I would very much assume that you have a strong genetic disposition to be smart and curious.
Do you think unschooling would work acceptably well for kids who are not smart and curious?
tailcalled
Karma: 2,376
I learned the vast majority of my useful knowledge through autodidactism, everything else (school, university) is pretty much noise. I would be open to alternatives, but I haven’t seen any kind of “teaching” so far that came anywhere close.
Collaborating with an expert/getting tutoring from an expert might be really good?
Well, I don’t know who Ben Hoffman is,
He’s a rationalist(-adjacent?) blogger who writes about power, economics, culture, and EA: Compass Rose. His post Oppression and production are competing explanations for wealth inequality might be a good place to start.
Not sure whether setting this up as a related question that is hidden from the front page was the best approach. Maybe I should have selected that it should be posted to the front page instead. First time I’m using “Ask Related Question”.
[Question] How difficult is it for countries to change their school curriculum?
How does society decide what subjects get taught in school?
What would happen if society reinstated child labour?
Hm not sure such damage commonly happens.
Much of the material covered has no conceivable use except as a wasteful signal.
What would you think of the argument that getting taught a bundle of random things practices learning, so that those who have been taught in school are better able to learn other things afterwards?
I don’t necessarily expect there to be a black-and-white answer to my question, it’s mainly that I was reading Ben Hoffman and was thinking about how schools are a pretty central crux to his writings, yet after having unupdated my beliefs about schools, I wasn’t sure what to think of this crux, so I wanted some opinions from smart informed people that I could dig into or reflect upon.
There would be a second organization, not embedded inside the school, verifying that in fact students know the things that the school was aiming for them to know, that year and at least several years afterwards.
How would this second organization go about verifying that?
Well, it’s not obvious to me for one. In particular I am not sure what the alternative you propose would look like.
I’m not sure I trust The Case Against Education. I had once heard a review of it mention how the book debunked the notion that education teaches thinking skills. This interested me as I was trying to understand some things about how psychometrics works, so I skipped to that part of the book and looked at his references.
However, it turned out that the references were unconvincing. For instance, one of the main arguments was based on a small, old study that used an ad-hoc test of critical thinking skills. It was unclear to me how good that test was, and the study did not give any of the usual measures of goodness like internal reliability.
[Question] Is school good or bad?
I agree that something like this is plausible. To be more specific:
My favorite terms for it, which I mainly use in private thoughts and haven’t really written up much about elsewhere, is covariant concepts vs contravariant concepts. If you are familiar with type systems or category theory, you can think of it as being that sense. Basically, a contravariant concept is defined by something like a utility function or a boolean predicate or a set of constraints, while a covariant concept is defined by something like a distribution/measure or a convex set or a schematic. Covariant = output, contravariant = input, essentially. Or in terms of AI, generative models vs classifiers.
It does seem like some concepts are very anti-natural to define in a contravariant way, and very natural to define in a covariant way. In fact I would dare to go further and say that contravariant concepts usually cannot really be defined without a library of covariant concepts to define them in terms of, because you need something to “grab onto” reality, and contravariant concepts can’t really do that. This makes it seem quite plausible that covariance will play a critical role in aligned AI.
Some further things:
A simple way to handle covariant concepts is to take them as primitive. This is what e.g. generative latent variable models tend to do. The trouble is that in the real world, they are not primitive, but instead reduce into smaller-scale components. Sometimes we want the AIs to respect the sanctity of those components (e.g. don’t wirehead people) while other times we want the AI to intervene in the components (e.g. do cure aging). (And sometimes it’s controversial what we want it to do, e.g. with em uploads.) Taking them as primitive on one level of analysis does not really support novel interventions on the lower levels of analysis, nor does it really support respecting their sanctity on a higher level of analysis.
Covariance and contravariance is not totally absolute. The point of search, for instance, is basically to convert contravariant things to covariant ones.
A program that can act in the world to achieve good things would be a covariant representation of goodness. However, this program needs to be created in some way, and the world seems much to complex for engineers to handle each part. Instead it seems that we need some sort of learning process to capture it. But I have trouble thinking of learning processes that can plausibly effectively learn novel useful things without involving contravariant reasoning somewhere along the line. So I don’t think the trouble with contravariance can be avoided for alignment.
Not sure if this aligns 100% with your way of thinking about things but the framing might be helpful?
In your view, are values closer to recipes/instruction manuals than to utility functions?
You indicated that there was a problem with me doing a spot check on something not directly related to co-ops, so let me instead do a spot check on something directly related to co-ops:
Let’s look at what the literature indicates as the benefits of socialist firms:
[...] 3. Workers put in more effort[11]
I believe this is the cited study?
