On another note, I have seen small no-name labs published impactful scientific results, and then went back to stealth mode again. It's super cool. Science labs, unlike a business entity, should not be measured by recurrent revenue and growth.
So, usually when people say this about sturgeon's law type stuff, it's because we genuinely don't know how to determine what's crap and what's not overall. For example, if we're talking about 90% of academic papers are crap, ok, it's true, but we pay the cost because we don't apriori which 10% will be the good ones.
The issue is that in academics, we actually have a pretty good idea which programs produce the best and most impactful work. Now it doesn't seem like a cost we have to pay, but rather a conscious decision not to cut out the crap (or to keep funding the crap).
I think the ideal outcome is that the culture and resources of all programs are approximately the same, and each produces 90% crap, and 10% good stuff.
Did the individual contributors stay at these labs?
Newton did first rate work, but he also spent a lot of time working on alchemy and theology which didn’t get anywhere.
Research grants are generally competitive, if you haven’t shown you’ve ever done something my significant or haven’t done something significant in a long time, it is difficult to survive. But indeed even the top scientists spend a lot of time on dead ends.
It is true that often only top 5 labs or so doing good work, but those top 5 labs are not constant over time, they wax and wane like fashion trends.
The lesson is just because over last few years the top work came from only a few labs, it’s premature to eliminate all other labs because those top labs could be lower in the next cycle.
Sometimes people do investigate the crap, and discover it's actually fertilizer. The trouble is, one cannot tell in advance.
This law seems to lack any kind of rigor or theoretical basis and seems like a purely rhetorical device.
There's all kinds of reasons to expect that too... Everything from funding, likely collaboration partners, talent, etc.
That would give it some theoretical basis.
In many fields the definition of "crap" depends, after all, on what the rest of the field's output looks like.
It’s not quite the Liar’s Paradox, I think Sturgeon’s fits into the other 10%.
That isn't by any means to suggest we shouldn't try to make science better by being more replicable and reducing fraud, but if it is an improvement, we're already doing something right and there is far less cause for despair and cynicism.
It also, of course, doesn't do any base rate comparison to non-scientific ways of generating facts. If 90% of published psychology research turns out to be wrong, but 99% of folk psychology turns out to be wrong, then science is still winning there.
It doesn't matter a whole lot with respect to 'how it compares to 50 years ago' if the assumption that science is supposed to be objective and credible.
It's supposed to be 'approaching 100%' i.e. they are publishing findings, not fake findings.
i.e. it's not a game of 'getting a higher percentage' it's a game of integrity.
The implications that 'most science is rubbish' is pretty scary actually and it has profound consequences for the future both in practical terms (i.e. we can't trust a lot of it), but especially in populist terms (i.e. people not trusting climate change, vaccine research etc.).
"If 90% of published psychology research turns out to be wrong, but 99% of folk psychology turns out to be wrong, then science is still winning there. "
Is it though? We spend bazillions on psychology and nothing on pop psychology, so we'd have to contemplate how valuable that '10%' is to society - especially in the context that we may not even know 'which 10%'.
This is kind of a big problem.
If, on the other hand, you just dumped all those psychology researchers out of the academic environment and sent them to marketing and advertising companies... are we better off?
Honestly? Very possibly.
The reason is that regardless of your opinion of marketing/advertising (mine is not overly positive while acknowledging that informing consumers of new products etc is a worthwhile endeavour broadly speaking) at least if the psychology researchers were trying to implement whatever their pet theory is in a company the market will quite quickly let them know whether or not it is horseshit. Markets have faults of course but at least if the psychological theories were implemented at companies you could reasonably easily tell if they were effective or not simply based on do they increase sales etc. If they turn out to be nonsense, this acts as a signal to the individual as well as others that this avenue of study is not worth further exploration.
Whereas Academia can often act as an artificial greenhouse environment for theories/hypotheses that won't survive contact with the real world but seem quite successful in the artificial environment.
