Papers and patents are becoming less disruptive over time
nature.com
nature.com
It's easy to make up a measure and then develop a story that explains why that measure is useful. Was Newton's theory of gravity disruptive, or consolidating. I'm thinking consolidating, since it allowed a lot of other things to make sense. Likewise, perhaps quantum mechanics was disruptive -- it sounds disruptive -- but it also consolidated a lot of confusing observations.
(And lets not even mention the problem that when there are 1,000-times as many papers between 1940 and 2010, we might not expect 1,000-times as many disruptions.)
This paper has a measure that sounds useful, and is certainly good for headlines, but it is very unclear that it provides any insight into the progress of science.
Pace of publishing and observations have dramatically as you rightly point out.
That recognition immediately led to a flurry of follow-on work confirming the mechanism and birthing molecular biology as we know it, which opened up vast new areas of science.
I think the more important question is, how do we validate the usefulness of the CD5 metric. And, why is it better to be "disruptive" than "consolidating"?
Based on the examples shown (and others that I can imagine), it is unclear that it measures anything related to scientific significance.
Probably best to just say some of the greatest discoveries are both disruptive, and consolidating.
The trend over time could equally be capturing changes in referencing practice (either that really important, influential papers/inventions in the past didn't reference other really important, influential papers/inventions nearly as often as equally novel papers/inventions today, or that modern referencing is much more exhaustive about referencing minor and tangentially relevant papers/inventions. There is, of course, also more significant prior art in journals and patent indices to actually cite in later years...). They carry out some robustness checks, but I'm not sure any of them really dent that hypothesis. It's also a conclusion based on proportions (ie. the effect mostly comes from by there being a lot more "non-disruptive" papers published rather than fewer "disruptive" papers) and there's actually a trend rise in the raw number of "most disruptive" academic papers from the 1950s through to the 1990s, at which point the numbers drop so sharply it looks like a discontinuity in the data (but would probably coincide with lots of journal content becoming easily accessible online...)
When I read a modern paper, I skim a lot more of the references. Getting to a cited paper is often just two clicks: there's a reference to the bibliography, and from there a link to arXiv (or some other public repository). As papers get older the bibliography might not link to arXiv, or might not link to a digitized paper at all. Those older papers are more time consuming to use as a reference. Scientists aren't lazy but they also don't have infinite time.
Put another way: papers form a graph, this result suggests that older nodes are traversed less frequently. I think the metric they propose is useful, but we could rephrase this all as measure of how useful papers are as a node (where CD = -1 is most useful, CD = 1 is least useful).
In short, this result might just mean paper discovery is getting easier. That's a good thing. We can wring our hands endlessly about how there's less low hanging fruit (there is) or about how academia is rotting away (as though it was ever different), but those conclusions are big jumps from what this paper says.
I didn't push back, because I had no power and I just wanted to get the fuck out of there and start making money after being scammed out of 5 years of my life where I poured a lot of effort and got shit pay. I lost respect for my advisor and I was stunned by how un-scientific his response was when it came to correcting his 1980s results. My measurements were more accurate and were strongly supported by structural measurements he never did and couldn't do back then. But he was more interested in protecting his "legacy" of his paper with 13 citations.
So petty and small. I dropped those efforts and was able to get out completely obviating those results, never mentioned them in my thesis even.
Anyway, my point is that I'm not surprised that scientific results are lame, vanilla, uninspired. When the committees that decide on what gets funded are all 65+ dinosaurs stuck in their ways, very few disruptive ideas will be founded because you're disrupting the very research that got the committee members clout and recognition.
As they say, change comes a funeral at a time.
Kind of like a Judge ruling one way or another for personal reasons, or ruling with a heavy conflict of interest.
There needs to be much stronger ethical principles around this and an advisor hinting at obfuscating work for such reasons should be up for review.
It can't all be blamed on corporatization, however. The control of funding agencies at the federal level by vested interests seems to be a major problem, i.e. innovative lines of research that threaten the status quo could threaten those cabals that control fund disbursement, so such efforts don't get funded. The stagnation of Alzheimer's research and the focus on the amyloid hypothesis seems to fall into this category. This kind of bureaucratic ossification is not new; the classic example is how Trofim Lysenko in the Soviet Union controlled the direction of agricultural research for about three decades, much to the detriment of the understanding of plant genetics and crop development research in the USSR over that period.
