Hacker News Reader

Comment by drgo | Hacker News Reader

The entire research field needs major reform urgently. There are so many corrupting incentives for its practitioners that nearly most research findings (at least in my field: medicine) are suspect. Most researchers have neither the time nor energy to work on the important problems, to perfect their craft and to understand the limitations of the scientific methods. The metrics used in evaluating grant applications and academic promotion push researchers to work on short-term sensationalist and easy-to-publish ideas. This is sad because if there was a time we need solid unimpugnable science is now.

View on HN

In clinical medicine, the system needs no reform, because almost nobody's goal is to do research. Clinical research is pretty unique in that it serves almost entirely hierarchical advancement rather than science. Be honest, when was last time you saw a clinical researcher formulate a consistent research hypothesis and constitute a solid project around it? Almost all projects are basically: 'I'll test that with homemade shitty stats because it's trendy!'

So IMO, the first reform to conduct is to turn clinical researchers into real scientists. That won't be easy.

> So IMO, the first reform to conduct is to turn clinical researchers into real scientists. That won't be easy.

Mostly, there are not enough hours in a life to be both a good doctor and a good scientist. The few MD-PhDs who give equal weight to both areas are both brilliant and extremely driven. Multidisciplinary teams seem to have a better chance at success.

I guess if the scientific education was done early (undergraduate at the latest) dual training could work. Once medical training and practice starts there isn't a lot of time left over. And early / non-practicing science education, for whatever reason, doesn't seem to be very effective.

I was told long ago that the first rule of evaluating a medical treatment is to ignore all the clinical studies because nothing useful comes out of them.

I don’t know if that’s actually the case but that seems to be the prevailing thought. Which is awful, because it should be the opposite. Clinical studies should be the gold standard. They’re often conducted by people who are better doctors than scientists though. A zero-blind trial of n=4 might perhaps not be the definitive study the doctor had hoped.

(Not to pick on doctors. My wife’s one. Software developers often fall into the same common “I’m good at this one hard in-demand skill so I must be good at all of them” trap.)

To ignore _all_ clinical studies because “nothing useful comes out of them” seems extreme. Yes, there are poorly run studies, and conclusions can be inappropriately drawn from studies (ex: when a study wasn’t meant to answer certain questions).

However, well-run randomized control trials (RCTs) in medicine absolutely exist and can be a very strong form of evidence when interpreted correctly. This is especially true in comparison to other disciplines (economics, etc.) where such RCT trials are impossible and researchers wish that they a research tool as valuable as an RCT.

Fair. It was mainly in the context of “clinical studies show that unlikely thing X cures Y”. No randomized double-blind trials? Sure.

Interested to hear what you mean when you say "turn clinical researchers into real scientists." Are you thinking of expanding the MSTP? Or just better statistical literacy for physicians? Or something else entirely?

I don't know of any magic formula. We'd have to force the change somehow, because it won't correct itself. But the path is unclear, if there is a path at all. The current situation is that most clinical research is performed by MDs/MD-PhDs with no scientific background else than what is taught in medical/doctoral school, i.e. they lack the basics, and are usually more interested about advancing their careers rather than science. That said, systems shape people, and not the other way around.

I am an MD/PhD, and I agree that a lot of the research done by clinicians is self-serving and useless. But most medical/health research is not done by physicians who are typically quite busy and generally not motivated to do anything that involves statistics and takes time and effort. Also, most clinician-led research is simply ignored because it tends to be overridden by the results of large clinical trials usually conducted by professional scientists. Where I see major problems are in two fields. The first is epidemiology (which receives about 30% of Research funding) where most of the research is done by non-clinicians, often involves data dredging and posthoc theorizing motivated by easy access to larger amount of pre-collected data and the availability of statistical software that permits fitting black box models that typically show whatever results are good for publication. This field is the source of contradictory results that we read every day in the news where something and its opposite are both good and bad for every kind of imaginable illness or symptom -. The other is a lot of basic sciences research which typically represent about 60% of all funded research. The great majority of that research is baseless as was shown by effort to replicate many of the findings. Billions of dollars are spent searching for genetic and other causes of illness using small poorly designed studies that can neither prove or disprove anything. So the problem is much bigger than just poorly trained and busy clinicians trying to prove a pet hypothesis. The problem is everywhere. Edited to fix typos

I also agree with what you wrote here. But I have much more experience with clinical research, and know less about other fields.

It has become a game of reputation, citations and the number of papers. It works the same way that a company does, with the actual work being done by students and postdocs and citations accumulated by management (a lot of them don’t do research, rather spend time putting their names in different papers, collecting more and more students, and playing academic politics). After a while, they become a luminary and collect awards.

It looks eerily similar to the problems of businesses that happen when all what the executives care about is the stock price growth in the current quarter.

I think the scale of the problem varies wildly between different fields of science. For instance, in computer science, double blind reviewing is the norm, and the biggest issue is that there is an incentive to produce more papers with smaller contributions, instead of a mixture of larger, better, papers, and 1-2 column notes that describe important, but simple results (such notes used to be published routinely).

