Does Science Advance One Funeral at a Time?
nber.org
nber.org
I don't think this is how it works at all. I think people are unable to challenge stars in their field.
When everyone is celebrating some star, good luck getting heard if you disagree with them. It is worse if, like so many people, this star will defend their territory by following the maxim "A good offense is the best defense."
My observation of behavior in online forums is that a typical pattern of behavior is that everyone seeks to either align themselves with one of the "stars" of the forum or position themselves as being "against" anything that person says. It is very much about pecking order, not truth, and if you have two or three really popular people, then you get camps that revolve around each person. All conversation tends to default into a polarizing back and forth of "I am for STAR!" and "I am against STAR!"
Since all conversation is framed as either for or against STAR, no conversation can occur that genuinely diverges from the framing given. Even if you genuinely try to diverge from this framing of for or against the idea set that this star individual represents, people will actively paint you into a corner as being in either the for or against camp. Good luck with saying "Yeah, no. That isn't what I am saying At All."
This only stops when that person exits the picture. Dying is the most final and absolute means to exit the picture.
As a specific example look at evolutionary optimization. It's existed for decades, has a bit of a non mainstream cult following, but now the leaders of the field are finding they can beat the fanciest deep reinforcement learning approaches with it. https://blog.openai.com/evolution-strategies/
The fact that an alternative perspective that proves simple, effective, and popular in alternative communities is now popularized from stars in the field of deep learning rather than the alternative communities is exactly this.
In fact, the resources available to the stars is much different than what these other communities may have access to such that they may never be able to demonstrate value themselves. It's not very often you have a 720 core machine to run these tests. Resource constrained science exists all over from chemistry, physics, and biotech to computer science. The rich get richer as they say.
However it's also a great instance of solid science to test alternative approaches and I don't mean to take anything away from anyone, but simply to point to a case where that it is extremely difficult for people to challenge stars in the field.
Evolution Strategies as a Scalable Alternative to Reinforcement Learning
by Tim Salimans, Jonathan Ho, Xi Chen, Ilya Sutskever
This seems like counter-evidence to the idea that people are unable to challenge the stars of the dominant paradigm.
It was great to hear from her how long it took for the right theories to overcome the barriers of adoption, how fashion works in science, and how many things we think as new are actually really old stuff in new clothes.
"Kuhn challenged the then prevailing view of progress in "normal science". Normal scientific progress was viewed as "development-by-accumulation" of accepted facts and theories. Kuhn argued for an episodic model in which periods of such conceptual continuity in normal science were interrupted by periods of revolutionary science. "
https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Re...
Philosophers and historians have long debated the extent to which the pragmatic success of a scientific theory determines how quickly it gains adherents, or its longevity (e.g., Kuhn [1970], Laudan [1977], and their many detractors).
A former fellow grad student was expressing their dis-belief that though there was a lot of work done in the brain with electrical signals, there is effectively no work done on measuring the magnetic properties of the brain or neurons. The much older and much more esteemed PI that was also present simply shut the student down, to the point of telling them to 'shut up'. "There are no magnetic fields in the brain", I think was the quote. The student, misunderstanding the situation, pressed onward and challenged the PI on fMRI and the like. The discussion turned into a 1-sided argument where the PI basically told the student that they were an idiot and that things like fMRI were useless (actually a debatable point at the time, re: the dead salmon experiments). I'll say I never quite trusted PIs after that dressing-down, they seem to be more concerned about their mortgages than their legacies.
When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong.
[1] https://en.wikiquote.org/wiki/Arthur_C._Clarke#On_Clarke.27s...
I've met countless young researchers who embraced Bayesian practices. And all the opposition seems to come from older hands. It's finally turning a corner... Because of retirement!
The part that has opposition is using a Bayesian justification for the experimental results in a paper -- presenting a conclusion about how we should update our beliefs because of the experiment, instead of a p-value.
Right now it seems pretty common to translate Bayesian experimental results sloppily into frequentist ones to get them past reviewers.
This is interesting. I'm glad for a different perspective. I chalk this up to differences across fields, because my experience is so different. My advisor learned Bayesian statistics from Arnold Zellner a few decades ago, and his synopsis of the class was: "It was great in theory, but we could't implement it." Once computers made if more possible, it seems to have spread rather quickly, possibly because computing power was the last piece of a mostly complete puzzle.
It's almost to the point that reviewers prefer the Bayesian approach, in my neck of the academic woods.
Consistent with previous research, the flow of articles by collaborators into affected fields decreases precipitously after the death of a star scientist (relative to control fields). In contrast, we find that the flow of articles by non-collaborators increases by 8% on average.
I haven't read the paper, and maybe this is just completely wrong, but wouldn't we expect the number of published papers by non-collaborators to rise no matter what?
I mean, a star scientist is dead now. There are no more papers coming from that person. Of course other people's papers fill the gap. The number of journals didn't change. The number of articles in an issue of each journal didn't change. These journals need to get articles from someone.
"COBOL with separate compilation and parameters" is what we seem to be stuck with, even though the advances in hardware and garbage collection software have made FP type stuff practical now, rather than something that was scoffed at back in the 80s with the advent of microprocessors posing as general purpose computers :-)
I imagine the same principle applies to networks of people as does to networks of neurons.
Certainly amonst the unintended consequences one might care to consider.
Just putting it out there. Killing scientists to stimulate progress is out for ethical reasons, obviously.
Unfortunately, I feel this is similar to that one dessert on the front page of the menu at an expensive restaurant. It looks delicious and tastes even better, but as soon as the bill comes you wondered what in the world you were thinking.
