I didn’t focus my post on this, but since you object, I’ll attempt to support my claim that “there are often large, unnecessary delays in theoretical discovery.” What I’m saying is that we can do this search in less time than we typically do.
That shouldn’t be surprising: there are all kinds of inefficiencies in theoretical discovery. Young researchers are incentivized to work on things that will get them tenure quickly, they often choose subject areas due to practical career concerns or personal interest rather than likely fruitfulness or social importance, they often pursue research out of momentum rather than due to an EV calculation, etc. This is all quite understandable, but it does introduce inefficiencies.
No, these are all spot-on criticisms (but I don’t think they are specific to theoretical research). I certainly agree that many problems of mainstream academia can be solved via the older patronage model, or perhaps even the newer crowdsourcing model.
I guess it is not clear to me that the failures in the OP’s list are due to a structural fault, or due to more excusable issues, like:
(a) “the scholarship coordination problem” (stuff gets forgotten and rediscovered over and over again, people don’t read other disciplines, etc.)
Yeah, as I said to joaolkf, I think my title was misleading: I wasn’t trying to contrast theoretical research with empirical research, but merely to look at these inefficiencies in the context of theoretical research, since that’s what MIRI does.
And you’re probably right that for any given example of apparently unnecessary delay in progress, it can be pretty hard to tell which inefficiencies deserve the most blame.
they often choose subject areas due to practical career concerns or personal interest rather than likely fruitfulness or social importance
Note that singling this out as a reason for inefficiency feels somewhat contradictory with the OP, where you suggested that there being no obvious profitable applications in the near-term was a reason for inefficiency. If people were choosing subjects areas based on “likely fruitfulness”, then we should expect areas with useful near-term applications to be prioritized.
I would not disagree. But then the claim seems trivial. Your comment’s second paragraph also applies to scientific research. Worse, in more applied areas graduate students have much less freedom to choose their own research topic and they seem to have a higher degree of overall social conformity.
We can then reinstate the question of what exactly about theoretical or philosophical research that makes it so particularly slow and unproductive. I say it is a huge, unknown search space, with no good search process and the fact that every time we find something, we lose it to another area. (Plus the fact that in the past our claims had political/practical consequences, as MIRI’s might have)
Oh I see what’s happening. Sorry, I think my title was accidentally misleading.
My post wasn’t trying to contrast the efficiency of theoretical research vs. empirical research. I just wanted to talk about those inefficiencies in the context of theoretical work specifically, since that’s what MIRI does. (E.g. I wanted to focus on examples from theoretical research.)
Anyway, the point about the large search space is an important one, and I hadn’t been thinking of the inefficiencies coming from political consequences until you mentioned it.
A book about Einstein and Godel claims both of them were able to identify a problem that became suddenly relevant and trackable due to other developments. I think there are certain ‘game changers’ that reshape discovery space producing low-hanging fruits. But, I do not think these low hanging fruits stay there for long. The possibility of AGI and X-Risks made some of your examples relevant, and they were addressed shortly after those game changers arose. But otherwise, some of your points seem similar to those on the Einstein-Godel book I read.
It is more of a biography of their friendship. I don’t think is worth reading. I almost summarized all his conclusions of the matter, except he applies it in more detail to history of science.
Ok. But you did say relative inefficiency. Relative to what? And still, I think many of your low hanging fruits were retrospective. I’m not sure that they were really obviously important and easy to obtain before, say, 1995.
One easy fix would be to just could about inviting some young possibly relevant philosophers for dinner and saying “do you see these 2 equally fun abstract problems? this one is more relevant because impacts the future of humanity!”
Ok, then the mistaken interpretation was my fault, you weren’t relevantly using the theoretical/applied dimension anywhere.
About decision theory. Perhaps utility maximizers were pulled towards game theory and thence economics and more narrow minded areas, while decision theory end up being maximized for oddness sometimes. That is, people who could attend to low hanging fruits were on areas where the background assumptions were unpopular,while people who could—perhaps—understand the background assumptions couldn’t care less for utility.
