“Then you may think that “Light is arglebargle” is a good explanation, that “arglebargle” is the correct password. It happened to me when I was nine years old—not because I was stupid, but because this is what happens by default. This is how human beings think, unless they are trained not to fall into the trap. Humanity stayed stuck in holes like this for thousands of years.”
Okay, but there’s one innocent interpretation even here. People learn language, and when we learn language we copy the verbal behavior of other people. Maybe “arglebargle” is a synonym for light in some language, or maybe it’s a supercategory of light (a category that includes light among other things). Maybe the teacher is still in the process of explaining to us what arglebargle means and the first step is to say that light is arglebargle—later on the teacher will tell us what else is arglebargle so that we will gradually build a good concept of it but initially we need to retain the point that light is arglebargle while not yet knowing what arglebargle is, because this is a step in learning what arglebargle is. In that case, we’re learning new language when we learn that “light is arglebargle”. That’s innocent, it’s not a mistake.
This suggests that the error may not be learning the teacher’s passwords per se, as such, but learning the teacher’s passwords when we should ideally be learning something else. The context matters.
But the student is the student and therefore ignorant by assumption, so it may in many cases be too much to expect the student to know when it is time to learn the passwords and when it is time to learn something else. If the student experiences academic success only from learning the passwords, then it may be that the student is not at fault, it’s the curriculum that is at fault—the teacher.
So the right recommendation may not be to tell the student to stop learning passwords. Passwords are a legitimate thing to learn, sometimes, and the student, being a student, doesn’t know ahead of time which times. The right recommendation may be to adjust the curriculum so that only the right kind of learning yields academic success.
This reminds me of my own experience as a student who loved chemistry. We were told a series of useful untruths about what matter is as we went through the system.
Molecules and atoms were like billiard balls.
No, that was an approximation—atoms are made of nuclei and electrons which can be visualised as little planetary systems.
No, that was an approximation—electrons, protons, neutrons are more usefully considered as probability functions.
I didn’t do science at university level, so I never got to the next level, but quantum theory was waiting for me there.
I did start an electronic engineering course, and there we learned another useful half-truth—the equations that describe the behaviour of a transistor. Only they don’t. They describe a manageable function which is something like the behaviour of a transistor—the real-world behaviour is non-linear and discontinuous (truly horrible—I didn’t finish the course...).
All of these useful untruths are like passwords—they allow us to reliably accomplish things in the world, but they do not give us real power over or understanding of the domain they address. Nevertheless, it would be hard to do without them.
Given how much got accomplished with prior models of the atom, I wouldn’t say these are necessarily good examples of passwords. They also weren’t approximations so much as older models. It’s sort of like learning the geocentric model first, and then later updating to the heliocentric model, and then finally learning that the sun actually revolves around the center of the universe as well.
I’m honestly a bit puzzled as to why we insist on teaching so many older models in science, without appropriately labeling them. Perhaps the math is easier to learn, and perhaps it’s just much easier to teach the models you grew up with originally.
Actually, when I learned these I learned them all at once, with the “older model” tag attached to them, and then I was given a “current model” that I was told that I wouldn’t understand yet, and so we worked with the planetary system thing.
That’s progress?
(You are trying to submit too fast. try again in 711 milliseconds. This website really values accuracy.)
Great example. I need to do something similar when teaching children photosynthesis. It’s helpful to start just by teaching a definition (when plants turn light into food). Do they need to learn about stomata and leaf anatomy? No, or at least not this year. That’s in next year’s textbook. If for now they can remember what is ‘photosynthesis,’ then their comprehension will be aided later.
I think what Yudkowsky is describing can be a problem sometimes when the point is thinking critically, not memorization of terms and definitions. Learning spanish is basically all about guessing the teacher’s password. In literature class, when we want to know what the White Whale is a symbol for, we don’t want a ‘password’ type response.
A password is a type of (usually partial) extensive definition (a list of the members of a set). What we want to teach is intensive definitions (the defining characteristics of sets). An extensive definition is not entirely useless as a learning aid, because an student could, in theory, work out the related intensive definition. Unfortunately, this is extraordinarily difficult when the definitions relate to wave dynamics, for example.
A password is an extensive definition being treated like the objective—a floating definition, where the intensive definition is no longer being sought. Extensive definitions do not restrict anticipation, and so can only ever be a step towards teaching intensive definitions. Learning passwords is wrong by definition—if it’s useful, it’s no longer a password.
In the case of a student learning a foreign language, providing extensive definitions is very useful because there is no difference between saying “the set {words-for-apple} includes apfel” and “apfel means apple, and the set {words-for-apple} contains all words meaning apple” - the intensive and extensive definitons are provided together, assuming the student knows what what an apple is.
