Yes, if you can’t do an unimportant thing X, I can’t judge whether you’ll be able to do important thing Y, because I don’t know what the relationship in difficulty is between X and Y.
But if you can’t do an easy thing, surely I can judge that you’ll have even more of a problem doing a difficult thing.
And there aren’t going to be many things in the world easier than cleaning your room or loading the dishwasher, right?
This feels very much like typical mind fallacy. Sure, for me, cleaning my room and loading the dishwasher are extremely easy, mindless things. But I know some people—my boyfriend, for instance—for whom household chores take up an undue amount of mental energy and are near-physically painful to do. On the other hand, he can happily spend hours trying to figure out a complex physics problem (while for me, this takes an undue amount of mental energy and is near-physically painful). Perhaps a more widespread example is reading books. Some people find it relaxing and do it all the time; other people have to exert a lot of mental violence to do it.
So, forcing my boyfriend to do ‘easy’ things like doing the laundry or reading a novel is going to be an uphill battle, but it could well be the case that doing AI safety research would, to him, feel like an endless stream of fun. I think that’s the point Katja’s making. Everybody’s different.
This presupposes that you know what the difficulty level is for the person in question. It also ignores a ton of stuff that can get between “easy thing” and “actual doing”, like what their priorities, interests, and abilities are.
Let’s say Bob has a really important project he needs to work on. He’s stuck and obsessed with it. Meanwhile, his room goes uncleaned and his dishwasher unloaded. He’s not accomplishing anything, but he’s not doing those simple things because he’s pouring energy into something else.
Now let’s consider Alice. Alice is a blind paraplegic computer programmer, who runs rings around her peers when it comes to coding. Programming for her is super easy, barely an inconvenience. But cleaning up the room or loading the dishwasher are not exactly her strengths.
And then there’s Carl. He spends hours playing video games at insanely high difficulty levels that nobody else can match. But putting away dishes is boring, and doesn’t get him that sweet sweet cred… or endorsement deals and advertising revenue. He’ll do it tomorrow, for sure. Maybe. Or maybe his mom will.
None of these people’s rooms are getting cleaned or dishwashers loaded, but that fact by itself tells you very little about what that person can accomplish. (After all, Bob could easily be a successful best-selling author who lets his place go to hell when he gets stuck in the middle of a book project.)
Those are all people who don’t really consider cleaning their room important (Alice, if she considered it important, could easily hire a cleaning service with her programmer salary).
I’m not talking about people who don’t clean up because they’re “pouring energy into something else” or because “putting away dishes is boring” or because they have a physical disability. I’m talking about the people from Katja’s post:
‘how can make a big difference to the world, when I can’t make a big difference to that pile of dishes in my sock drawer?’
This sounds to me like someone who wants to load the dishwasher, but finds themselves unable to. Like someone who’s frustrated with themselves; not someone who’s happy with the state of the affairs because they have better things to do.
In this case, I would expect this to be a pretty good predictor of not being able to do things that are more difficult (for an able-bodied person) than loading the dishwasher. And while there will not be much of a correlation between difficulty and importance, I would still say that virtually all non-trivial accomplishments in the world will be over the “loading the dishwasher” threshold of difficulty.
You’re overloading “want” here. If all of your sub-agents want to load a dishwasher, then surely you will load the dishwasher. If some of your sub-agents want to load a dishwasher, but need to get other sub-agents on board in order to do so, then you might not. It depends on how good your dishwasher agent is at recruiting the other agents. But this recruitment problem is not a subproblem of every other task you might care about.
Yes, if you can’t do an unimportant thing X, I can’t judge whether you’ll be able to do important thing Y, because I don’t know what the relationship in difficulty is between X and Y.
But if you can’t do an easy thing, surely I can judge that you’ll have even more of a problem doing a difficult thing.
And there aren’t going to be many things in the world easier than cleaning your room or loading the dishwasher, right?
This feels very much like typical mind fallacy. Sure, for me, cleaning my room and loading the dishwasher are extremely easy, mindless things. But I know some people—my boyfriend, for instance—for whom household chores take up an undue amount of mental energy and are near-physically painful to do. On the other hand, he can happily spend hours trying to figure out a complex physics problem (while for me, this takes an undue amount of mental energy and is near-physically painful). Perhaps a more widespread example is reading books. Some people find it relaxing and do it all the time; other people have to exert a lot of mental violence to do it.
So, forcing my boyfriend to do ‘easy’ things like doing the laundry or reading a novel is going to be an uphill battle, but it could well be the case that doing AI safety research would, to him, feel like an endless stream of fun. I think that’s the point Katja’s making. Everybody’s different.
This presupposes that you know what the difficulty level is for the person in question. It also ignores a ton of stuff that can get between “easy thing” and “actual doing”, like what their priorities, interests, and abilities are.
Let’s say Bob has a really important project he needs to work on. He’s stuck and obsessed with it. Meanwhile, his room goes uncleaned and his dishwasher unloaded. He’s not accomplishing anything, but he’s not doing those simple things because he’s pouring energy into something else.
Now let’s consider Alice. Alice is a blind paraplegic computer programmer, who runs rings around her peers when it comes to coding. Programming for her is super easy, barely an inconvenience. But cleaning up the room or loading the dishwasher are not exactly her strengths.
And then there’s Carl. He spends hours playing video games at insanely high difficulty levels that nobody else can match. But putting away dishes is boring, and doesn’t get him that sweet sweet cred… or endorsement deals and advertising revenue. He’ll do it tomorrow, for sure. Maybe. Or maybe his mom will.
None of these people’s rooms are getting cleaned or dishwashers loaded, but that fact by itself tells you very little about what that person can accomplish. (After all, Bob could easily be a successful best-selling author who lets his place go to hell when he gets stuck in the middle of a book project.)
Those are all people who don’t really consider cleaning their room important (Alice, if she considered it important, could easily hire a cleaning service with her programmer salary).
I’m not talking about people who don’t clean up because they’re “pouring energy into something else” or because “putting away dishes is boring” or because they have a physical disability. I’m talking about the people from Katja’s post:
This sounds to me like someone who wants to load the dishwasher, but finds themselves unable to. Like someone who’s frustrated with themselves; not someone who’s happy with the state of the affairs because they have better things to do.
In this case, I would expect this to be a pretty good predictor of not being able to do things that are more difficult (for an able-bodied person) than loading the dishwasher. And while there will not be much of a correlation between difficulty and importance, I would still say that virtually all non-trivial accomplishments in the world will be over the “loading the dishwasher” threshold of difficulty.
Does that make sense?
You’re overloading “want” here. If all of your sub-agents want to load a dishwasher, then surely you will load the dishwasher. If some of your sub-agents want to load a dishwasher, but need to get other sub-agents on board in order to do so, then you might not. It depends on how good your dishwasher agent is at recruiting the other agents. But this recruitment problem is not a subproblem of every other task you might care about.