I didn’t focus my post on this, but since you object, I’ll attempt to support my claim that “there are often large, unnecessary delays in theoretical discovery.” What I’m saying is that we can do this search in less time than we typically do.
That shouldn’t be surprising: there are all kinds of inefficiencies in theoretical discovery. Young researchers are incentivized to work on things that will get them tenure quickly, they often choose subject areas due to practical career concerns or personal interest rather than likely fruitfulness or social importance, they often pursue research out of momentum rather than due to an EV calculation, etc. This is all quite understandable, but it does introduce inefficiencies.
Do you disagree?
No, these are all spot-on criticisms (but I don’t think they are specific to theoretical research). I certainly agree that many problems of mainstream academia can be solved via the older patronage model, or perhaps even the newer crowdsourcing model.
I guess it is not clear to me that the failures in the OP’s list are due to a structural fault, or due to more excusable issues, like:
(a) “the scholarship coordination problem” (stuff gets forgotten and rediscovered over and over again, people don’t read other disciplines, etc.)
(b) the standard exponential search for insight
Yeah, as I said to joaolkf, I think my title was misleading: I wasn’t trying to contrast theoretical research with empirical research, but merely to look at these inefficiencies in the context of theoretical research, since that’s what MIRI does.
And you’re probably right that for any given example of apparently unnecessary delay in progress, it can be pretty hard to tell which inefficiencies deserve the most blame.
Note that singling this out as a reason for inefficiency feels somewhat contradictory with the OP, where you suggested that there being no obvious profitable applications in the near-term was a reason for inefficiency. If people were choosing subjects areas based on “likely fruitfulness”, then we should expect areas with useful near-term applications to be prioritized.
I would not disagree. But then the claim seems trivial. Your comment’s second paragraph also applies to scientific research. Worse, in more applied areas graduate students have much less freedom to choose their own research topic and they seem to have a higher degree of overall social conformity.
We can then reinstate the question of what exactly about theoretical or philosophical research that makes it so particularly slow and unproductive. I say it is a huge, unknown search space, with no good search process and the fact that every time we find something, we lose it to another area. (Plus the fact that in the past our claims had political/practical consequences, as MIRI’s might have)
Oh I see what’s happening. Sorry, I think my title was accidentally misleading.
My post wasn’t trying to contrast the efficiency of theoretical research vs. empirical research. I just wanted to talk about those inefficiencies in the context of theoretical work specifically, since that’s what MIRI does. (E.g. I wanted to focus on examples from theoretical research.)
Anyway, the point about the large search space is an important one, and I hadn’t been thinking of the inefficiencies coming from political consequences until you mentioned it.
A book about Einstein and Godel claims both of them were able to identify a problem that became suddenly relevant and trackable due to other developments. I think there are certain ‘game changers’ that reshape discovery space producing low-hanging fruits. But, I do not think these low hanging fruits stay there for long. The possibility of AGI and X-Risks made some of your examples relevant, and they were addressed shortly after those game changers arose. But otherwise, some of your points seem similar to those on the Einstein-Godel book I read.
Which book?
http://www.ams.org/notices/200707/tx070700861p.pdf
It is more of a biography of their friendship. I don’t think is worth reading. I almost summarized all his conclusions of the matter, except he applies it in more detail to history of science.
Ok. But you did say relative inefficiency. Relative to what? And still, I think many of your low hanging fruits were retrospective. I’m not sure that they were really obviously important and easy to obtain before, say, 1995.
One easy fix would be to just could about inviting some young possibly relevant philosophers for dinner and saying “do you see these 2 equally fun abstract problems? this one is more relevant because impacts the future of humanity!”
Relative to financial markets, to which I was analogizing.
Ok, then the mistaken interpretation was my fault, you weren’t relevantly using the theoretical/applied dimension anywhere.
About decision theory. Perhaps utility maximizers were pulled towards game theory and thence economics and more narrow minded areas, while decision theory end up being maximized for oddness sometimes. That is, people who could attend to low hanging fruits were on areas where the background assumptions were unpopular,while people who could—perhaps—understand the background assumptions couldn’t care less for utility.