Commercial research was responsible for conceptual refinements of existing insights, and for innovations developed as industrial and consumer applications, but not for the original intellectual breakthroughs that made specific classes of inventions and devices thinkable.
Computing itself was state-sponsored - first with Turing+Church's theoretical work, then with various practical implementations whose develop was accelerated by WWII, then with state subsidies and grants for further military and industrial development.
Recently state funding has become less and less effective because of the financialisation of academia. Universities now have explicit financial goals and lean heavily on the grant funding game.
Previously the culture was far more tolerant of blue sky inquiry that might not have an immediate payoff - or any payoff at all.
I would also suggest that computing began no later than Babbage; Turing definitely contributed to the field, but Babbage and Clement kicked it off (largely with government funding).
That, along with the decrease in overall amount. State funding for research seems to be more limited than it used to be during much of the last century; especially if we normalize by population size.
Are discoveries harder in bigger countries? So would we expect a single researcher to be more productive in England than in the US?
Higher GDP is correlated implies higher per person productivity and more specialization to decommoditize themselves vs the global labor. Plus pushing the envelope often requires more funding for novel projects as there isn't an off the shelf for something novel to reduce costs.
They aren't perfect measures by any means (there exist things which boost either not research) but are decent enough proxies.
If the number of corporations needing new discoveries keeps growing but the funding is fixed, that means either the researchers are spread thin or there’s a centralization of research that has a hard time transitioning into the wild. (Both situations exist, and it sucks.)
Your argument about a minimum size research group for productivity is interesting. I am not quite sure whether that means we should normalise by population size, though.
Ever heard of the Solvay conferences?
While not embedded in one company, these confs were definitely sponsored by an industrial. And strongly impactex 20th century advancements in physics.
Langmuir flow, plasma physics, and fundamental surface chemistry. Basically all of organic chemistry, too. Hormone biochemistry, etc.
People seem to get confused on what actually goes into bringing a new drug to market. Yes, a significant amount of basic research is done through government funding. But discovering a new drug target is about 20% of the way to a new drug.
You still need someone to develop new candidates, screen them, identify promising compounds, bring it through years of clinical trials, develop a manufacturing process, get FDA approval and then sell it.
A good rule of thumb is that the cost of R&D is typically 1/3 research and 2/3 development. Development is almost always entirely private. And a significant amount of research is too.
From talking to some of the old-timers, universities started to get better and better funding (NIH grants, royalties, etc) and eventually could buy the equipment they needed themselves, no need to rely on drug companies.
So it might be more accurate to say that basic research funding has grown and most of that growth happened on the university side. Pharma companies have either held steady or decreased their spend on basic research (it really depends on the company).
Certainly with the costs associated with drug approvals being exponentially higher today, only Federal government could possibly be in the drug business.
A single phase 3 clinical trial can run into the hundreds of millions of dollars. You can fund a lot of basic research for that kind of money.
When I was in grad school, my PI got grants of ~$2M a year and that supported a lab of ~10 researchers.
There's a big pay difference between grad students, industry researchers, and government researchers, with grad students being the lowest on the totem pole. Which should make sense because they are researchers in training. There's many different types of research too.
They reportedly were so successful in a large part because of re-using a lot of data already pain-stakingly produced by publicly funded research labs.
But the state (at least the US state) has been cutting spending on research over the last several years - and those cuts have accelerated under the current administration which does not seem to understand the value of basic scientific research.
It seems obvious that they haven't.
I don't think it's anything nefarious about government funding (or lack thereof) that did it; it's just we've already picked all the low hanging fruit. Plucking all the undiscovered ideas at the frontier of knowledge will be increasingly expensive, because if they were cheap to discover, we already would have.
Though I always wonder if there's some trove of cheap ideas hanging out somewhere...
Also, if I my toot my own horn, I think better programming languages has the potential to revolutionize all knowledge work, making everything else more productive, and helping unlock whatever other basic research might remain woefully untapped.
I have to admit some skepticism. What follows is my opinionated thoughts as a naif: new languages have helped a lot in creating new models of social organization. To be concrete, the significant 90s languages were forgiving enough, powerful enough, and easy enough to learn that they allowed for a kind of triumph of the programmer proletariat over the artisans. That allowed for the explosion of information technology that we've been observing for the past 3 decades, which enables new business models and makes old ones substantially more efficient.