So it can be attributed to the rise of monopolistic corporatization on university campuses on one hand, and the growth of entrenched bureaucracies at the federal funding level on the other. Neither group wants the rug pulled out from under them by 'disruptive innovation'. There are many historical antecedents for this kind of situation, incidentally.
This is a fundamental problem with the nature of science and the nation state in that modern science is constituted by the nation state, not merely imposing into or upon it.
Take physics. As I read it, modern fundamental physics is mostly some combination of general relativity and quantum mechanics, which are both early 20th century inventions.
They weren't necessarily genius in particular in-depth analysis, rather they were genius new concepts that created a whole new blank canvas to fill in.
What did the field look like just before them? I always forget the details, but was it lord Kelvin who said, physics just needs to figure out the ultraviolet catastrophe and it's done? I wonder if it looked similarly uninnovative. The canvas they had was full, and they needed a new canvas for true innovation.
Now the thing is, I think true disruption, like GR and quantum physics, is not limited by the number of postdocs, but by truly genius insights. Einstein wasn't even working as an academic when he came up with Special Relativity. So the number of papers written isn't a proxy for pace of revolution.
Not to mention, since number of published papers has become a target, it has become useless as a measure of progress.
I think perhaps we've also partially lost the meaning of ground breaking research given the constant news articles on "ground breaking science" and the constant overinflation and exaggeration of one's papers and achievements.
Besides the current structure of research and the known weaknesses (misaligned incentives, funding issues, etc.) science in general today ran out of low hanging fruit. The last unexplored or unanswered riddles may need prerequisites that are outside of the current boundaries so someone needs to open the door do a vast new bubble that has more relatively low hanging fruit.
Also humans probably (don't quote me on that :) ) can't come up with any new questions or answers unless they are already at the edge of that bubble. And they each see and understand a smaller patch of that surface as it expanded. Stumbling onto new thinks is exponentially harder.
This is what those revolutionary discoveries did for us in the past: Opened the door to a new world where the hard answers were suddenly a bit easier, they made sense.
Yes, and we can’t expect it to occur in a socially confined over-structured environment either. Academia has changed a lot since then, with peer review hell, publishing pressure, etc. I mean, are we really pretending to be surprised that a system of incrementalist incentives produces incrementalist results?
It’s amazing upright apes can figure out the speed of light or postulate and find the Higgs Boson in the first place.
How much physics do we expect the smartest dog in the world to do? What if we 10x’d its lifespan?
At some point either:
1. There is no more physics
2. We build beings that can do physics that we can’t
3. We’re stopped dead in our tracks at the limit of our ability to comprehend things that apes weren’t built to comprehend
A bunch of blind people could theoretically construct a model of an elephant, if they work together well. https://en.wikipedia.org/wiki/Blind_men_and_an_elephant
It would be cool to have a brain that sees time, just as an incidental consequence of what’s happening in higher dimensions.
Rather, journals promote well-connected kooks at the expense of less well-connected kooks and happen to occasionally publish some science as a byproduct.
So "peer review" isn't holding back kook theories at all on its own. At best, journal prestige is bolstered by it, which the journal can then translate into holding them back. The track record is a bit unclear, though.
Journal aren't just a magic upload into scientific consciousness; it takes time and energy to read and process them.
And we nevertheless pay the price of adhering closer to orthodoxy for it, since effort being put in is a rather superficial "this paper sounds reasonable to me" filter, perhaps with some boxes that must be checked like p-value or sample size.
The problem is that inherently peer review is designed to act as a brake: >Peer review, in which a scientist's paper is scrutinized by a group of colleagues, is designed to find errors and weed out half-baked ideas.
But in blocking you are going to reject ideas. >And although some have criticized peer review for rejecting new ideas just because they're too radical for the establishment to stomach
How much of a problem and how many revolutionary ideas are stopped, I don't know. I am more concerned about the reproducibility problem in science.
Peer review is also obviously a bad solution for the reproducibility problem, given that we've got the problem now, concurrent with generations of peer reviewed manuscripts.
This article mentions Einstein being offended that his paper was submitted to peer review by Physical Review, and submitting it elsewhere instead:
"We (Mr. Rosen and I) had sent you our manuscript for publication and had not authorised you to show it to specialists before it is printed. I see no reason to address the – in any case erroneous – comments of your anonymous expert. On the basis of this incident I prefer to publish the paper elsewhere."
