6 ms·

I love these people's work. At the same time I don't think any of this really addresses the root causes of the problems, which are career, university administr

I love these people's work. At the same time I don't think any of this really addresses the root causes of the problems, which are career, university administrative, and field (i.e., recognition and attention) incentives. The whole paradigm we use, and the way we treat scientists and fund research, is broken. It feels a little bit like bailing water out of your boat with teacup on one side while ignoring the gaping hole letting rushing water in on the other side.

anonymouskimmer 3y ago
The root cause it too many people for the funding available. If a single scientist can spend $250k per year it's obvious that even rich economies can only support so many of them. Whenever you have this situation you inevitably lead to selection criteria which can, and will, be gamed.
- buitreVirtual 3y ago
  It's not just the funding but the jobs themselves. Universities and even the funding bodies want to show shiny easy-to-digest numbers to their stakeholders to justify tuition and government funds. Which means these institutions in turn expect those numbers from their researchers, who then are happy to comply and build their paper mills and citation cartels, solving no real problems at best and producing fake data at worst.
  - agumonkey 3y ago
    I think this issue is universal to any human work group. It would be worth studying it to better understand these patterns.
armchairhacker 3y ago
But how do we “solve” those incentives? There will always be winners and losers, and since successful fraud is by definition indistinguishable from genuine success, there will always be people trying to game the system. The only effective solution I see is making fraud harder to hide. (There’s also making the penalty for fraud higher and trying to instill integrity, but the problem is some people will still take the risk.)
- User23 3y ago
  > since successful fraud is by definition indistinguishable from genuine success This is absolutely false in real science and that you say it seriously is proof of just how far gone the academy is. Needless to say no matter how much you fake your data or results, nature is not so easily fooled.
  - nitwit005 3y ago
    You're misinterpreting the quote You're responding to.
- scherlock 3y ago
  The major issue with research is how only successful results are viewed as worth while, and study results without a "positive" out come is viewed as a failure. So this means researchers are incentized to p-hack or falsifying results since they need more funding. The reality is these negative results are just as useful and can either help other researchers avoid dead ends or they might have a not great result but with some refinement to methodology might be worthwhile for further study. Researchers should be judged on the quality of the research they perform, not their outcomes. A well crafted study that shows that further research shouldn't be done should be just a well received as a well crafted study showing more research should be done. The focus should be on the design of the study first, since no matter what the results it adds to the corpus of knowledge in a meaningful way
  - armchairhacker 3y ago
    You are 100% correct. The problem is, even without the incentive of money and employment there's still the incentive of fame, and negative results simply won't get the same attention and impact as positive results. We can give people with negative results PhDs and publications, we can continue funding them. But at the end of the day, people remember and researchers cite the team who invented the LLM, not the many teams who tried and invented the less-successful models.
- nwhitehead 3y ago
tkgally 3y ago
One positive move would be for more institutions to publicly endorse the recommendations of the Declaration on Research Assessment. It won’t solve all of the incentive problems, but it should be a step in the right direction. I wrote about it briefly here a couple of weeks ago [1]. [1] https://news.ycombinator.com/item?id=39064322 https://news.ycombinator.com/item?id=39064322
gustavus 3y ago
We've had these incentives since we started science. Remember Albert Einstein wasn't able to get a teaching position until after he was winning the Nobel. It turns out the people that really love science will do it even if it doesn't pay anything and brings them nothing. We just need to make sure the salesman scientists and parasitic beuracrats don't squash the efforts of those doing real work.
gmd63 3y ago
The solution is unfortunately for everyone, more transparency in research. When you conduct an experiment, why on earth is the report composed months, sometimes years later, instead of being streamed live and recorded? Just the same as in business. We need a new more prestigious stock exchange for companies that are willing to open their books up to continual audit. The government can step in to aid businesses and researchers that act openly, in the form of grants to researchers or tax breaks for members of the more restrictive stock exchange, and shift the cost of policing destructive fraud to rewards for constructive conduct.
- analog31 3y ago
  We're not all that different than tech workers. Who wants to be on stage 100% of the time?
  - gmd63 3y ago
    That's why the incentives exist. Folks who can stomach it get tax breaks or bigger grants. Evidently, we're not able to stop fraud, so it makes sense to analyze how much we're spending fighting fraud, and spend that paying people to be honest.
- jltsiren 3y ago
  Today I "conducted an experiment". Which means that I went to a grocery store, did laundry, and took a walk in a park. Between all that, I checked once in a while that things were proceeding as expected. The outcome of that experiment is probably never going to be reported, except maybe if something unexpected happens. Real science is full of messy exploratory work. It's sometimes done remotely, and often in short stretches regardless of what time and day it is. And you usually don't know in advance if it's going to be relevant. At some point, you may start believing that you have discovered something other people may care about. Then you start thinking about the normative experiments that are supposed to demonstrate the result and perhaps even validate it. But at that point, most of the work is often already done. The discovery has already been made (or at least you believe so) and you are just trying to replicate it. Maybe you could record those experiments if you are really into it. But the recordings would likely be meaningless without the actual written report.
verisimi 3y ago
Even if you resolve fraud, you still fail to resolve the driver of the fraud, which is money. There are basically 3 sources of funding for universities - government, military, corporations. These 3 have aligned interests. Where their interests align (or are not in conflict) you will have funding of one side only. If you fund one side only, you will generate lots of information to support whatever-it-is, but this will fail to be a fair representation of the issue. I'll just run through a simple, silly example. If I wanted to prove cats cause stormy weather, I could fund say, 10 studies. Perhaps 2 are supportive (cats and storms correlate in some way). I can then fund even more studies in that direction. I would soon have a huge body of supportive data .. but I haven't said anything meaningful, I haven't got to the truth of the matter. I just have a huge body of evidence. To me, this is chicken feed. It's not that the information is a lie necessarily, it's just that it's isn't important. Nevertheless, the body of data can be referred to, in order to help steer public opinion. Everyone is getting paid, the thing is true .. what's to be upset about? Anyway, whoever has the money determines the outcome. Truth is a secondary consideration.