Could be a good system, but I don't see the professors endorsing it.
The established institution of a sabbatical year (every 7 or so years) fits into this. China also tried a ... not entirely unrelated approach in the 1960s. With mixed results.
I will say this though. Young scientists are hungry and there not enough positions for them. Every department has that old rockstar or two who doesn't really push the boundaries of science anymore, but takes up available positions. When they retire or die, new people come in who are still young and full of energy.
Sounds to me like you believe that science proceeds without politics.
> Young scientists are hungry and there not enough positions for them. Every department has that old rockstar or two who doesn't really push the boundaries of science anymore, but takes up available positions. When they retire or die, new people come in who are still young and full of energy.
This process you've described here resembles the political process we observe in government (or any human social system that involves a competitive hierarchy), with the exception that in government the problem of 'taking up available positions' is attenuated by the existence of term limits for most leadership positions.
The question is whether politics and science
are equivalently grounded...
No one seriously debates whether politics is a factor in science. AFAICT, the the question is entirely to what degree politics influences the progress of science. In my experience, even people with relatively radical views understand this.
My first claim in great grandparent was that the argument that science on progresses when scientists die is used to argue that politics and science are equivalent, that science is reducible to politics. Since we agree that is not so reducible to politics, let us agree that we agree and move on.
None of the major exponents or scholars of this concept of whom I'm aware -- Bohr, Kuhn, Oreskes, and others -- have as their primary motivation any desire to demote or debunk science. If you're going to make that claim, you'll have to substantiate it. You've failed to do so. Twice.
So my argument was underspecified, which led you to believe I was referring to the like of Bohr, Kuhn, Oreskes and other serious thinkers, which I was not.
I'll offer as an excellent case study in development and establishment of new scientific theory, the emergence of plate tectonics (or continental drift theory). Naomi Oreskes, a tireless advocate of real science, and debunker of pseudoscience and disinformation (see Merchants of Doubt) documents the four-century-long battle to establish the concept in her books PLATE TECTONICS: The Insiders’ History of the Modern Theory of the Earth and The Rejection of Continental Drift, along with several further articles on the topic.
https://www.kirkusreviews.com/book-reviews/naomi-oreskes/pla...
http://www.worldcat.org/title/rejection-of-continental-drift...
Which is basically the norm for a lot of industries. I'm not even sure I mind this sort of thing so long as they are still contributing and producing. I just wish there were enough space for the young folks to do their stuff too.
This kind of ingrained conservatism can delay progress both of a field and those examining it. It is a catch-22 as you want qualified reviewers to examine a paper so it isn't bunkum but equally those reviewers may slow it down because it is a valid advance in contrary to their position.
Perhaps a move to post-publication peer review and greater pre-print deposition would help, but that would take a deep cultural shift.
But as always it doesn't have to be that way, it has arrived at this point due to a combination of sometimes perverse incentives and natural human tendencies.
I was thinking for a while that having a mandatory retirement age could be a good idea, but retirement is becoming obsolete for economic reasons, and scientists are now employed in so many different ways by so many different institutions that I don't think it is possible.
Next up: a curve showing that friends of WWI vets decrease their tuna consumption when the WWI vet dies, relative to their non-friends.
It's a really tricky topic...
It sounds like they want CC-BY, but as it stands they haven't really licensed the digest in an understandable way.
I've seen worse. Went a few rounds back in the 2000s with a developer who'd claimed simply "this has no copyright". I pointed out the several problems with that claim, and the (almost wholly) trivial approaches to fixing them.
He ... wasn't interested.
It is, legally, difficult to divest yourself of copyright. Some people say it's possible, some people say it isn't.
It's not difficult to divest yourself of your exclusive rights under copyright. Which is what the WTFPL accomplishes. It's a subtle but significant distinction.
Similarly, you could:
1. Simply disclaim any copyright interest. "I disclaim all copyright interest in this work."
2. Disclaim any exclusive rights: "I disclaim all rights under copyright in this work."
3. Allow others any exclusive rights: "I grant to all parties any exclusive right under copyright."
Etc.
The fundamental problem is that under Berne Convention copyright, which dates to the early 20th / late 19th century (I'd have to check) in at least some jurisdictions, copyright is manifest at creation in any work. Which means these very words as I'm writing them. And there's no procedure for disclaiming those rights.
The Berne Convention rules have been adopted widely particularly since the 1970s.
Again: there are those who disagree that you can't claim a work is "uncopyrighted", but the ambiguity of such a comment (and the trivial means by which that ambiguity is removed) makes certain uses highly imprudent.
Sadly most people think waiting till the nonbelievers die is the way to go and not much money is spent in immortality research.
1. Future-value discounting. Even an immortal may well value present experiences over future ones.
2. Consolidation of power. There's little evidence that any system based on perpetuities won't result in the accumulation of wealth and/or power in a limited set of hands.
3. Strategic intentional long-term power consolidation. If you live forever, you'll be concerned forever about upstarts seeking to claim / disrupt your space / power base.
The backstory here is that SSRN, a site for open dissemination of largely pre-publication papers, was bought by Elsevier in 2013, to the loud dismay of the open-access community.
It seems that the predictions of what would transpire are being fully born out.
https://svpow.com/2016/07/18/elsevier-has-started-destroying...
The Sci-Hub workaround is to chase the NBER PDF download link:
The whole "Minsky and Papert killed the perceptron" lesson-in-a-nutshell was a historical anecdote at the time the connectionist resurgence started in the 1980s with backprop. I.e., Minsky was already in the distant past by the 1990s, having no influence on connectionism or the NIPS crowd, from which large-scale machine learning eventually sprung.