I just recently ended up there via this blog: https://experimentalhistory.substack.com/p/the-rise-and-fall...
which argues, not unconvincingly IMO, that peer review is a failed experiment which tries to stop bad science at the price of also stopping brilliant, unconventional science, where instead we would profit more from letting the scientific public do the sorting instead of hand-picked specialists. The author phrases it as science being as strong as its strongest links, not as weak as the weakest one, and we should do everything to get the brilliant crazy ideas.
Also currently on the front page: https://news.ycombinator.com/item?id=34257150
In brief, London 'man off the street' solves 20k year Ice Age drawings mystery, published in Cambridge Archaeological Journal via contact with Durham professor. (And he is first author.)
I don't say that to disagree with you by the way - it's undoubtedly a rare example.
I tend to agree that peer review has flaws. But given the ease with which someone can publish anything with no review whatsoever it seems reasonable that peer review might shift to fill a niche that is slower and less disruptive.
Most recent example: https://www.quantamagazine.org/long-out-of-math-an-ai-progra...
This is apparently a common misconception regarding a lecture given by Lord Kelvin in 1900 concerning "two clouds" in physics.[1] In fact, Max Planck solved the ultraviolet catastrophe, also in 1900, by assuming electromagnetic radiation can only be emitted or absorbed in discrete packets of energy called quanta. Albert Einstein also solved the ultraviolet catastrophe in a paper he published in 1905, for which he was awarded the Nobel Prize in 1922, by, as always, standing on the shoulders of giants, hypothesizing that Planck's quanta were actual particles. These particles are known today by a word coined by physical chemist Gilbert Newton Lewis in a letter to Nature in 1926, namely, photons.
It’s not even just going against headline consensus issues that will get you in trouble. Any findings that overturn what a large group of scientists have spent their careers working on won’t be well received.
It’s always been a problem but I’d say these factors make it worse:
1. It’s harder for individuals or small groups to make breakthroughs. We need larger groups and more expertise as well as expensive equipment to continue making discoveries. More hands in the pie make more gatekeepers and less risk taking.
2. In general our entire society is “circling the wagons”. I’m not sure why but we’re more tribal than normal. Science isn’t immune from the trend.
I could go on. It’s a big topic.
Hasn't that always been the case? It's said science advances one grave at a time.
I suspect that there isn't a single cause but a cluster of them.
One possible issue is that high level science needs more and more energy (like LHC) and more and more intelligence to crack (most science prizes are teams).
[0] https://en.wikipedia.org/wiki/Planck%27s_principle?wprov=sfl...
> it takes a PhD before people can even understand the frontier of most scientific fields
It’s interesting that we have this implicit view that hyper-specialization is the only way to advance. I always thought so too, but why really? If you think about it, specialization is very close to incrementalism. Groundbreaking paradigm-shift stuff can often be written down on a napkin, and rarely does it require a PhD.
That said, to find treasure you have to go down a LOT of wrong paths before you (or someone else) ends up in the right one.
How to slow down scientific progress According to Leo Szilard
https://rootsofprogress.org/szilard-on-slowing-science
Discussed in https://news.ycombinator.com/item?id=34264436
The focus is clearly on "getting the grant". And the grant acts as an entity in itself, as it becomes a line in the CV of scientists, which will help them "get the next grant". The outcome of the grant becomes thus irrelevant. Maybe competition is missing?
It almost seems as a new model would take the world by storm, but it doesnt seem to arise. Even private research often falls for the same pitfalls, like how Deepmind keeps seeking to publish proprietary research in Nature, instead of creating an open ecosystem that will drive reinforcement-learning science forward.
That's the major problem humanity faces. Individuals wielding concentrated power are dangerous, but they're also at least capable of thinking about something other than their immediate short-term interest. The masses, as aggregated through votes or purchasing decisions? They're structurally incapable of it.
The problem is more parochialism than it is short-term thinking. And both sets of entities (powerful individuals and aggregate masses) are capable of parochialism.