In other fields, not only are submissions not anonymous, but editors sometimes override the reviewers (in both directions), for what appear to be political purposes.

Double blind reviewing is not the norm in computer science. You will sometimes see it, but it depends on the subfield, the specific conference/journal, and even the year. In my experience as an author/reviewer, the submissions have been anonymous maybe 10% of the time.

And the biggest issue I see in CS is the focus on gatekeeping. When many (most?) papers are published in conferences with a single round of reviews, peer reviewing becomes more about accepting/rejecting the paper than improving it. The average review I receive in bioinformatics is more useful to me than the average review in CS.

Uhh, the elephant in the room for CS/ML publications is that a lot of people hardcore lie on their results page since almost no one tries to reproduce anything.

I've witnessed folks straight out of their undergrad with literally 1 workshop paper at an ACL track conference get thrown 150K starting offers by no-name startups. Imagine how important it must be to get your first NeurIPS main conference publication. The ROI likely amounts to 300K+ per year, so if anything, I'm surprised that cheating/lies in the academy aren't even more brazen. Dynamics like this are why both Stanford and Harvard's presidents had to step-down for gross academic plagerism/misconduct.

1. https://www.science.org/doi/10.1126/science.359.6377.725 (2018)

2. https://www.technologyreview.com/2020/11/12/1011944/artifici... (2020)

3. https://www.nature.com/articles/d41586-023-03817-6 (2023, Nature)

I've seen it happen even when source is available. I've also seen a lot of really terrible math mistakes in ML because people don't understand the underlying assumptions of the metrics they use.

The rot is everywhere because people forgot the goal. We all get caught up in the metrics that are used to judge us that we forgot what we were proxying in the first place.

What are the current corrupting incentives? And what would an ideal research world look like?

What kind of reform?

I don't personally see a solution other than student loan money being much harder to get.

There is so much bad research because there are too many researchers.

For myself, I think I am smart enough to have worked on a PhD in economics but not smart enough to have done anything worthwhile or to have advanced the field. I am certain I would not have. Then in the face of this massive student loan debt I would have produced junk research and try to game the system to make the investment worthwhile. The incentives are all ridiculous in the current system for most.

The idea politicians will cut student loan money in order to produce better scientific research would be a good way to optimize for losing an election.

Personally, I don't think there is a way out. If anything it probably just gets worse.

There are countries where student loans are not a thing and the science there arguably faces the same problem.

PhD is not paid by a student even in US, it's all scholarships.

Get paid to answer important questions instead of getting paid to provide sought-after results. I think that's the end-point, the "how we get there" is now the question!

Who decides what's important?

Great question. Who decides currently?

Nobody? The whole premise is that we should somehow start awarding people for working on important questions which we do not do now.

The usual critique of academia is that people are expected to publish a lot (which creates wrong incentives) and that nobody really accesses quality of the publications (which gives opportunities to cheat). At the same time there is huge PhD overproduction, which creates competition.

I don't think the problem is that people are working on wrong problems, the problem is how they work on problem of their choice and why.

I also don't think there is a simple fix to this problem.

Yes, you are absolutely right, and this in turn has led to people disbelieving science and things which were once a bipartisan issue, such as climate change, has now been termed a conspiracy by people on the right. This has only happened because people do not trust scientists and science anymore.

I'm not sure about the direction of causality here. Anti-vax people or climate change deniers also object to well-replicated results and would believe what they want to believe anyways.

It does give them some ammo for low-brow dismissal of the entire field of science. Which the more extreme may use to convince the more reasonable people. Even though the argument is rarely done in good faith, we do need some strategy for dealing with it...

Being perfect is not really a realistic option though - mistakes will be made in such a complex endeavour as science. But if we can no longer legitimately say "it only _a few_ bad apples", then we must improve things drastically - or risk loosing all the credibility.

There's an interesting documentary about flatearthers called "Behind the Curve" that gets into individual flatearther origin stories. The interesting thing is they are not stupid it's more than they have learned to not trust people in positions of authority on topics of science (often due to an experience in grade school).

Towards the end there are a few funded research projects that are erily more similar to the social dynamic of actual "legit" academic science than the film makers probably realize.

Can you please define terms so that I can understand exactly what you mean? Meanwhile I decided to check Wikipedia. https://en.wikipedia.org/wiki/Climate_change_denial says

Climate change denial (also global warming denial) is a form of science denial characterized by rejecting, refusing to acknowledge, disputing, or fighting the scientific consensus on climate change.

I cannot assume this corresponds with your own use of the term which is why I suggest defining terms but I'm wondering if anyone reading this sees some slight problem in the Wiki wording? Let's assume any disputation will have to include peer-reviewed evidence if it's to be seriously considered though according to Wiki, it won't be, whatever the evidence. This is science - according to an anonymous editor of Wikipedia? As described by Wiki, this is a religion.

