1/Here are my thoughts about whether science is slowing down, what that means, and what we can do about it.
@patrickc and @michael_nielsen recently wrote an Atlantic article claiming that science is slowing down: theatlantic.com/science/archiv…
@patrickc and @michael_nielsen recently wrote an Atlantic article claiming that science is slowing down: theatlantic.com/science/archiv…
2/This article has been receiving pushback from scientists, which ishardly surprising...
https://twitter.com/neurobongo/status/1063879812773748737
3/So, let's take a look at the Atlantic article's methodology.
The authors surveyed a bunch of scientists and asked them to rank Nobel-winning discoveries in their field by importance. The result: a flat or even declining rate of "important" discoveries.
The authors surveyed a bunch of scientists and asked them to rank Nobel-winning discoveries in their field by importance. The result: a flat or even declining rate of "important" discoveries.
4/There are at least two big problems I see with this methodology.
The first is that Nobel Prizes are rate-limited - there can only be up to 3 per year, usually for just one or two discoveries.
The first is that Nobel Prizes are rate-limited - there can only be up to 3 per year, usually for just one or two discoveries.
5/Suppose scientific progress were actually accelerating.
That doesn't necessarily mean you'd see an increasing *level* of importance for the Nobel-winning discoveries. You might just see an increasing *number* of important discoveries, many of which couldn't win Nobels.
That doesn't necessarily mean you'd see an increasing *level* of importance for the Nobel-winning discoveries. You might just see an increasing *number* of important discoveries, many of which couldn't win Nobels.
6/Accelerating progress would therefore generate a backlog of Nobel-worthy discoveries, causing the Nobel committee to award prizes to older and older discoveries.
Interestingly, the committee *has* been awarding prizes to older and older discoveries, as the article notes!
Interestingly, the committee *has* been awarding prizes to older and older discoveries, as the article notes!
7/The authors interpret the paucity of prizes for discoveries since 1990 as evidence of declining #'s of important discoveries.
But it could indicate the committee scrambling to award prizes to a huge # of important discoveries before the discoverers die (and become ineligible)!
But it could indicate the committee scrambling to award prizes to a huge # of important discoveries before the discoverers die (and become ineligible)!
8/The second methodological problem is that a scientific discovery's actual, real importance is not fixed in time.
New discoveries build on old ones. Each new discovery makes the old discoveries it's based on grow in actual importance.
New discoveries build on old ones. Each new discovery makes the old discoveries it's based on grow in actual importance.
9/Consider all the recent discoveries that MIGHT lead to great things, or MIGHT be relatively dead-endish.
Will CRISPR allow ubiquitous safe genetic engineering? Or will it be a mostly useless tool due to massive side effects? We don't know yet.
Will CRISPR allow ubiquitous safe genetic engineering? Or will it be a mostly useless tool due to massive side effects? We don't know yet.
10/EVEN IF scientists are capable of accurately assessing a discovery's importance, graphs of importance vs. time are naturally biased against recent discoveries.
It'll be decades before we know what discoveries from the 2010s really changed the world.
It'll be decades before we know what discoveries from the 2010s really changed the world.
11/So I believe the Atlantic article's methodology is deeply flawed.
But that said, I believe their conclusion is probably true - at least, in specific fields.
Because we have other evidence.
bloomberg.com/opinion/articl…
But that said, I believe their conclusion is probably true - at least, in specific fields.
Because we have other evidence.
bloomberg.com/opinion/articl…
12/A recent paper by Bloom, Jones, Van Reenen, and Webb looks at specific scientific fields, and tries to measure research productivity - i.e., discoveries per researcher - for each one.
The find declining productivity in each field.
web.stanford.edu/~chadj/IdeaPF.…
The find declining productivity in each field.
web.stanford.edu/~chadj/IdeaPF.…
13/The authors are careful. They use a variety of measures of discoveries.
Real measures like Moore's Law, crop yields, drug discoveries, etc.
Measures of company performance in various fields, patents, etc.
All measures show slowing productivity-per-researcher in all fields.
Real measures like Moore's Law, crop yields, drug discoveries, etc.
Measures of company performance in various fields, patents, etc.
All measures show slowing productivity-per-researcher in all fields.
14/Now, does falling research productivity mean science as a whole is slowing down? Or that our investment in science has become uneconomical?
NO. It does not!
Read this post:
bradford-delong.com/2018/08/nichol…
NO. It does not!
Read this post:
bradford-delong.com/2018/08/nichol…
15/So why DO we care about slowing within-field productivity in scientific fields?
Because it's a sign that we need to actively search for new fields to open up.
bloomberg.com/opinion/articl…
Because it's a sign that we need to actively search for new fields to open up.
bloomberg.com/opinion/articl…
16/Science doesn't progress by simply doing more of the stuff that worked in the past.
It progresses by branching out in new directions.
AI. Neurotech and biomechanical engineering. Genetic engineering. etc. Fields that were science fiction 50 years ago.
It progresses by branching out in new directions.
AI. Neurotech and biomechanical engineering. Genetic engineering. etc. Fields that were science fiction 50 years ago.
17/And that means that our granting agencies, the NIH and the NSF, need:
1) More funding rather than less
2) More money allocated to lesser-known institutions, and more for up-and-coming researchers rather than established ones.
1) More funding rather than less
2) More money allocated to lesser-known institutions, and more for up-and-coming researchers rather than established ones.
18/Whether we're seeing "the end of science" is a question that will never be answered.
But each specific line of inquiry eventually sees diminishing returns, so we need to always be opening up new lines.
(end)
But each specific line of inquiry eventually sees diminishing returns, so we need to always be opening up new lines.
(end)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
