I thought you gave superb explanations for the intuitions underlying both the burden-of-knowledge thesis and the institutional decay thesis. But then you sort of just assert that the institutional decay thesis seems more plausible to you, even though it seemed from your own presentation like very strong arguments can be mustered on both sides with neither having a knock-down! Can you suggest some experiment or observational result that could clinch the matter one way or the other?
One big observational result is that over the past several centuries progress has been accelerating. That suggests there are forces which cut against the burden of knowledge and that those forces were dominant for several centuries. In the past 100 years growth rates have been ~constant or slightly decreasing in the US. It's possible that this is just where the curves of decreasing returns to scale and the increasing burden of knowledge cross but it seems less likely.
The burden of knowledge also doesn't explain cross-country differences in productivity growth or negative productivity growth. The burden of knowledge isn't higher in Italy than South Korea or the US but there are huge differences in their productivity growth rates. I think that institutional differences are a good explanation.
Similarly, growth rates in Italy are negative but this isn't because they're forgetting ideas. It's because they are getting worse at deploying the ideas.
Say we're pretty confident that institutional differences can explain cross-country differences in productivity growth. Then when we observe within-country over-time differences in productivity growth of a similar magnitude to the cross-country differences and we also observe large institutional changes I think it's reasonable to attribute the change over time to institutions.
Thank you for reading and commenting, really appreciate it!!
- To whatever extent there’s been a slowdown in scientific progress, it seems to have happened similarly all over the world. It’s not as if there’s one country still making fundamental advances in a way that no other country can understand or replicate.
- You give excellent examples of where the burden of knowledge got *lessened* over time, as with the replacement of the Ptolemaic by the Copernican system. But I notice that all of your examples were from several hundred years ago! Might it be that, as we learn more about what makes a good explanatory theory, we unfortunately also add to the burden of knowledge, since fewer of our theories can then be radically simplified or discarded? Our current understanding of atomic physics and chemistry seems like an excellent candidate for something that could just be permanent.
- More generally, there might be different stories about the rate of progress in different areas of science. In pure math, there doesn’t seem to be much reason to have expected a slowdown in the past 50 years, and indeed there doesn’t seem to have been a slowdown. In particle physics, by contrast, we reached a point where the accelerators basically hit the limit of what civilization was willing to fund, and that coincided with what many would say WAS a slowdown, and maybe that’s had knock-on effects for other areas as well.
1. That's a fair observation which cuts against the institutional explanation; we see a worldwide trend but institutions have a massive variation across the world so it's hard to find a worldwide commonality between them that can explain a worldwide trend.
However, I think this relies on a claim that potential scientific progress is thickly distributed all around the world so that we have hundreds of chances to combine good institutions and scientific potential in lots of countries. I think that scientific potential is actually clustered in just a few places so that we only have a couple of chances to combine it with good enough institutions. E.g some local variation in the institutions of Florence or London in the 15th or 18th centuries would have worldwide effects on progress because so much of the world's potential was concentrated in those places. Things are perhaps less concentrated today but you could capture most of it with the US, UK, Japan, and China. So I think a more local institutions based explanation can work because we only need to change things in these few places to have a worldwide effect. Plus, the institutions of the US, UK, and Japan are highly correlated.
2. Ha! :) Fair enough, I do tend to reach for historical examples. I think time and separation makes these things easier to see. For a more recent example I would think about computing. This also has to do with my specialization point, but it seems very possible to make big advances in this field without learning the mathematical and engineering background. E.g Steve Jobs didn't have learn about turing machines or transistors to make the iPhone. There also may be something happening currently with computational biology. If computer scientists can come in and make big progress in biology without much field-specific training I would call that a lessening of the burden of knowledge.
3. I agree with this. I think it would be surprising if both the drags and accelerants of science did not vary by field.
I would like to know this too! You note that metascience could be an avenue for improving institutional decay. Are there specific interventions we could try?
Someone (I thought Scott Alexander but I can't find it in his archives) proposed a model where researchers are foraging for ideas on a landscape, and as more and more ideas get discovered, you have to travel farther to find new ones. I think they originally thought of "traveling farther" as the burden of knowledge problem, but it doesn't have to be - you can think of it as just "you have to think for a really long time to come up with something genuinely new" or something like that.
If that's the case, you'd still have the ideas-getting-harder-to-find problem without any of the burden-of-knowledge indicators you mentioned, and without institutional decay playing a big role.
If you have not done so, I would recommend reading "Range" by Daniel Epstein. He makes some interesting observations as to how generalists can make real contributions in a world dominated by specialists. I think he conclusions are very applicable to Progress Studies.
