I had a similar conversation with a friend yesterday, about how infuriatingly boring and incremental much of science has become - people ask the safe questions, instead of the fun risky ones. I 100% agree that the system is very, very broken, and I like the analysis you present here.
but - I think I have some counter examples. there are countries in Europe where the funding system is different; people don't have to continuously compete for small grants like in the US. yet, it seems that at least in some cases, because of the "comfortable" situation and the lack of competition, people... stop trying, and still largely pursue the easy, comfortable, incremental questions. now, admittedly, this is mostly an outsider observation, the academic system I'm by far the most familiar with is the UK one, which is similar to the US.
if this observation is correct (which it might not be), then removing the tollbooth alone might not do much. but maybe there's some kind of middle ground?
That's a great point. Going too far in the other direction may have another set of unintended consequences. Maybe a working theory is: no competition leads to stasis, some competition is healthy, but hypercompetition, like we have in the US at least, really starts to distort things?
this got me wondering whether it isn't just about going too far in the other direction, but whether this is a case of macro environments being different, but people perceiving the same micro-constraints regardless.
probably a combination of both? either way, I love this way of thinking about the problem in micro and macro metascience terms.
The progress ‘feedback loop’ (more of a swirl) you describe sounds exactly like a description of modern social media, or regular news media for that matter. There is a ‘safe’ route of homogeny, which the AI algorithms then purify and exploit, leaving us with Walter White curated content, but little deviation or variety.
Funny how the capitalist conquest for ever more efficient and voluminous production does not pair particularly well with scientific advancement. 🤔
totally agree that the underlying dynamics are similar. the parallel to social media algorithms is on target in the sense that both systems have tendencies to optimize for what's measurable (engagement, citations) rather than the underlying quality of the content, and ultimately have a homogenization effect. I think this is relevant to capitalism but also even more general, given that the NIH grant system isn't really a market. Sure, market forces influence science in lots of ways, but in this particular case the perverse incentives emerged more from institutional inertia and historical quirks than from profit motive. My read is that this is really a story about incentives and selection pressures - capitalist or otherwise - which I think is a general theme in AI impact: it will amplify whatever dynamics already exist in a system, whether that's a weird government bureaucracy or a competitive market.
There are a lot of ideas out there (including some the NIH hase tried) but I think regardless of the entity, the key is to create an environment that isn't so focused on short-term outcomes. This is a hard sell these days, public or private.
The tollbooth analogy is perfect. Ive seen this play out in pracitce where AI tools help scientists write grants faster, but that just means more proposals competing for the same limited funding, which doesn't actually accelerate anything meaningful. The historical detail about the NIH borrowing a Depression-era risk-minimization model and never updating it shows how institutional inertia compounds over decades.
Yeah, i absolutely love that analogy (and to be clear, it is from Arvind and Sayash). I think about it constantly when people suggest throwing AI at a problem: Is this a number of lanes problem or a tollbooth problem?
Oh and yes, i found the history in that interview to be just so revealing
I had a similar conversation with a friend yesterday, about how infuriatingly boring and incremental much of science has become - people ask the safe questions, instead of the fun risky ones. I 100% agree that the system is very, very broken, and I like the analysis you present here.
but - I think I have some counter examples. there are countries in Europe where the funding system is different; people don't have to continuously compete for small grants like in the US. yet, it seems that at least in some cases, because of the "comfortable" situation and the lack of competition, people... stop trying, and still largely pursue the easy, comfortable, incremental questions. now, admittedly, this is mostly an outsider observation, the academic system I'm by far the most familiar with is the UK one, which is similar to the US.
if this observation is correct (which it might not be), then removing the tollbooth alone might not do much. but maybe there's some kind of middle ground?
That's a great point. Going too far in the other direction may have another set of unintended consequences. Maybe a working theory is: no competition leads to stasis, some competition is healthy, but hypercompetition, like we have in the US at least, really starts to distort things?
have you seen this post about micro and macro metascience? https://goodscience.substack.com/p/the-economy-of-knowing
this got me wondering whether it isn't just about going too far in the other direction, but whether this is a case of macro environments being different, but people perceiving the same micro-constraints regardless.
probably a combination of both? either way, I love this way of thinking about the problem in micro and macro metascience terms.
The progress ‘feedback loop’ (more of a swirl) you describe sounds exactly like a description of modern social media, or regular news media for that matter. There is a ‘safe’ route of homogeny, which the AI algorithms then purify and exploit, leaving us with Walter White curated content, but little deviation or variety.
Funny how the capitalist conquest for ever more efficient and voluminous production does not pair particularly well with scientific advancement. 🤔
Great article!
totally agree that the underlying dynamics are similar. the parallel to social media algorithms is on target in the sense that both systems have tendencies to optimize for what's measurable (engagement, citations) rather than the underlying quality of the content, and ultimately have a homogenization effect. I think this is relevant to capitalism but also even more general, given that the NIH grant system isn't really a market. Sure, market forces influence science in lots of ways, but in this particular case the perverse incentives emerged more from institutional inertia and historical quirks than from profit motive. My read is that this is really a story about incentives and selection pressures - capitalist or otherwise - which I think is a general theme in AI impact: it will amplify whatever dynamics already exist in a system, whether that's a weird government bureaucracy or a competitive market.
Insightful thanks. How best to fund risky research?
There are a lot of ideas out there (including some the NIH hase tried) but I think regardless of the entity, the key is to create an environment that isn't so focused on short-term outcomes. This is a hard sell these days, public or private.
Great piece.
The tollbooth analogy is perfect. Ive seen this play out in pracitce where AI tools help scientists write grants faster, but that just means more proposals competing for the same limited funding, which doesn't actually accelerate anything meaningful. The historical detail about the NIH borrowing a Depression-era risk-minimization model and never updating it shows how institutional inertia compounds over decades.
Yeah, i absolutely love that analogy (and to be clear, it is from Arvind and Sayash). I think about it constantly when people suggest throwing AI at a problem: Is this a number of lanes problem or a tollbooth problem?
Oh and yes, i found the history in that interview to be just so revealing
Do you have good recommendations for similar content to the Lauer interview, discussing potential new models?
Here's one overview on alternative models that considers both incremental and more radical reforms: https://www.pnas.org/doi/10.1073/pnas.2401237121
Also this substack by @stuartbuck if you don't already subscribe! https://goodscience.substack.com/