anonymouskimmer 3y ago

The root cause it too many people for the funding available. If a single scientist can spend $250k per year it's obvious that even rich economies can only support so many of them. Whenever you have this situation you inevitably lead to selection criteria which can, and will, be gamed.

buitreVirtual 3y ago
It's not just the funding but the jobs themselves. Universities and even the funding bodies want to show shiny easy-to-digest numbers to their stakeholders to justify tuition and government funds. Which means these institutions in turn expect those numbers from their researchers, who then are happy to comply and build their paper mills and citation cartels, solving no real problems at best and producing fake data at worst.
- agumonkey 3y ago
  I think this issue is universal to any human work group. It would be worth studying it to better understand these patterns.

armchairhacker 3y ago

But how do we “solve” those incentives? There will always be winners and losers, and since successful fraud is by definition indistinguishable from genuine success, there will always be people trying to game the system. The only effective solution I see is making fraud harder to hide. (There’s also making the penalty for fraud higher and trying to instill integrity, but the problem is some people will still take the risk.)

User23 3y ago
> since successful fraud is by definition indistinguishable from genuine success This is absolutely false in real science and that you say it seriously is proof of just how far gone the academy is. Needless to say no matter how much you fake your data or results, nature is not so easily fooled.
- nitwit005 3y ago
  You're misinterpreting the quote You're responding to.
scherlock 3y ago
The major issue with research is how only successful results are viewed as worth while, and study results without a "positive" out come is viewed as a failure. So this means researchers are incentized to p-hack or falsifying results since they need more funding. The reality is these negative results are just as useful and can either help other researchers avoid dead ends or they might have a not great result but with some refinement to methodology might be worthwhile for further study. Researchers should be judged on the quality of the research they perform, not their outcomes. A well crafted study that shows that further research shouldn't be done should be just a well received as a well crafted study showing more research should be done. The focus should be on the design of the study first, since no matter what the results it adds to the corpus of knowledge in a meaningful way
- armchairhacker 3y ago
  You are 100% correct. The problem is, even without the incentive of money and employment there's still the incentive of fame, and negative results simply won't get the same attention and impact as positive results. We can give people with negative results PhDs and publications, we can continue funding them. But at the end of the day, people remember and researchers cite the team who invented the LLM, not the many teams who tried and invented the less-successful models.
nwhitehead 3y ago