(Fudging here around the 0.05 threshold threshold, since stuff on that boundary is the most likely to produce false positives, not sure what how to use a more realistic range... I think this would just be upper-bound on false positives? Please correct/clarify if your probability/statistics knowledge here is deeper than mine!)
Let's say twenty different researchers have a similar idea, they all pursue it, and one of them gets (un)lucky and gets apparently statistically significant results. They're the only ones that try to publish, and nobody ever has the knowledge that it only worked 1 time out of 20 (or more?) - even the people who's experiments failed who see it are only aware of their own attempt, and maybe they just messed something up.
Let's say 90% of all published results are non-reproducible. What sort of pipeline would be required to create that? For every reproducible, significant finding, there are 9 false-positives. For each of those false-positives there were 19 true-negatives. So about 1 out of 180 research attempts are reproducible truly-meaningful novel results. Does that sound reasonable? It's hard, right, especially to keep finding new things on top of everyone else's continued progress too? But it isn't necessarily malicious or fraudulent every step of the way - we just have our system structured in a way that biases in favor of creating false positives.
(Put in other numbers, say 10% of all research has significant results, then 90% of it doesn't, and at the 0.05 level, that's still (up to?) 4.5 false positives for every 10 truly statistically significant things, meaning close to 1 out of every 3 papers would still be expected to be a false positive.)
Publishing negative results would require a bigger publication industry, and more people spending more time filtering through it all, but you'd at least be able to compare to the literature to say "this looks like it worked, but ... it hasn't ever worked for anyone else."
People look to get 3 things in a job: Compensation, Interest and Reputation.
The professor is a researcher who teaches. However, they do not want to merely be a teacher. Especially when it comes with a substantial pay cut in comparison to the industry. So, you offer them repute and interest by giving tenure and something to research. The university gets a better teacher for cheap and the professor gets an inflated sense of ego and self-actualization.
From the student's point of view, it opens up alternate avenues for promotion. If only the top 5 labs get funded enough to do good work, then entering those top 5 labs early is absolutely essential. Anyone who doesn't get on the bandwagon early, will be left to the wolves like medical residencies. It will greatly exacerbate the competition in standardized exams and sideline students who don't abide by a rigid template early in life. (eg: a lot of friends went to tier 2 universities for their masters, did excellent work and got into the top 5 labs purely because the top 50 lab was funded enough for them to make up for lost time early in life)
I have also seen lab impact scores jump around substantially enough, that while the 90% hypothesis may be true, we have no way to picking out who will create the 10% that's gold. While half of the 10% might get reliably produced by the top 5 labs. The other half can be notoriously hard to pin down. You will see this in top industry groups (MSR, Brain) that hire exclusively out of the top 5 labs. Even there, the 90% number holds, despite there being a hiring system specifically built for capitalistic impact.
On the flip side, any academic field whose employability exists only within academia is a dictionary-definition pyramid scheme. It is a common accusation that's correctly levied on the liberal arts, but STEM should not be immune to it.
Of course - I think if anyone act on the belief that 90% of their bets are duds, they would not be in the venture business. It's really interesting that a process could lead to a distribution - but knowing the outcome distribution is actually...useless?
This seems like an non-thorough and fairly uninteresting answer to the question.
I would want to know how much useful-but-not-groundbreaking stuff still comes out of the "make work" crowd, how wide you need the funnel to be at the top to make sure you catch the people who can end up in the top labs and who make breakthroughs, how much of a "safety net" that you need to make sure people are motivated to jump into the funnel in the first place rather than going into something safer, etc.
A world where we have fewer "wasteful" make-work research jobs but also have fewer breakthroughs isn't great. Hell, you could call the majority of today's service- and information-oriented non-academic jobs "make-work" too (let's up those clickthrough rates! let's stream this video in 8k instead of 4k! let's move this money around slightly faster!). So is there necessarily much utility in moving "make-work" researchers into other fields?
You need to fund avenues that don't look promising now. Some new developments may make them good. E.g. deep learning before 2011.
That's also true for software engineers. A handful of 10x can carry a company a long way.
