The inversion of the usual ranking is weird. ‘B’ is usually worse than ‘A’—B list, B movies, plan B—and here ‘B’ happens to be the supreme level of knowledge/operation...
Eliezer attended a pretty serious and wide diplomatic simulation game, that made him appreciate how difficult is to just maintain a global equilibrium between countries and avoid nuclear annihilation
Eliezer is being an idiot who forgot his own maxim to not rely on fictional evidence.
This was a game. Let me repeat this: game. I rather doubt its goals were to educate the players as to how the world really works. On the contrary, I would guess that the goals were to entertain the players and persuade them that the experts (=those running the game) are very valuable and indispensable people. It was a PR exercise and Eliezer fell for it hook, line, and sinker.
The whole point of the Level B concept seems to be “Keep us in power or really bad things will happen”.
High variance
High variance is a good thing if you’re sleepwalking off a cliff already.
Most substantial variations from the equilibrium are disasters
Most random variations are disasters. Aren’t we lucky that most variations are not random?
This is an argument for stagnation and for never ever changing anything. Did EY become an old man already?
Eliezer is being an idiot who forgot his own maxim to not rely on fictional evidence. This was a game. Let me repeat this: game.
It was a simulation. Simulations can be incredibly powerful educational tools, when they reflect reality. These sorts of simulations are meant to reflect reality (in as much as is possible in a 4-hour window).
I rather doubt its goals were to educate the players as to how the world really works.
That is entirely the point of these simulations. I’ve been in them before, and helped construct a simulation once. The entire point of the exercise is to educate the players as to how the world really works. The designers, who themselves have relevant real world experience, spend a massive amount of time obsessing over this.
You’re doing content-free handwaving. “Can be”, “meant to”—sure, a lot of things are possible, but we’re talking about EY suddenly gaining a formative insight into geopolitics on the basis of a 4-hour simulation. Conveniently, it matches his existing political bias. I wonder if playing a couple of Civ games should shape your view of history… :-/
The entire point of the exercise
Are you talking in general or about the specific scenario playthrough that EY participated in? How do you know what is the entire point of that exercise was?
What I was trying to do is to point out that you might not be understanding the situation correctly. I think you are getting hung up on the word “game.” This is not a game in the sense of a video game, or even dungeons and dragons. Rather, it is standard practice within the national security apparatus to transfer knowledge by means of simulation of events, and that is what is being described here. This should not be surprising—the realm of governance is that of human actions and responses in a world constrained by time and limited data. A typical mistake of someone “too smart for their own good” is to assume that they can just get better data, or think their way out of a situation, when the data itself is confused and possibly suspect and deep consideration carries with it the cost of inaction and giving your opponent more time. These sorts of simulations teach both how decision making occurs under time and resource pressures, and the various common failure modes. It is an educational exercise designed to draw accurate references from.
I wonder if playing a couple of Civ games should shape your view of history… :-/
Absolutely not, because Civ is not meant to accurately simulate historical decision making; it’s a game meant for entertainment, not decision-theoretic education. Am I really failing to make this point?
Are you talking in general or about the specific scenario playthrough that EY participated in? How do you know what is the entire point of that exercise was?
In general because I’m not sure what specific exercise he went through. But as I did say, this is standard practice in the industry, and industry I have been involved in. EY says at the very beginning of the narrative that it was a “game” (read: simulation) constructed and run by national security insiders. That was the clue that he was talking about a training exercise and not some live action role playing nonsense.
you might not be understanding the situation correctly
Sigh. It would simplify things if you were to assume I’m not stupid. I did explicitly mention “games/simulations/scenario playthroughs”...
I’m not getting hung up on the word “game”. I’m getting hung up on it being not empirical reality.
standard practice within the national security apparatus to transfer knowledge
I don’t think EY holds a security clearance which makes any simulations he’s allowed to participate in… very incomplete. This still looks like a mostly PR exercise to me.
These sorts of simulations teach both how decision making occurs under time and resource pressures, and the various common failure modes.
I agree. Therefore you design the simulation to teach what you want the participants to learn. As I mentioned, accuracy is merely an instrumental goal. And if the participants are not actually people who would be expected to make such decisions, your goals might well be different from just transferring knowledge.
I don’t think EY holds a security clearance which makes any simulations he’s allowed to participate in… very incomplete.
I don’t know specifically that he has a clearance but there might be good reasons for him to have a clearance. The purposes of MIRI call for him being able to talk with people about classified AI projects.
Palantir has classified AI and Thiel is likely capable of making the necessary introductions.
