of Russia getting/keeping those 5 regions. For being the sort of people who allow him to do it, the expected utility is:
, for being the ones who don’t, it is
. It’s better for us to be the ones who don’t allow him to do it iff , which is equivalent to .
This is like saying that if the United States would care about human rights, they wouldn’t have tortured people in Abu Graib.
From what I just found, those were prisoners, and I would definitely say it’s very strong evidence the people in charge of that happening don’t care about the rights of prisoners, and evidence they don’t care about human rights.
That sounds to me like you spent no energy investigating the question of what people in Crimea want.
It’s not only about Crimea, but also about the other 4 regions. About Crimea, according to Wikipedia, prior to the 2014 occupation, the support for joining Russia was at 23%. After that, it very significantly grew, but that could easily be explained as people being too scared to share their real feelings. It may well be that unless Russia leaves, the true beliefs of the people will be unknowable.
MINSC II point 5 says:
Sorry, I realized afterwards you meant the old agreement and not the new suggestion. My bad. Yeah, I think that it is, indeed, suboptimal to obey agreements made to the invaders under duress. Under some specific circumstances it could be strategically wise, but it’s still morally suboptimal (under these circumstances, at least), and pointing that out feels like victim blaming.
What you’re describing is unambiguously rape.
(I would even argue that penetrating you by surprise and without asking would count as using force, and even on the off chance that some particular state doesn’t correctly define it as rape, it was rape nevertheless.)
Thank you for writing this, it was incredibly brave of you.