The fact that there is even a term like "Climate change denial", means that it feasible to deny "the science" (science is about doubt), because the climate scientists actually can't prove anything, fundamentally. They only have stats, and stat models, and there is no way, at our technological level to even model a single mole of carbon 12 (12 grams of atoms), so you think they can model an entire planetary system? Accurately? No, it's not possible.

They also like to muddy the waters, climates change, because climates aren't even a thing, but an aggregate, obviously. Anthropogenic climate change, is what's being questioned, and that there is no certainty that it will cause disasters any more so than if there was still 300ppm CO2 in the atmosphere. Maybe 400ppm is bad, doesn't seem so tho, we are living the experiment.

BUT also, I would bet Air Conditioners are far worse for the planet than 400ppm CO2, even if they act together. ACs, are, absolutely abysmal thermodynamically. Heating up one space to make another space colder temporarily, with high likelihood of CO2 release.

Slight tangent. Have you heard the sixth Dimmu Borgir album? Good stuff. https://www.youtube.com/watch?v=TFRwNYPsHwY

I wonder will it generate any discussion, or just be castigated via downvotes, your choice.

We also can't model every single one of the atoms splitting in a nuclear reactor core, but thanks to stats and models, we can predict how they behave to very high precision.

Knowing whether or not anthropogenic greenhouse gas emissions impact global climate can be reasonably achieved with some modeling assumptions.

We may not be able to predict the weather in each location for the indefinite future, but we certainly can and do understand the general impact of bringing billions of tonnes of new greenhouse gas into a system.

Similarly, “We can’t even model the inside of a proton. How can you possibly know when a pot will boil?”

Turns out you can get pretty far modeling the macro behavior without perfect knowledge of the micro details.

> The fact that there is even a term like "Climate change denial", means that it feasible to deny "the science" (science is about doubt), because the climate scientists actually can't prove anything, fundamentally.

There are people out there denying the shape of the planet, despite that it has been observed directly.

Am I talking about those people?

You're conflating things that are empirically verifiable with things that are not.

I was responding to what you wrote:

> The fact that there is even a term like "Climate change denial", means that it feasible to deny "the science"

This is clearly untrue. Some people, such as the 'flat Earthers', deny plain facts.

Yeah good point, I'll try better

"Psychology can't produce real results because we can't even model a single brain cell perfectly. They only have stats, and stat models."

You want people to think you're a specious disingenuous person? It's of your own doing.

> Anti-vax people or climate change deniers also object to well-replicated results and would believe what they want to believe anyways.

Accusing your enemy of being beyond argument is just an underhanded way to avoid making an argument.

There will always be bad beliefs, we will not always know at the time which beliefs are bad, and the response to what we think are bad beliefs should be good science, not massive fraud studies on hydroxychloroquine published in one the most prominent journals in the history of medicine.

Replicable is not a synonym for correct even though some discourse about the science crisis treats it that way. It's necessary but not sufficient. The academic literature is full of studies that will replicate every time but the methods are lacking rigor or are just outright invalid. The classic correlation/causation fallacy pops up everywhere, to pick one example. A paper that shows a correlation then declares a causation will replicate just fine but still yield wrong conclusions.

Funny how the unabashedly partisan are the ones who make partisan statements in the same breath as their denouncements of partisanship.

Declining trust in science is not because of lack of open-access publications.

The right is not ignoring research because it is behind a paywall or because it hasn’t been reproduced. In fact, the right has chosen to willingly ignore research that has been painstakingly reproduced, all in service of the god of greed and ideology.

In my opinion the declining trust is science is due to science being far less apolitical and far less objective and less validated than in the past.

Which era of science was apolitical, exactly? The one where Galileo was prosecuted by the church? Or one where the US military invested billions of dollars into early computer science? Or one where nearly every physicist was staunchly anti-proliferationist ?

Science has not, and will never be, apolitical. It is just that you are noticing that the politics of some scientists does not align with yours or with some politicians, and so you’re noticing the difference.

Do you have any evidence for this?

I suspect in the 60s and 70s science was (in the public imagination) wrapped up in a positive story of the future, full of space travel and prosperity for all. And now science is wrapped in a story (at least for some) of doom and gloom. Climate wars, or creeping authoritarian governments telling people what to think.

I suspect the experience of actual scientists, in labs trying things out, hasn’t changed anywhere near as much as the public’s stories of science have changed.

How is science political? I am happy to work from that premise, but it needs more than an assertion to be taken seriously.

My experience is that people complaining about science being partisan (which they tend to mean by “political”) do not understand much about science or how it works. I am willing to believe that you are not one of those people, but what you write is a bit close. So, could you give some particular examples or facts rather than empty phrases?

Yeah definitely dude. Climate disbelievers are actually doing good faith deep dives into the studies and identifying specific, real problems, and pushing for better science to be done.

This is evidenced by… wait… what exactly?

What if science and politics really are corrupt enough to corroborate the conspiracy theories? We need an International Journal of Conspiracy Studies right now!

Reply on news.ycombinator.com