I think much the same problems are going on in social science and history. I believe new fields of inquiry that build upon the findings of multiple disciplines are necessary. This enables us to see "old data" in new ways by making hither to unnoticed connections.
Given the current state of academia, however, this is unlikely to happen in universities. We need to find ways to do so via digital technologies.
I believe that is exactly what Progress Studies should be trying to do. Here is a series of posts on how we can do that:
Here is one models on how the burden can still show up: https://www.strangeloopcanon.com/p/innovation . The question of how much you need to learn before making a breakthrough is dependent on how quickly and well you can learn, and that "node capacity" can manifest as a burden of knowledge
You are absolutely correct that learning the fundamentals of a field is necessary for a breakthrough. The key knowledge, however, is understanding what the most important problems in the field that need to be solved. Then understanding the characteristics of that problem so one can look for potential solutions that have already been applied. This is called "analogic" thinking.
It is very likely that a similar problem has already been solved in a different field, but none of the specialists realize it. I believe the way to overcome that is to spend 80% of your time learning specialized knowledge, but the other 20% in general knowledge in many different fields. Even just reading university textbooks can achieve that basic knowledge. It is not hard, it just takes time.
It is very likely that the next breakthrough will come from applying a solution from one of those other fields to your chosen field.
I'm extremely attracted to the idea that university science is the barrier to success. It sucks. But if so why haven't drug compabies smashed progress out of the park? They have none of the same intitutional problems. Sure, there's been progress via these companies, like glp-1. But not as much as you'd expect.
I imagine it has something to do with the connection between academia and drug companies. After all, drug companies hire scientists that got their PhDs in academia (after 9 years in academic institutions!) I read somewhere that many industry scientists also rely on basic research from academia but that these are often impossible to replicate (see the following for some related evidence: https://www.bmj.com/content/344/bmj.e2555 ).
I have occasionally thought that organisations should be subject to random accidents, as humans are. And/or, they should have a fixed but indeterminate lifespan, as humans do, with the chance of death increasing yearly after forty years or so. Incentives in academia would be a little different, I think.
I thought you gave superb explanations for the intuitions underlying both the burden-of-knowledge thesis and the institutional decay thesis. But then you sort of just assert that the institutional decay thesis seems more plausible to you, even though it seemed from your own presentation like very strong arguments can be mustered on both sides with neither having a knock-down! Can you suggest some experiment or observational result that could clinch the matter one way or the other?
One big observational result is that over the past several centuries progress has been accelerating. That suggests there are forces which cut against the burden of knowledge and that those forces were dominant for several centuries. In the past 100 years growth rates have been ~constant or slightly decreasing in the US. It's possible that this is just where the curves of decreasing returns to scale and the increasing burden of knowledge cross but it seems less likely.
The burden of knowledge also doesn't explain cross-country differences in productivity growth or negative productivity growth. The burden of knowledge isn't higher in Italy than South Korea or the US but there are huge differences in their productivity growth rates. I think that institutional differences are a good explanation.
Similarly, growth rates in Italy are negative but this isn't because they're forgetting ideas. It's because they are getting worse at deploying the ideas.
Say we're pretty confident that institutional differences can explain cross-country differences in productivity growth. Then when we observe within-country over-time differences in productivity growth of a similar magnitude to the cross-country differences and we also observe large institutional changes I think it's reasonable to attribute the change over time to institutions.
Thank you for reading and commenting, really appreciate it!!
Thanks!! Three quick thoughts:
- To whatever extent there’s been a slowdown in scientific progress, it seems to have happened similarly all over the world. It’s not as if there’s one country still making fundamental advances in a way that no other country can understand or replicate.
- You give excellent examples of where the burden of knowledge got *lessened* over time, as with the replacement of the Ptolemaic by the Copernican system. But I notice that all of your examples were from several hundred years ago! Might it be that, as we learn more about what makes a good explanatory theory, we unfortunately also add to the burden of knowledge, since fewer of our theories can then be radically simplified or discarded? Our current understanding of atomic physics and chemistry seems like an excellent candidate for something that could just be permanent.
- More generally, there might be different stories about the rate of progress in different areas of science. In pure math, there doesn’t seem to be much reason to have expected a slowdown in the past 50 years, and indeed there doesn’t seem to have been a slowdown. In particle physics, by contrast, we reached a point where the accelerators basically hit the limit of what civilization was willing to fund, and that coincided with what many would say WAS a slowdown, and maybe that’s had knock-on effects for other areas as well.
1. That's a fair observation which cuts against the institutional explanation; we see a worldwide trend but institutions have a massive variation across the world so it's hard to find a worldwide commonality between them that can explain a worldwide trend.