tkgally 3y ago

One positive move would be for more institutions to publicly endorse the recommendations of the Declaration on Research Assessment. It won’t solve all of the incentive problems, but it should be a step in the right direction. I wrote about it briefly here a couple of weeks ago [1]. [1] https://news.ycombinator.com/item?id=39064322 https://news.ycombinator.com/item?id=39064322

gustavus 3y ago

We've had these incentives since we started science. Remember Albert Einstein wasn't able to get a teaching position until after he was winning the Nobel. It turns out the people that really love science will do it even if it doesn't pay anything and brings them nothing. We just need to make sure the salesman scientists and parasitic beuracrats don't squash the efforts of those doing real work.

gmd63 3y ago

The solution is unfortunately for everyone, more transparency in research. When you conduct an experiment, why on earth is the report composed months, sometimes years later, instead of being streamed live and recorded? Just the same as in business. We need a new more prestigious stock exchange for companies that are willing to open their books up to continual audit. The government can step in to aid businesses and researchers that act openly, in the form of grants to researchers or tax breaks for members of the more restrictive stock exchange, and shift the cost of policing destructive fraud to rewards for constructive conduct.

analog31 3y ago
We're not all that different than tech workers. Who wants to be on stage 100% of the time?
- gmd63 3y ago
  That's why the incentives exist. Folks who can stomach it get tax breaks or bigger grants. Evidently, we're not able to stop fraud, so it makes sense to analyze how much we're spending fighting fraud, and spend that paying people to be honest.
jltsiren 3y ago
Today I "conducted an experiment". Which means that I went to a grocery store, did laundry, and took a walk in a park. Between all that, I checked once in a while that things were proceeding as expected. The outcome of that experiment is probably never going to be reported, except maybe if something unexpected happens. Real science is full of messy exploratory work. It's sometimes done remotely, and often in short stretches regardless of what time and day it is. And you usually don't know in advance if it's going to be relevant. At some point, you may start believing that you have discovered something other people may care about. Then you start thinking about the normative experiments that are supposed to demonstrate the result and perhaps even validate it. But at that point, most of the work is often already done. The discovery has already been made (or at least you believe so) and you are just trying to replicate it. Maybe you could record those experiments if you are really into it. But the recordings would likely be meaningless without the actual written report.

verisimi 3y ago

Even if you resolve fraud, you still fail to resolve the driver of the fraud, which is money. There are basically 3 sources of funding for universities - government, military, corporations. These 3 have aligned interests. Where their interests align (or are not in conflict) you will have funding of one side only. If you fund one side only, you will generate lots of information to support whatever-it-is, but this will fail to be a fair representation of the issue. I'll just run through a simple, silly example. If I wanted to prove cats cause stormy weather, I could fund say, 10 studies. Perhaps 2 are supportive (cats and storms correlate in some way). I can then fund even more studies in that direction. I would soon have a huge body of supportive data .. but I haven't said anything meaningful, I haven't got to the truth of the matter. I just have a huge body of evidence. To me, this is chicken feed. It's not that the information is a lie necessarily, it's just that it's isn't important. Nevertheless, the body of data can be referred to, in order to help steer public opinion. Everyone is getting paid, the thing is true .. what's to be upset about? Anyway, whoever has the money determines the outcome. Truth is a secondary consideration.