But I don't think a contemporary PL person would think of Java as a prime exemplar of the value that the research community brings to the world. And when I think of more recent language innovation (Go, Rust), they've only been incremental improvements that were anticipated decades ago. And although I love them both, neither has shown any signs of enabling new social modes like earlier languages did: my like of them is purely aesthetic. All the real, productive low hanging PL fruit has already been taken.
Opinionated, like I said =)
For someone focused on the language itself, I agree. But consider the JVM: an enormous amount of research has been done on, and applied to, the JVM. Garbage collection and just-in-time compilation are two particular fields that the JVM has benefited from enormously.
Glad you agree!
> To be concrete, the significant 90s languages were forgiving enough, powerful enough, and easy enough to learn that they allowed for a kind of triumph of the programmer proletariat over the artisans.
In terms of practice, yes the languages that would go on to be popular and didn't drag one down with manually memory management were created in the 90s. The technology was older.
> Java as a prime exemplar of the value that the research community brings to the world
Agreed. Still not as un-innovative as Go though!
> And when I think of more recent language innovation (Go, Rust), they've only been incremental improvements that were anticipated decades ago.
Rust fully admits it is recycling old research. It's productivity gains are diminished by the fact that it is targeting systems programming which is less productive. If somebody made a a Gc<T>, and made an implicitly garbage collected version of Rust to compete with Go and Java (of course with great FFI to regular Rust for free), it would already be a better candidate than any other popular language, and that's no research required.
> I'd love to hear some more details of the untapped engineering potential!
So I am Haskeller. I would say there are 3 tracks that interest me:
1. The "classic track" of fancier type systems. The Dependent Haskell work is interesting because it should allow the hodge-podge of language extensions we have today to be streamlined into fewer features, reducing complexity. Writing proofs means less productivity, however, combining libraries with lemmas might be extremely productive. Think being able to blindly mash stack overflow answers together with extreme fidelity. That should be a goal.
2. The "categorical strack". Lambda the ultimate....mistake :D. Our programs are higher order, but there is often a first order steady state, like the things we put on whiteboards for our managers. Our current practices utterly fail to capture that first order steady state anywhere in the code, and as languages become "more functional" (java 8+, ES whatever, etc.) we're basically throwing more lambdas at problems without principle.
Using some categorical primitives---and I mean the "realer ones" in https://hackage.haskell.org/package/categories not the ones in https://hackage.haskell.org/package/base that need a lot of squinting to make out the original math---gives us a chance to perhaps fix this. Categories putting the output and input on equal footing helps, and the simple wire diagram makes the plumbing / dataflow way of thinking that every programmer should employ much more apparent. "point free" programming sucks, but I ultimately think we can get something like Scratch that makes sense for programmer-adjacent types, and yet is useful for "actual work".
3. The "incremental track". There is a decent amount of literature around incremental / reactive programming and models of computation, but it needs to be properly synthesized. I think this is will be huge because it will improve techniques that actually match what real world programs do (toy "everything terminates" stuff from school is actively harmful when students extrapolate that theory has nothing to do with practice). This will defeat an entire class of concurrency woahs that currently is a huge source of productivity loss---I don't mean just race conditions, but the more general "how can I analysis (with the ooriginal "break apart" connotations) my program into simply peaces and then synthesize a holistic understanding". In terms of how this will actually happen, this strongly ties into the above.
Let me say 2 more things:
1. we should have "one language to rule them all", in that I can write my firmware, kernel, end applications, and everything in between without compromise in it. People think this is silly, "engineering is tradeoffs, amirite?" but it isn't. The language will begin a "dialect continuum" that supports vastly different clashing idioms for those different domains (GC everywhere? Manual memory management? No call stack even?), but do support flawlessly type-safe FII.
2. Non-programmers think "applied math" means plugin in numbers, and writing regular natural language prose to with it. Math isn't numbers, and people's notion of it must be fixed accordingly. Proof theory means the whole argument, not just the numerical proessing, can be formalized. This is how many jobs should work.
Do you have any references to share to learn more about potential practical uses of categories? Assume the knowledge of someone who isn't intimidated by the wiki page on categories but who learns by reading it.
Sure! Hope it wasn't too rambly.
> I was expecting a bunch of stuff in the "fancier type system" category and was pleasantly surprised.
Yeah after enough functional programming, one's sense of the new lowest-hanging fruit switches.