However, I think this relies on a claim that potential scientific progress is thickly distributed all around the world so that we have hundreds of chances to combine good institutions and scientific potential in lots of countries. I think that scientific potential is actually clustered in just a few places so that we only have a couple of chances to combine it with good enough institutions. E.g some local variation in the institutions of Florence or London in the 15th or 18th centuries would have worldwide effects on progress because so much of the world's potential was concentrated in those places. Things are perhaps less concentrated today but you could capture most of it with the US, UK, Japan, and China. So I think a more local institutions based explanation can work because we only need to change things in these few places to have a worldwide effect. Plus, the institutions of the US, UK, and Japan are highly correlated.
2. Ha! :) Fair enough, I do tend to reach for historical examples. I think time and separation makes these things easier to see. For a more recent example I would think about computing. This also has to do with my specialization point, but it seems very possible to make big advances in this field without learning the mathematical and engineering background. E.g Steve Jobs didn't have learn about turing machines or transistors to make the iPhone. There also may be something happening currently with computational biology. If computer scientists can come in and make big progress in biology without much field-specific training I would call that a lessening of the burden of knowledge.
3. I agree with this. I think it would be surprising if both the drags and accelerants of science did not vary by field.
I would like to know this too! You note that metascience could be an avenue for improving institutional decay. Are there specific interventions we could try?
Someone (I thought Scott Alexander but I can't find it in his archives) proposed a model where researchers are foraging for ideas on a landscape, and as more and more ideas get discovered, you have to travel farther to find new ones. I think they originally thought of "traveling farther" as the burden of knowledge problem, but it doesn't have to be - you can think of it as just "you have to think for a really long time to come up with something genuinely new" or something like that.
If that's the case, you'd still have the ideas-getting-harder-to-find problem without any of the burden-of-knowledge indicators you mentioned, and without institutional decay playing a big role.
aha: https://www.astralcodexten.com/p/the-low-hanging-fruit-argument-models?s=r
Excellent article.
If you have not done so, I would recommend reading "Range" by Daniel Epstein. He makes some interesting observations as to how generalists can make real contributions in a world dominated by specialists. I think he conclusions are very applicable to Progress Studies.
Here is a summary of the book:
https://techratchet.com/2020/03/13/book-summary-range-why-generalist-triumph-in-a-specialized-world-by-daniel-epstein/
I think much the same problems are going on in social science and history. I believe new fields of inquiry that build upon the findings of multiple disciplines are necessary. This enables us to see "old data" in new ways by making hither to unnoticed connections.
Given the current state of academia, however, this is unlikely to happen in universities. We need to find ways to do so via digital technologies.
I believe that is exactly what Progress Studies should be trying to do. Here is a series of posts on how we can do that:
https://frompovertytoprogress.substack.com/p/what-is-progress-studies
Here is one models on how the burden can still show up: https://www.strangeloopcanon.com/p/innovation . The question of how much you need to learn before making a breakthrough is dependent on how quickly and well you can learn, and that "node capacity" can manifest as a burden of knowledge
You are absolutely correct that learning the fundamentals of a field is necessary for a breakthrough. The key knowledge, however, is understanding what the most important problems in the field that need to be solved. Then understanding the characteristics of that problem so one can look for potential solutions that have already been applied. This is called "analogic" thinking.
It is very likely that a similar problem has already been solved in a different field, but none of the specialists realize it. I believe the way to overcome that is to spend 80% of your time learning specialized knowledge, but the other 20% in general knowledge in many different fields. Even just reading university textbooks can achieve that basic knowledge. It is not hard, it just takes time.
It is very likely that the next breakthrough will come from applying a solution from one of those other fields to your chosen field.
I'm extremely attracted to the idea that university science is the barrier to success. It sucks. But if so why haven't drug compabies smashed progress out of the park? They have none of the same intitutional problems. Sure, there's been progress via these companies, like glp-1. But not as much as you'd expect.
I imagine it has something to do with the connection between academia and drug companies. After all, drug companies hire scientists that got their PhDs in academia (after 9 years in academic institutions!) I read somewhere that many industry scientists also rely on basic research from academia but that these are often impossible to replicate (see the following for some related evidence: https://www.bmj.com/content/344/bmj.e2555 ).
Thanks for the great content Maxwell! Found your blog through Malcolm Cochran (https://antheros.blog/). I wrote a post largely as a response to this one: https://nematobe.substack.com/p/science-friction-is-the-burden-of
I have occasionally thought that organisations should be subject to random accidents, as humans are. And/or, they should have a fixed but indeterminate lifespan, as humans do, with the chance of death increasing yearly after forty years or so. Incentives in academia would be a little different, I think.