I think this is a dangerous delusion:
Historically, centralized control has been a disaster because those individuals wielding power believed they could think of something greater… but actually lacked the skill. So we ended up with simplistic concepts that failed at scale, ideas that never could work because they missed key details, etc. By contrast, society gets on with the business of building the future in spite of and not due to that meddling by centralized power.
I think there's plenty enough of competition in academia. And competition alone doesn't help, because the metrics over which one loses or wins the race are not correlated with short or long-term worthiness. And while we can describe, even if somewhat vaguely, what makes worthy science, there's no way I know of to have a stable system in which that metric drives funding.
> Deepmind keeps seeking to publish proprietary research in Nature, instead of creating an open ecosystem that will drive reinforcement-learning science forward
This research at least has some feedback: Google funds it because it expects to make money off using its results in practice, so the research has to be at least somewhat correlated with reality. Unfortunately, this structure also means that "open ecosystem" isn't pursued.
The volume of papers has increased significantly, and publish or perish kinda stinks, but it's not just an issue of funding, but some (many?) researchers publish and exaggerate about the importance and difficulty of their research to receive an "I am smart" badge on social media. Although, despite this increase in volume of crap tier papers, the article seems to think it's not correlated:
"Declines in disruptiveness are also not attributable to changing publication, citation or authorship practices..."
People can point to incentive structures like grants etc, and to the implications in terms of the meaning of metrics etc but the real harm is in how it shifts mental focus and attention collectively.
It's not ok to let minor things or things you think are obvious go anymore to focus on bigger picture issues, that might be "riskier" but nonobvious. The way that plays out too is incredibly dependent on your institutional social environment etc.
So much of this is difficult to easily quantify in ways that can be easily studied. I loved this paper for example, but it's easy for me to think of papers in my own field that would look "disruptive" in terms of citation networks but are really the same content. There's these weird shifts I've seen happen in reading old literature and during my career, where big shifts in who is being cited will happen for quixotic social reasons. Usually it's basically politics or ignorance of big parts of a field, who will get introduced to an idea by a particular author or group. Lots of chaotic citation patterns and feedback loops.
Add in shifts in what's motivating grant and paper topics and it really damages authentic scientific discourse. So much is driven by looking like a brilliant scientist and not by scientific progress. People aren't dumb either and they are very very good at looking like brilliant scientists.
Take any revolutionary discovery. Compare making it freely available to anyone, or limiting the availability to those holding the patents. EDIT: And perhaps more importantly give people the freedom to innovate on top versus stopping anyone who does. Shouldn't the former always be more disruptive of the status quo? I can imagine this is an oversimplification though so I'm open to hearing counterintuitive examples where the opposite is true.
And to be clear, either way I'm not saying this is good or bad in and of itself - in some cases it might slow down essential improvements, in others it might be good to have speed limits for change so that things don't break.
And you can see this play out in the history of invention. For example, Edison stopped other people from doing research on the light bulb. Meanwhile the Netherlands was a country without IP laws at the time so Phillips quickly ended up producing the superior and cheaper product. No patent laws, bigger disruption. Similarly the Wright brothers were notorious for holding back aviation with their patents.
On the other hand it is also case for more innovation due to necessity, e.g. his patent on beam connection to a shaft (linear-to-rotary movement) required others to come up with a different solutions (e.g. sun and planet gear).
- There are as many revolutionary discoveries with and without patents
- Without patents, discoveries would be freely avalaible
As far as I know, two (related) arguments are generally made for patents:
- Patents create an indirect (by preventing the competition from using your invention) or direct (by licensing it) monetary return to innovation, potentially leading to more innovation
- If a company wants competitors not to copy their innovation, they can 1. keep it secret or 2. disclose it and patent it; without patent the choice is between 1. keep it secret or 2. disclose it and have everybody copy them. In this case, patents lead to more innovation being made freely avalaible (with a delay!).
Whether patents lead to more or less innovation is, as far as I know, contentious.
I explicitly don't: I mentioned the patent stops others from innovating on top of the patent. The necessity of patents to be public is one I had not considered though, being a force against trade secrets is a good counterpoint.
> Whether patents lead to more or less innovation is, as far as I know, contentious.
Honestly, I really have trouble accepting this. Not that I have an answer, but with the idea that researchers haven't been able to find some sensible way of measuring this by now. I already mentioned the example of light bulbs. Surely there are enough similar historical scenarios available to analyze where one can make use of global differences in IP laws and other variables to simulate control groups?