I’m confused about this study because the abstract seems to say that conventionally owned firms are better than co-ops:
In our experiments, subjects in conventionally-owned firms exhibited higher productivity, perceived greater fairness in the pay they received and the method used to pay them, reported higher levels of involvement in their tasks, had more positive evaluations of their supervisors, and showed a greater propensity to interact with and provide assistance to their co-workers than did those in employee-owned firms.
However, this seems to contradict table 3, where co-ops did better. So maybe they swapped around the labels in the abstract or something.
I’m also confused about what role the owner/supervisor was supposed to play in the work. It looks like the experimental simulation-company had workers do fairly independent tasks (spellchecking), and that there were not significant coordination or risk-management tasks for the owners to solve.
A spot check is supposed to take a number of random sources and check them, not pick the one claim you find most suspicious (that isn’t even about co-ops)
There seem to be three parts to this objections:
I did not spot-check sufficiently many claims
I filtered the claim to spot-check based on being the most suspicious
I did not filter the claims based on being about co-ops
With regards to point 1, I agree that I cannot know your accuracy very precisely without doing more checks, but the problem is that each check takes time and there are a lot of posts on the internet to read, so I have to limit how much I search.
With regards to point 2, it’s not that I spot-checked the most suspicious one, rather it’s that I spot-checked the first suspicious one. This is still a filter on suspiciousness, but a much weaker one. I think some filter on suspiciousness is appropriate since suspicious claims are also the ones I can learn the most from if they turn out to be true, as claims become suspicious through a combination of being unlikely and having big implications.
With regards to point 3, if you put much more effort into verifying the accuracy of your claims about co-ops than your claims about other stuff, then your accuracy on co-ops might not be that correlated with your accuracy on other stuff, and I ought to do a spot-check specifically on your claims about co-ops. I don’t know if that is true. If it is true, it might also be helpful to mention it as a disclaimer in the post so people know what claims to mostly focus on.
and use that to dismiss the entire literature on co-ops.
So it’s not so much that I’m dismissing the entire literature on co-ops. (Or well, I would generally dismiss any social science that I haven’t done some surface checks of. But that’s different from my comments here.) It’s more that I’m dismissing your literature review of the literature.
This is dishonest, the vast majority of the sources are primary scientific studies and the few times I do refer to secondary sources it isn’t irrelevant.
I wasn’t talking about the vast majority of the sources, I was talking about source 3, which turned into source 2, which turned into some other source that I had to find myself.
You did handle it right, especially your deleted comment.
My deleted comment was nearly identical to one of the non-deleted comments. It’s just that I realized there was a problem with one of my comments after posting it and I needed to take some time to look at it.
Yeah, because the primary source is right there?! What value would me explaining in my second language bring to the explanation, when you can click on the link and immediately download the primary source?
The problem is that the primary source does not seem credible without additional information.
When people do something, they tend to become better at that thing by picking up tricks relevant to it. If the thing they are doing is learning lots of random things, presumably some of the tricks they pick up would be tricks for learning lots of random things.
I don’t know. I’ve talked with some people who are interested in intelligence research about how to measure learning ability. It would essentially require measuring people’s ability to do lots of things, then teaching them those things, then measuring their ability on those things again, and looking at something like the difference in ability. The trouble is that it is simultaneously really expensive to perform such measurements (as having to teach people things makes it orders of magnitude more expensive than ordinary psychometrics), and yet still too noisy when performed at reasonable scales to be useful.
So measuring learning ability would be difficult. And even if we found out how to do that, we would still need some sort of randomized trial or natural experiment to test school’s effect on learning ability.
Maybe. This assumes ability to learn when younger is as valueable as ability to learn when older, which might not be true because you have much more information about what you need to learn when you are older. For instance at my job I had to learn KQL, but KQL did not exist when I was a child, so in order to teach it to me as a child, we would have to be able to accurately forecast the invention of KQL, which seems impossible.
I suspect it depends on the person.
The sort of person who watches science documentaries on TV, who builds redstone computers in Minecraft, who reads LessWrong and scientific papers when surfing the web, and who talks with friends about topics like the theoretical arguments for and against school would probably have a much more intellectually stimulating environment outside of school than within it.
But such people are extremely rare, so we can to good approximation say they don’t exist. I’m less sure about how it would work out for the median person, who spends their time on other stuff. I think they might tend to learn things that are less intellectually varied, specializing deeply into keeping track of social relations, doing exciting things, or similar? Idk, I don’t know very much about the median person.