It is and it was a distraction in getting to the core argument, for me at least.
High variance is a good thing if you’re sleepwalking off a cliff already.
In the range of possible ways to describe the status quo, from 0⁄100 to 100⁄100, sleepwalking off a cliff is oddly specific and on one end of an extreme assessment.
Isn’t the main argument that for most cases high variance is bad?
Sorry for that. But there could be a tower of meta-level, and if someone had invoked a level above level B, I wouldn’t have known how to call it if it was named level A.
This was a game. Let me repeat this: game. I rather doubt its goals were to educate the players as to how the world really works.
On the other hand, one safe bet is that a game is always less complicated than the real thing is trying to simulate. Without knowing the details, it’s impossible to tell if the game was a sort of “big diplomacy game with ex diplomats just for fun” or a Dem publicity stunt. While possible, assuming it for certain is unwarranted. Anyway, your argument would fall under the category: level B isn’t real / doesn’t matter.
Most random variations are disasters. Aren’t we lucky that most variations are not random?
In the mind of Eliezer, as far as we know, he considers Trump pretty clueless (level B speaking), so that necessarily all his moves would be random.
a game is always less complicated than the real thing is trying to simulate
I see no reason to believe that this game was trying to accurately simulate reality.
One notable difference is that in this game nothing was a stake. You launch your nukes, obliterate Moscow, get obliterated in return, and they all y’all go have a beer and discuss the whooshing sound that ICBMs make as they rise out of their silos.
he considers Trump pretty clueless (level B speaking), so that necessarily all his moves would be random.
Huh? Even under this assumption, since when clueless people behave randomly?
I see no reason to believe that this game was trying to accurately simulate reality.
Do you know something more about it that I don’t? Because neither I know if it was trying to simulate reality, and I also have no idea if it wasn’t, so it’s 1⁄2.
One notable difference is that in this game nothing was a stake.
That is a general argument against any game, simulation or drill.
Huh? Even under this assumption, since when clueless people behave randomly?
Well, nothing is onthologically random, but from a Bayesian perspective, random is something about which you have no information. So if Trump has no information about the state space of international equlibria, from the point of view of the system his actions will be random.
No, but my prior is neutral—I said “I have no reason to believe” otherwise. In general, I think that in most games/simulations/scenario playthroughs of this sort accurate simulation is merely an instrumental goal and the actual terminal goals vary.
That is a general argument against any game, simulation or drill.
Not so, you can play for stakes including high stakes.
So if Trump has no information about the state space of international equlibria, from the point of view of the system his actions will be random.
Don’t think in black and white. Even assuming Trump is clueless, he has some information about geopolitics and will act according to his goals and information available to him which is not zero.
Let me point out some examples of random moves: sending troops to occupy New Guinea; imposing trade sanctions on Kiribati; signing a mutual defence treaty with Uruguay. Do you think any of this is likely?
The inversion of the usual ranking is weird. ‘B’ is usually worse than ‘A’—B list, B movies, plan B—and here ‘B’ happens to be the supreme level of knowledge/operation...
Eliezer is being an idiot who forgot his own maxim to not rely on fictional evidence.
This was a game. Let me repeat this: game. I rather doubt its goals were to educate the players as to how the world really works. On the contrary, I would guess that the goals were to entertain the players and persuade them that the experts (=those running the game) are very valuable and indispensable people. It was a PR exercise and Eliezer fell for it hook, line, and sinker.
The whole point of the Level B concept seems to be “Keep us in power or really bad things will happen”.
High variance is a good thing if you’re sleepwalking off a cliff already.
Most random variations are disasters. Aren’t we lucky that most variations are not random?
This is an argument for stagnation and for never ever changing anything. Did EY become an old man already?
It was a simulation. Simulations can be incredibly powerful educational tools, when they reflect reality. These sorts of simulations are meant to reflect reality (in as much as is possible in a 4-hour window).
That is entirely the point of these simulations. I’ve been in them before, and helped construct a simulation once. The entire point of the exercise is to educate the players as to how the world really works. The designers, who themselves have relevant real world experience, spend a massive amount of time obsessing over this.
You’re doing content-free handwaving. “Can be”, “meant to”—sure, a lot of things are possible, but we’re talking about EY suddenly gaining a formative insight into geopolitics on the basis of a 4-hour simulation. Conveniently, it matches his existing political bias. I wonder if playing a couple of Civ games should shape your view of history… :-/
Are you talking in general or about the specific scenario playthrough that EY participated in? How do you know what is the entire point of that exercise was?