"Take any revolutionary discovery. Compare making it freely available to anyone, or limiting the availability to those holding the patents. [...]"
You take the situation where there _is_ a revolutionary discovery, with or without patents, and then wonder about the effect of patents on the next innovations. In doing that you do not consider that may be a revolutionary discovery with patents, and none without.
Extruded plastic took off before resin printers because the patents expired a few years earlier.
As far as I can tell, we could have had the 3D printing revolution in the 90s or early 2000s, but instead we locked the technology away to only be used in a few esoteric commercial applications until the patents expired.
How much is the license cost?
Perhaps if there were no patent, there'd be more third-party research to improve all these things faster. But as far as adoption of the existing stuff goes, I don't see how license costs are a direct blocker.
The original idea of a patent is that it must reveal the "secrets" behind an invention, in return for a "temporary monopoly". It's supposed to help diffusion of technology by publishing trade secrets, not to "incentivise" research.
I think what is actually happening is that patents are made obtuse/opaque containing little meaningful information. They are then used as playing cards in patent licensing deals between big corporations.
There is no longer any hope for patent clerks to do meaningful inspection of patents to determine if they are legitimate. Instead, it's left to lawyers/attorneys to play something out in court.
The fact that I was looking into battery materials (a notoriously over-hyped sector) certainly didn't help, but some of the approved parents were laughable. They included core materials which had been historically known to be useful in batteries (and hence shouldn't qualify for protection), but then also every possible element-wise substitution of the same structure - many of the resulting materials would not even have been stable, never mind viable as a battery component. There were also instances of multiple patents covering the same material, suggesting that any prior-art searches were not being done rigorously.
Everyone is free to innovate on top of any pre-existing, current patent. They simply can't profit on their new innovation without licensing the patent. However what they can do with their new innovation is patent the innovation itself. This newly patented innovation prevents the original patent holder from profiting off of the new innovation, and thus incentivizes the original patent holder to license their patent to the new patent holder (or vice versa).
IMO, it'd be much better to look at the Avg. CD of the top N% of CD scores over time - that'd give me a much better idea of if disruptiveness is changing, as disruptiveness is a "best paper" phenomenon anyway.
How much of this is due to how much easier it is to publish something these days? Every statistics package spits out publication-ready charts and tables, Latex/Word has automated much of the typesetter's art, drafts can be communicated instantly via email, there are instantly searchable and linked databases of papers, etc, etc.
Think about what it was like in the 1950's, when many scientists were hiring typists to convert their longhand manuscripts into something an editor could read, attaching hand drawn figures, then heading down to the post office to mail the draft in. That draft then had to be copied and mailed to reviewers, who then mailed their opinions back, every step taking at least a few days of communication overhead. It was a different world and it gradually changed with widespread adoption of different technologies.
Also how does this interact with another pair of common suggestions for improving science: publishing negative results and doing more replications? Is this a sign that those two good practices are already occurring? Is it possible that the higher rate of disruption was due to shoddier work in the past?
This was done well into the 1980s, if not longer.
Still, many less techy problems remain, at least climate change and wealth inequality come to mind. Perhaps we'll see the biggest innovations in solving those in the future.
Also, wealth inequality isn’t a problem: how wealthy the people at the bottom are is. Wealth inequality is associated with fewer poor people and a higher standard of living because wealth inequality requires a wealthy society.
What exactly are you “fixing”?
Concerning wealth inequality, I do think it is a fundamental problem of society if a fraction of a percentage controls majority of the available resources even if the lowest half has their immediate needs met, but we can disagree on that all day and arrive nowhere.
My concern with your view is that in trying to “squash the gap”, you end up doing so by making everyone poorer — and society worse overall.
Happy to agree to disagree; hopefully we can consider each other’s perspectives in crafting policies.
Wealth inequality implies that some people disproportionately control the life production of other people. If you're in favor of a "great man" idea of progress this is a benefit. If you're in favor of individual power to direct production and innovation (along the lines of de Tocqueville or it's left-wing equivalent) this is a detraction.
I think wealth inequality is too ill defined to even be a problem. If anything i would argue it is only a problem if class mobility is low and that is where the focus should be.
