I think ideas get progressively harder to find within any given field as it matures. However, when we create new fields or find new breakthrough technologies, it opens up whole new orchards of low-hanging fruit.
When the Web was created, there were lots of new ideas that were easy to find: “put X on the web” for many values of X. After penicillin was invented, there was a similar golden age of antibiotics: “check out X mold or Y soil sample and check it for effectiveness against Z disease”. At times like this you see very rapid progress in certain applications.
Similarly, imagine if we got atomically precise manufacturing (APM). There would be a whole set of easy-to-find ideas: “manufacture X using APM.” Or if we got an easy way to understand and manipulate genes, there would be a set of easy-to-find ideas of the form “edit X gene to cure Y disease or enhance Z trait.”
I think the Great Stagnation is not a failure to extract all the value from existing fields, but rather a failure to open up new fields, to have new breakthroughs decades ago.
Thank you for these interesting answers. Do you think the creation of new fields is also subject to diminishing returns? e.g. are new fields harder to find as well? Or do you think that only technologies are subject to diminishing returns?
On this note, do you think progress is likely to be open to us indefinitely, or would you expect that eventually we will reach a level of technological maturity where all meaningful low-hanging fruit (be they individual technologies or S curves) have been picked and there is little further technological progress? If so, why? If not, why not?
“Are new fields getting harder to find?” I think this is the trillion-dollar question! I don’t have an answer yet though.
Is progress open indefinitely? I think there is probably at least a theoretic end to progress, but it’s so unimaginably far away that for our purposes today we should consider progress as potentially infinite. There are still an enormous number of things to learn and invent.
Quick thought here Jack and Jason (caveat—haven’t thought about this much at all!).
Yes, the creation of new fields is important. However, even if there are diminishing returns to new fields (sidenote—I’ve been thinking about ways to try and measure this empirically), what’s more important is the applicability of the new field to existing fields.
For example, even if we only create one new field but that field could be incredibly powerful. For example, APM (atomically precise manufacturing), or an AGI of some sorts, then it will have major ramifications on progress across all fields.
However, if we created a lot of new insignificant fields, then even if we create hundreds of them, progress won’t be substantially improved across other domains.
I guess what I’m trying to say is the emphasis is not just on new fields per se.
Thanks for doing this Jason. I agree with your response here. Seems natural to think that there are diminishing marginal returns to ideas within a sector.
You mention APM, which would spur progress in other sectors. Are there ways to identify which sectors open up progress in other domains, i.e. identifying the ideas that could remove the constraining factors of progress (small and big)?
I think basically you have to look at where an innovation sits in the tech tree.
Energy technologies tend to be fundamental enablers of other sectors. J. Storrs Hall makes a good case for the need to increase per-capita energy usage, which he calls the Henry Adams Curve: https://rootsofprogress.org/where-is-my-flying-car
But also, a fundamentally new way to do manufacturing, transportation, communication, or information processing would enable a lot of downstream progress.
I agree with Jason about the S-curves and the importance of distinguishing between within-area progress and between-area progress and he’s making some really good points about ways to think about these issues in the linked posts. I also have a giant essay about this paper coming out soon and I’m very skeptical of its findings—lmk if you’d be interested in reading the draft
In particular, I don’t think you can measure “research productivity” as percent improvement divided by absolute research input. I understand the rationale for measuring it this way, but I think for reasons Scott points out, it’s just not the right metric to use.
Another way to look at this is: one generative model for exponential growth is a thing that is growing in proportion to its size. One way this can happen is that the growing thing invests a constant portion of its resources into growth. But in that model, you expect to see the resources used for growth to be exponentially increasing. IMO this is what we see with R&D.
Another place you can see this is in the growth of a startup. Startups can often grow revenue exponentially, but they also hire exponentially. If you used a similar measure of “employee productivity” parallel to “research productivity”, then you’d say it is going down, because an increasing number of employees is needed to maintain a constant % increase in revenue.
Further, what these examples ought to make clear is that exponentially increasing inputs to create exponential growth is actually totally sustainable. So, I don’t see it as a cause for alarm at all, but rather (as Scott says) the natural order of things.
Alexey, I’m also skeptical of the findings but haven’t had time to dig deeper yet, so it’s just hunches at the moment. I have already asked you for the draft :). Honestly, can’t wait to read it since you announced it last week!
Are ideas getting harder to find?
I think ideas get progressively harder to find within any given field as it matures. However, when we create new fields or find new breakthrough technologies, it opens up whole new orchards of low-hanging fruit.