> Do you have any references to share to learn more about potential practical uses of categories?
I'm afraid not. I suppose there is plenty of blogs on Haskell and category theory, and N lab for the actual math, but I don't of a resource that zooms out from the neat tricks and tries to discuss broad needs for better programming and architecture.
But I'm relieved to say that where I work, we are working on some things in the vein of things for tracks 2 and 3. It will be open sourced when it's ready, so... stay tuned, I guess?
NIH budget for example increased from $32 billion in 2016[1] to $41.68 billion in 2020. Trump has proposed a 6% cut for the 2021 NIH budget but that is not finalized yet.[2]
NSF budget increased 2.5% from 2019 to 2020, to $8.3 billion. [3]
NOAA +4%. [4]
DoD Science and Tech: +1%. [5]
[1] https://www.nih.gov/about-nih/who-we-are/nih-director/fiscal...
[2] https://www.the-scientist.com/news-opinion/trump-proposes-si...
[3] https://www.aip.org/fyi/2020/final-fy20-appropriations-natio...
[4] https://www.aip.org/fyi/2020/final-fy20-appropriations-natio...
[5] https://www.aip.org/fyi/2020/final-fy20-appropriations-dod-s...
I guess the flipside of this is that, for a business, it makes way more sense to give money to a university than it does to start your own lab. You'd be paying real (and probably decent) wages to people who wouldn't be nearly as motivated as the phds who do the same work for less than minimum wage.
Everyone likely agrees that most large companies should have some longer-term speculative projects going on. As those companies get wealthier and fatter, those research organizations want to grow, they help position the company as an "innovator," and they may even have an OK payback over a long enough time horizon.
As to that time horizon, people have been arguing about that and it's swung back and forth at different forms for as long as I've been in the industry.
I think it's also true that there were always a relatively few large true research labs. There's a reason that, say, Bell Labs, Xerox PARC, and IBM Research come up (in roughly that order) and then most people have trouble naming examples.
Chiplets: https://www.darpa.mil/program/common-heterogeneous-integrati...
PIPES - optical chip io : https://www.darpa.mil/news-events/2020-03-16
Automated ML chip generation: https://www.darpa.mil/attachments/Real%20Time%20Machine%20Le...
A more dynamic view might better explain what’s going on: the flow of public money, the grants criterias, the state funded researchers gone to work for private companies for some years and then coming back…
Some example of this entanglement:
- in France any company can ask for a dedicated Credit Impôt Recherche, a tax credit for research purposes. Most seem granted for projects that are very light on fundamental research. That allowed many small companies to set up research/lab teams. It seems a more lightweight continuation of the European policy to help companies to integrate research and to create consortiums.
- when looking up public AI research in France, I found that nearly all the top AI/ML researchers based in France have been recruited by Google and Facebook. They are still fully part of their original, state-funded labs, but now work and are paid by Google/Facebook. They seem to still do the same kind of research they were doing at their state-funded labs.
I’m sure there’s other examples in other countries
Not trying to be some fascist statist, let's just recognize that since the 1980s, we've not been in a new gilded age so much as a gilded age cosplay (e.g. we are not actually closer to well functioning free-market capitalism for the "real" economy), and growth and large-scale investment have stalled accordingly. A lot of libertarian and decentralization ideology has grown up in the era since under false pretenses.
I would love devolved power and decentralization to actually work, but we need better, more honest theory on how to make it good, and less blind faith that centralization and monopolization will fail because that bad and "the arc of history bends towards justice".
I mean come on. The only way you will get to that theory is if you have brave people who try (maybe that's self serving), and if you aren't at least questioning the narrative of "the only way forward is through the state", especially when there is (increasingly sporadic, anecdotal) evidence to the contrary, then you are tacitly discouraging people to try.
I share my doubts here because HM tends to overly romanticize small organizations. I don't mean to tell people to give up and work for FAANG.
Without assigning causality (I happen to think that the primary cause is "low hanging fruit" phenomenon), the transition to state sponsored research has coincided with a general loss of scientific productivity in many fields, and even in some fields with explosive growth, like genomics, there has been major corporate involvement (half the human genome was sequenced by a not just private, but a wall-street traded, entity).
Yet the narrative of late has been "if the state doesn't sponsor it, nobody will". I don't feel like I have a solid understanding of why corporate research labs were a thing, or even, how science got done in the pre-manhattan-project/Vannevar Bush era.