- Tenure requirements
- Job duties (both requirements and limitations)
- Elimination of corporate skunk works like PARC
- Credentialism
- A "Pain point" focus would tend to be more incremental than truly disruptive
On the good side there may be some decrease of disruptive inventions and discoveries because we are also focusing on eliminating the externalities of current and previous inventions and discoveries. It's a lot easier to make wildly new things if you aren't worried about their side effects.
This is a big problem. PARC, Bell Labs, IBM Research are the first that come to mind. Same is true for mostly government-funded initiatives like the one where Marc Andreessen invented Mosaic browser, that eventually became Netscape Navigator. I don't recall the project name. Do we even do that anymore?
A variety of DOE national labs and core facilities (including NCSA where Mosaic was developed) still exist and are funded.
We may need more to take up the slack of the corporate research facilities.
My mind is genuinely struggling...
Certainly not Curie, Einstein, Fleming worked in a hospital.
Maybe the problem is...
Undergraduates!
Your income level is probably higher after leaving academia and going to work in industry. So you probably starve more by staying in academia.
I guess part of the reason I exited academia was because I was bad at conducting research :)
" What is a 'Seed of Science'? It is a speculation, an idea for an experiment, a novel observation, a thought-provoking question and discussion, the highlighting of an under-appreciated problem, or an unorthodox research study"
They got an ACX Grant last year from AstralCodexTen to pursue this.
---
CoI Disclaimer: My work is published in SoS.
Albert Einstein failed peer review the one time he was subjected to it. Newton was not subject to it at all.
His 1905 seminal papers were not peer reviewed per-se but the journal's editor who accepted the paper was Max Planck! So it was reviewed by another legendary physicist.
These days science has branched off into such specialized areas that another specialist in the field is needed for review. Whether that should be an anonymous referee, or someone else it is still valuable to catch errors and determine the soundness of the techniques. Also the selections of those peers is something for debate too.
[1] https://physicstoday.scitation.org/doi/full/10.1063/1.211782...
All these geniuses... people are giving them jobs!
Because it's boring and depressing. And one thing money can't buy, unless you're solely interested in accumulating money, is satisfaction with your life.
Obviously not true for everyone, but we've got 8.8 million "scientists" (I'm not sure what they mean by the word). https://sciencebusiness.net/news/number-scientists-worldwide...
Most scientists don't work in academia. Some of those who do have stable jobs as staff instead of faculty positions.
And this guy did it in his spare time while working at FAANG: https://news.ycombinator.com/item?id=34236889
I hate to be a doomsayer, but this seems to be a general trajectory that most of the western world is on.
But this is an effect of the simplification of history, and when you read the detailed technical histories you can see how remarkably incremental the old innovations were.
We can easily see the precedents of recent innovations but not for older ones.
This is pointed out in the book I am reading on the development of the turbojet, which only looks more unprecedented than it really was.
This intuition seems to overlook the culture of citations in the patent space. There, the incentive is for applicants to be over inclusive with citations. This is combined with the increasing ease with which references can be found.
But let’s accept that we are seeing less fundamental breakthroughs in science. Is that a failure of the system…or, depressingly, what if we really are hitting the limits of what we can model and describe?
In a roughly 200 year timespan, we figured out the at least the rough outlines of…wow, where even to start…electromagnetism, general and special relativity, quantum mechanics as it relates to field theories and chemistry, genetics, and computability. We really picked a lot of the low-hanging fruit, and what if there just isn’t any fruit left to easily get at? There’s certainly things like dark energy that we don’t have particularly satisfying theories for at all, but, the basics? We seem to have good models.
Famous Last Words
This and many other pieces of very useful knowledge are there for the taking in the metaphysical realm, just as was the case hundreds of years ago in the physical realm, prior to the rise of science (a metaphysical phenomenon itself).
In the early 20 century new ideas in physics and mathematics sparked a revolution and we are slowly unraveling the benefits of it ever since. For 40 years we had evolution in the AI research field, now we have a revolution, I do not know when it will die down but after a while we will just start to build over the ideas we discovered and a long period will pass before we do another jump. Same with psychology, economics, biology...
shrug
There was a time in the 19th century when people thought physics discoveries were complete. A few decades later all hell broke lose.