“Then you may think that “Light is arglebargle” is a good explanation, that “arglebargle” is the correct password. It happened to me when I was nine years old—not because I was stupid, but because this is what happens by default. This is how human beings think, unless they are trained not to fall into the trap. Humanity stayed stuck in holes like this for thousands of years.”
Okay, but there’s one innocent interpretation even here. People learn language, and when we learn language we copy the verbal behavior of other people. Maybe “arglebargle” is a synonym for light in some language, or maybe it’s a supercategory of light (a category that includes light among other things). Maybe the teacher is still in the process of explaining to us what arglebargle means and the first step is to say that light is arglebargle—later on the teacher will tell us what else is arglebargle so that we will gradually build a good concept of it but initially we need to retain the point that light is arglebargle while not yet knowing what arglebargle is, because this is a step in learning what arglebargle is. In that case, we’re learning new language when we learn that “light is arglebargle”. That’s innocent, it’s not a mistake.
This suggests that the error may not be learning the teacher’s passwords per se, as such, but learning the teacher’s passwords when we should ideally be learning something else. The context matters.
But the student is the student and therefore ignorant by assumption, so it may in many cases be too much to expect the student to know when it is time to learn the passwords and when it is time to learn something else. If the student experiences academic success only from learning the passwords, then it may be that the student is not at fault, it’s the curriculum that is at fault—the teacher.
So the right recommendation may not be to tell the student to stop learning passwords. Passwords are a legitimate thing to learn, sometimes, and the student, being a student, doesn’t know ahead of time which times. The right recommendation may be to adjust the curriculum so that only the right kind of learning yields academic success.
That’s probably not easy.
This reminds me of my own experience as a student who loved chemistry. We were told a series of useful untruths about what matter is as we went through the system.
Molecules and atoms were like billiard balls.
No, that was an approximation—atoms are made of nuclei and electrons which can be visualised as little planetary systems.
No, that was an approximation—electrons, protons, neutrons are more usefully considered as probability functions.
I didn’t do science at university level, so I never got to the next level, but quantum theory was waiting for me there.
I did start an electronic engineering course, and there we learned another useful half-truth—the equations that describe the behaviour of a transistor. Only they don’t. They describe a manageable function which is something like the behaviour of a transistor—the real-world behaviour is non-linear and discontinuous (truly horrible—I didn’t finish the course...).
All of these useful untruths are like passwords—they allow us to reliably accomplish things in the world, but they do not give us real power over or understanding of the domain they address. Nevertheless, it would be hard to do without them.
Given how much got accomplished with prior models of the atom, I wouldn’t say these are necessarily good examples of passwords. They also weren’t approximations so much as older models. It’s sort of like learning the geocentric model first, and then later updating to the heliocentric model, and then finally learning that the sun actually revolves around the center of the universe as well.
I’m honestly a bit puzzled as to why we insist on teaching so many older models in science, without appropriately labeling them. Perhaps the math is easier to learn, and perhaps it’s just much easier to teach the models you grew up with originally.
Actually, when I learned these I learned them all at once, with the “older model” tag attached to them, and then I was given a “current model” that I was told that I wouldn’t understand yet, and so we worked with the planetary system thing.
That’s progress?
(You are trying to submit too fast. try again in 711 milliseconds. This website really values accuracy.)
I once read something like “This page was generated in 0.[fourteen digits] seconds.”
Great example. I need to do something similar when teaching children photosynthesis. It’s helpful to start just by teaching a definition (when plants turn light into food). Do they need to learn about stomata and leaf anatomy? No, or at least not this year. That’s in next year’s textbook. If for now they can remember what is ‘photosynthesis,’ then their comprehension will be aided later.
I think what Yudkowsky is describing can be a problem sometimes when the point is thinking critically, not memorization of terms and definitions. Learning spanish is basically all about guessing the teacher’s password. In literature class, when we want to know what the White Whale is a symbol for, we don’t want a ‘password’ type response.
A password is a type of (usually partial) extensive definition (a list of the members of a set). What we want to teach is intensive definitions (the defining characteristics of sets). An extensive definition is not entirely useless as a learning aid, because an student could, in theory, work out the related intensive definition. Unfortunately, this is extraordinarily difficult when the definitions relate to wave dynamics, for example.
A password is an extensive definition being treated like the objective—a floating definition, where the intensive definition is no longer being sought. Extensive definitions do not restrict anticipation, and so can only ever be a step towards teaching intensive definitions. Learning passwords is wrong by definition—if it’s useful, it’s no longer a password.
In the case of a student learning a foreign language, providing extensive definitions is very useful because there is no difference between saying “the set {words-for-apple} includes apfel” and “apfel means apple, and the set {words-for-apple} contains all words meaning apple” - the intensive and extensive definitons are provided together, assuming the student knows what what an apple is.