What I was trying to do is to point out that you might not be understanding the situation correctly. I think you are getting hung up on the word “game.” This is not a game in the sense of a video game, or even dungeons and dragons. Rather, it is standard practice within the national security apparatus to transfer knowledge by means of simulation of events, and that is what is being described here. This should not be surprising—the realm of governance is that of human actions and responses in a world constrained by time and limited data. A typical mistake of someone “too smart for their own good” is to assume that they can just get better data, or think their way out of a situation, when the data itself is confused and possibly suspect and deep consideration carries with it the cost of inaction and giving your opponent more time. These sorts of simulations teach both how decision making occurs under time and resource pressures, and the various common failure modes. It is an educational exercise designed to draw accurate references from.
Absolutely not, because Civ is not meant to accurately simulate historical decision making; it’s a game meant for entertainment, not decision-theoretic education. Am I really failing to make this point?
In general because I’m not sure what specific exercise he went through. But as I did say, this is standard practice in the industry, and industry I have been involved in. EY says at the very beginning of the narrative that it was a “game” (read: simulation) constructed and run by national security insiders. That was the clue that he was talking about a training exercise and not some live action role playing nonsense.
Sigh. It would simplify things if you were to assume I’m not stupid. I did explicitly mention “games/simulations/scenario playthroughs”...
I’m not getting hung up on the word “game”. I’m getting hung up on it being not empirical reality.
I don’t think EY holds a security clearance which makes any simulations he’s allowed to participate in… very incomplete. This still looks like a mostly PR exercise to me.
I agree. Therefore you design the simulation to teach what you want the participants to learn. As I mentioned, accuracy is merely an instrumental goal. And if the participants are not actually people who would be expected to make such decisions, your goals might well be different from just transferring knowledge.
I don’t know specifically that he has a clearance but there might be good reasons for him to have a clearance. The purposes of MIRI call for him being able to talk with people about classified AI projects.
Palantir has classified AI and Thiel is likely capable of making the necessary introductions.
It is and it was a distraction in getting to the core argument, for me at least.
In the range of possible ways to describe the status quo, from 0⁄100 to 100⁄100, sleepwalking off a cliff is oddly specific and on one end of an extreme assessment.
Isn’t the main argument that for most cases high variance is bad?
I have no idea. Which “most cases”? From whose point of view? There is no Law of Nature which states that high variance is bad.
Not to mention that for sufficiently fat-tailed distributions variance does not exist
Sorry for that. But there could be a tower of meta-level, and if someone had invoked a level above level B, I wouldn’t have known how to call it if it was named level A.
On the other hand, one safe bet is that a game is always less complicated than the real thing is trying to simulate.
Without knowing the details, it’s impossible to tell if the game was a sort of “big diplomacy game with ex diplomats just for fun” or a Dem publicity stunt. While possible, assuming it for certain is unwarranted.
Anyway, your argument would fall under the category: level B isn’t real / doesn’t matter.
In the mind of Eliezer, as far as we know, he considers Trump pretty clueless (level B speaking), so that necessarily all his moves would be random.
I see no reason to believe that this game was trying to accurately simulate reality.
One notable difference is that in this game nothing was a stake. You launch your nukes, obliterate Moscow, get obliterated in return, and they all y’all go have a beer and discuss the whooshing sound that ICBMs make as they rise out of their silos.
Huh? Even under this assumption, since when clueless people behave randomly?
Do you know something more about it that I don’t? Because neither I know if it was trying to simulate reality, and I also have no idea if it wasn’t, so it’s 1⁄2.
That is a general argument against any game, simulation or drill.
Well, nothing is onthologically random, but from a Bayesian perspective, random is something about which you have no information. So if Trump has no information about the state space of international equlibria, from the point of view of the system his actions will be random.
No, but my prior is neutral—I said “I have no reason to believe” otherwise. In general, I think that in most games/simulations/scenario playthroughs of this sort accurate simulation is merely an instrumental goal and the actual terminal goals vary.
Not so, you can play for stakes including high stakes.
Don’t think in black and white. Even assuming Trump is clueless, he has some information about geopolitics and will act according to his goals and information available to him which is not zero.
Let me point out some examples of random moves: sending troops to occupy New Guinea; imposing trade sanctions on Kiribati; signing a mutual defence treaty with Uruguay. Do you think any of this is likely?
You’ll have to ask Eliezer, I’m afraid. But I understand where you’re coming from, I guess neither Yudkowsky believed in such level of cluelessness.