The problem we have right now is that we are stymied on the complexity front (materials, chemistry, biology, medicine etc)
Our mental / mathematical models that we use to understand the world and even enable us to ask original scientific questions are losing their edge when things get complex. These tools are 19th century stuff (including all the AI junk) and we have milked them for all they are worth
I would hypothesise that a new period of discovery will correlate with some breakthrough in the mathematics of complexity. We had a few false dawns but not the real thing
I'm not really sure what you're looking for here. I mean, you could claim that modern cryptography is based on ideas from the ancient Greeks (factoring numbers), and is therefore old and busted. Is every idea just derivative unless it uses a completely new kind of math?
Software developers (maybe not you, but a large percentage of the users of this site) tend to automatically think older ideas are worse. But that's not true in math. Mathematical ideas don't expire. The idea of the number 7 is aeons old; the number 0 over a thousand years; the complex number i 450 years. But the numbers 7 and 0 and i are still extremely useful, in infinite combinations. If you think we have milked those numbers for all they're worth, you have an extremely limited imagination.
Sure, most of the "AI junk" is a combination of old ideas like linear algebra and gradient descent, but used at a scale orders of magnitude larger than their creators ever imagined. (And quantity has a quality all its own.) It's that combination of ideas that is so powerful. Besides, backpropagation is a pretty damned novel idea. If you're looking for the one truly new idea from AI, it's backpropagation.
You are vastly underestimating the volume of the idea space available to explore simply by combining the ideas and materials we already have. Crossover is a better source of variety than mutation is. You may be right that truly new concepts are the "raw materials" of invention, but we never run out of the raw ideas we have, and they prove to be useful in their combination up to the limits of our imaginations. It's actually surprising how likely you are to come up with a reasonably good idea by simply considering a few randomly-selected ideas together.
> we are stymied on the complexity front (materials, chemistry, biology, medicine etc)
Again you seem like your standard is: if it's not an entirely new thing where nothing like it has ever existed before, then it's boring!. If you think the volume of the idea-space from combining the materials, chemistry, biology, and medicine we already have/understand is X, it's almost certainly at least 2^X, or even X! (factorial).
Besides, we are coming up with brand new materials all the time! Materials science is one of the most exciting frontiers of human technology.
Old math can still have an impact at the margin (eg speeding up protein folding calculations etc) but its not the cognitive leap that would allow us to steer discoveries in more effective manner. Materials, medicine etc all very exciting and very active but also hampered by their intrinsic complexity.
>Although the proportion of disruptive research dropped significantly between 1945 and 2010, the number of highly disruptive studies has remained about the same.
So... maybe this is because having more people doing scientific research doesn't yield more breakthroughs. More people doing scientific research DOES yield more incremental achievements, as the study found, i.e. word frequency analysis showing more occurrences of "improve", "expand", "extend".
Why not phrenology?
That's just intentionally crippling our own understanding of the world, for fear that we might not like what we discover. That's hardly an unprecedented behavior, to say the least, but the cultures that have gone that direction are probably not the sort we might want to emulate.
I assume only people like me and maybe a few like you will know what this means, but doesn't it upset you that I can freely say it?
The war may have had something to do with it too, I know a lot of advances (like radio / microwave engineering) have roots in the war efforts. But I suspect it's more of just a dry spell because we pushed forward in the 18th-20th century and ran out of ideas
plenty of counter-examples, I'm sure, but like once you know the structure of DNA and it's fundamental operations, the remaining discoveries will have a smaller impact.
Let's take AI for example. Research papers that were published in 2019-20 have become obsolete within a year or less, and consequently get cited less and less, as more recent and derivative papers introduced better models. Does that make the older papers any less 'disruptive'? No!
On the patent front, well patents are for parasites.
Interesting observation.
Also I suspect most innovation are being done outside of papers, especially when you have "reviewer 2" ready to be nitpicky at every minor detail, so you make it more bland and maybe skip one juicy part or another
Lower corporate taxation incentivizes the hoarding of cash. Prior, corporations were incentivized to allocate more funding towards their industry research (IBM, Bell Labs, etc).
Would love some counterarguments, here!
How to slow down scientific progress, according to Leo Szilard - https://news.ycombinator.com/item?id=34264436 - Jan 2023 (20 comments)
Gone into universities as bright young people. They come out of them brain dead, not even knowing what science means. They think it means peer-reviewed papers etc. No, that's academia. And if a paper is peer-reviewed it means everybody thought the same therefore they approved it.