(But this should be obvious to a VC so I may be reading it wrong)
New business ventures are inherently risky.
Research is inherently risky, and we would expect 90% to 'fail' but for 90% of papers to be rubbish would be a matter of 'extreme corruption' along the lines of 90% of judges taking bribes etc..
"I don't see a better system."
Well, one thing we could do is punish people for knowingly publishing crap results, i.e. have a 'secondary review system' that investigates 'why' results were rubbish, and if there was malfeasance, well, put in place some kind of retribution.
We could also change our tune with respect to expectations, and come to 'expect' that the majority of research should publish then 'null' result and that should just be it.
It's possible for 90% of papers to be rubbish without any ill intent or dishonesty on the part of the experimenter. They might simply suffer from "works for me" syndrome, as it afflicts software developers. In science, this equates to failing to control for all the possible confounding factors that might affect the result of the experiment. They might have carefully controlled for everything they thought of and gotten methodologically valid results, but then when another experimenter tries to replicate it, they find their experimental environment is different in a way that neither considered important but alters the results.
Finding all these conflaters is the point of science. If you can't replicate the experiment, it means the results don't hold, but it doesn't mean the investigators were bad people - and assuming they are is unlikely to get better science.
Without being able to independently verify results, a lot of journals rely on signals like your research & results being noteworthy, which is anticorrelated with them being true. If our understanding of the world is at all correct, you'd expect most experiments to give boring, status-quo results, and your first impulse when you get something noteworthy is to assume the experimenter made a mistake. (Interestingly, this is how things work at the undergrad level - if you measure the speed of light at 1 million meters/sec, your prof will assume you screwed up.)
Probably we need a separate verification step after publication, with researchers evaluated on what fraction of their published results survive verification, not just how many papers they publish. Publish all the noteworthy results you get, but assume that published research doesn't mean much until it's been independently reproduced.
So let's be thankful of the crappy medium haha
I think there is a difference between "being shit" and "being fake" though. 'do you really think it’s true that 50-70% of biomedical research is fake?'
Fake seems more insidious and problematic.
Also this brings up a very troubling issue. If 90% of the "research is fake/shit" and lets say that means that 90% of the researchers are "fake/shit", then it means that 'scientific' consensus is also likely "fake/shit".
This is why I'm always skeptical of 'scientific' consensus. Science is about evidence, testing, etc. Not consensus - which is the realm of politics, law, etc.
And I'd suspect 99.99% of social 'sciences' is probably 'fake/shit' and that is used to push/change society/government/etc.
"Anyway, science right now is in an existential crisis. This is a real, real issue, and there’s now a generation of scientists who specialize in pointing this out and analyzing it. Andrew Gelman and others. I’ll give you an example: I had a conversation with the long-time head of one of the big federal funding agencies for healthcare research who is also a very accomplished entrepreneur, and I said, “do you really think it’s true that 50-70% of biomedical research is fake?” This is a guy who has spent his life in this world. And he said “oh no, that’s not true at all. It’s 90%.” [Richard laughs]. I was like “holy shit,” I was flabbergasted that it could be 90%.
He’s like “well look, 90% of everything is shit”, which is literally this thing called Sturgeon’s law which says that 90% of everything is bad. 90% of every novel written is bad, 90% of music, 90% of art… 90% of everything is bad. So his analysis was, anything you get in the field of medical experimentation, biomedical development, and this is going to be true of any field, there’s like five labs total in the world that are really good at what they’re doing and doing really cutting edge work. And this is true of quantum computing; pick any field for advanced technology you want.
So those five labs have a pretty good shot at doing interesting work, even some of that is going to reproduce and some of it isn’t. But once you get out of those top five labs, it’s pretty much make-work, incremental, marginal improvements at best, and a complete waste of time otherwise. And I said “good God, why does the other 90% continue to get funded if you know this?” And he said, “well, there are all these universities and professors who have tenure, there are all these journals, there are all these systems and people have been promised lifetime employment.” Anyway, a longwinded way of saying that we have pretty serious structural and incentive problems in the research complex. "