When the Web was created, there were lots of new ideas that were easy to find: “put X on the web” for many values of X. After penicillin was invented, there was a similar golden age of antibiotics: “check out X mold or Y soil sample and check it for effectiveness against Z disease”. At times like this you see very rapid progress in certain applications.
Similarly, imagine if we got atomically precise manufacturing (APM). There would be a whole set of easy-to-find ideas: “manufacture X using APM.” Or if we got an easy way to understand and manipulate genes, there would be a set of easy-to-find ideas of the form “edit X gene to cure Y disease or enhance Z trait.”
I think the Great Stagnation is not a failure to extract all the value from existing fields, but rather a failure to open up new fields, to have new breakthroughs decades ago.
Further reading: https://rootsofprogress.org/teasing-apart-the-s-curves
Also: https://rootsofprogress.org/where-is-my-flying-car
Thank you for these interesting answers. Do you think the creation of new fields is also subject to diminishing returns? e.g. are new fields harder to find as well? Or do you think that only technologies are subject to diminishing returns?
On this note, do you think progress is likely to be open to us indefinitely, or would you expect that eventually we will reach a level of technological maturity where all meaningful low-hanging fruit (be they individual technologies or S curves) have been picked and there is little further technological progress? If so, why? If not, why not?
“Are new fields getting harder to find?” I think this is the trillion-dollar question! I don’t have an answer yet though.
Is progress open indefinitely? I think there is probably at least a theoretic end to progress, but it’s so unimaginably far away that for our purposes today we should consider progress as potentially infinite. There are still an enormous number of things to learn and invent.
Quick thought here Jack and Jason (caveat—haven’t thought about this much at all!).
Yes, the creation of new fields is important. However, even if there are diminishing returns to new fields (sidenote—I’ve been thinking about ways to try and measure this empirically), what’s more important is the applicability of the new field to existing fields.
For example, even if we only create one new field but that field could be incredibly powerful. For example, APM (atomically precise manufacturing), or an AGI of some sorts, then it will have major ramifications on progress across all fields.
However, if we created a lot of new insignificant fields, then even if we create hundreds of them, progress won’t be substantially improved across other domains.
I guess what I’m trying to say is the emphasis is not just on new fields per se.
Thanks for doing this Jason. I agree with your response here. Seems natural to think that there are diminishing marginal returns to ideas within a sector.
You mention APM, which would spur progress in other sectors. Are there ways to identify which sectors open up progress in other domains, i.e. identifying the ideas that could remove the constraining factors of progress (small and big)?
I think basically you have to look at where an innovation sits in the tech tree.
Energy technologies tend to be fundamental enablers of other sectors. J. Storrs Hall makes a good case for the need to increase per-capita energy usage, which he calls the Henry Adams Curve: https://rootsofprogress.org/where-is-my-flying-car
But also, a fundamentally new way to do manufacturing, transportation, communication, or information processing would enable a lot of downstream progress.
I agree with Jason about the S-curves and the importance of distinguishing between within-area progress and between-area progress and he’s making some really good points about ways to think about these issues in the linked posts. I also have a giant essay about this paper coming out soon and I’m very skeptical of its findings—lmk if you’d be interested in reading the draft
Oh, I should also point to the SSC response to “ideas getting harder to find”, which I thought was very good: https://slatestarcodex.com/2018/11/26/is-science-slowing-down-2/
In particular, I don’t think you can measure “research productivity” as percent improvement divided by absolute research input. I understand the rationale for measuring it this way, but I think for reasons Scott points out, it’s just not the right metric to use.
Another way to look at this is: one generative model for exponential growth is a thing that is growing in proportion to its size. One way this can happen is that the growing thing invests a constant portion of its resources into growth. But in that model, you expect to see the resources used for growth to be exponentially increasing. IMO this is what we see with R&D.
Another place you can see this is in the growth of a startup. Startups can often grow revenue exponentially, but they also hire exponentially. If you used a similar measure of “employee productivity” parallel to “research productivity”, then you’d say it is going down, because an increasing number of employees is needed to maintain a constant % increase in revenue.
Further, what these examples ought to make clear is that exponentially increasing inputs to create exponential growth is actually totally sustainable. So, I don’t see it as a cause for alarm at all, but rather (as Scott says) the natural order of things.
Alexey, I’m also skeptical of the findings but haven’t had time to dig deeper yet, so it’s just hunches at the moment. I have already asked you for the draft :). Honestly, can’t wait to read it since you announced it last week!