An unintended consequence is that when new knowledge emerges, new scientific insights, they can never ever be peer-reviewed. So we're blocking all new advances in science that are big advances. If you look at the breakthroughs in science almost always they don't come from the center of that profession, they come from the fringe. The finest candle makers in the world couldn't even think of electric lights. They don't come from within they often come from outside the bricks. We're going to kill ourselves because of stupidity."
— Allan Savory, Zimbabwean ecologist, research biologist, livestock farmer, and president and co-founder of the Savory Institute. He originated the Holistic management of agriculture to fight desertification.
"Holistic management of agriculture to fight desertification" is a non-quantitative science and falls under a very different regime from, say, people working at the state of the art in particle physics or biology.
Einstein was only subject to it once and didn't pass. Newton, Galileo and the rest were not subject to it at all.
Newton and others all published and then discussed the details in person, a lot of the peer review was "off-paper" and "post-publication".
Also, the vast majority of scientific breakthroughs, by count, occurred after peer review- mainly due to the unbelievable pace of breakthroughs due to technological advancement.
According to this [1], Einstein's paper that didn't pass peer review turned out to be wrong (it said gravitational waves didn't exist), and was later published elsewhere with different conclusions and fixing problems that the referee's report had already pointed out.
> The irony, of course, is that Einstein could have found that escape route months earlier, simply by reading the referee’s report that he had dismissed so hastily. The referee had also observed that casting the Einstein–Rosen metric (as we now call this solution of the Einstein equations) in cylindrical coordinates removes the apparent difficulty.
[1] https://physicstoday.scitation.org/doi/10.1063/1.2117822
Only knowledge that originates in academia, and even then only a portion of it.
> because for every important problem in academia, there's a competitor who is the best person to point out problems with the theory.
While this may be generally true, it is not absolutely true.
> "Holistic management of agriculture to fight desertification" is a non-quantitative science...
Partially non-quantitative.
Also: it'a (partially) true, but so what?
> ...and falls under a very different regime from, say, people working at the state of the art in particle physics or biology.
Right: one is known to be important, and the others are (at this point in time) hoped to be important (as they have been in the past).
it's called bureaucracy
Major breakthrough that disrupts the paradime. Followed by normal science with diminishing returns until the next paradime.
Consider for example deep sea archaeology. When you find a shipwreck, there's going to be a flurry of activity as people study the shipwreck, and the more they answer the questions they started with, the more new questions they are going to want to ask, so even long after the initial discovery there's going to be a continuous string of studies refining knowledge about that wreck. And this is true of every wreck. But finding one shipwreck doesn't generally make it easier to find the next, assuming the same resources, new wrecks are going to be discovered at a pretty steady rate. So the number of discovered wrecks grows linearly, but with each being revisited regularly, the cumulative amount of research done on all wrecks grows geometrically.
Similarly discovering a new chemical opens up many possibilities for studying that chemical, likewise for a new phylum of bacteria, a new type of star, you can only see something for the first time once, but you can take closer looks forever. We should expect the ratio of new things to look like a graph of 1/x.
Perhaps it's weird that the absolute number of disruptive papers is holding steady. While there are competing forces at work - on the one hand we have more scientists with better instruments, but on the other hand the low hanging fruit is gone and they must go ever further to find something new - there's no fundamental reason these two forces should perfectly counterbalance. I might posit our society has a certain appetite for discovery and we adjust our allocation of resources to science and innovation to maintain this continuous stream of disruptions, but I at least feel that scientific breakthroughs tend to excite people about science and desire to double down on science, not pull back. Perhaps more realistically we just move the goal posts as to what we consider disruptive such that the top X discoveries of the year are always "groundbreaking." Regardless of the why, this definitely the more interesting part of the story.
Really?
Too much talent going into trivial tech ventures. The shift towards rentier capitalism in the US, especially in terms of cheap money/QE being used for share buybacks instead of R&D. The colossal cost of higher education in the US, with grad stipends barely budging until literally this month. Fewer people willing to tolerate the (mostly) toxic culture of academic advisors overseeing Phds and post-grads. Google Scholar skewing papers to exist more for popularity/citation prevalence. The Internet as the biggest distraction in the history of humanity. etc. etc.