A big part of the "DARPA formula" that doesn't seem to get much treatment here is consistency. DARPA is going to spend a few billion this year, and it spent a few billion last year, and it spent a few billion twenty years ago, and you can feel confident they'll spend a few billion next year. That means that if an academic researcher or similar can position their lab in an area that DARPA likes, they have a good shot at reliable funding for the length of a career. With all the economic uncertainties around research, this is tremendously appealing and will cause a lot of smart people to re-orient their entire research programs around DARPA-friendly subjects [0]. A funding organization that might disappear in a few years, by contrast, will get plenty of grant applications, but won't attract nearly the same level of researcher devotion.

(As a side note, I want to pull out a quote I thought was really nice, hidden a ways in:

"DARPA funds wacky things that go nowhere. DARPA programs have a 5—10% success rate and have included things like jetpacks, earthworm robots, creating fusion with sound waves, spider-man wall climbing, and bomb detecting bees. You can’t cut off just one tail of a distribution.")

[0] Which may, incidentally, go a ways towards explaining why DARPA heads describe their projects as "idea-limited"

IMO this analysis gets caught up looking for answers in process when the real problem that DARPA solves is a political-economic one.

ATT didn't invent the internet not for lack of process, they were extremely innovative. They held back on innovations because they were making huge sums of money overcharging long distance mainframe-mainframe data link rates in the 50's-70's. DARPA succeeded as a trust buster.

They have the resources and legality to make whatever they want, regardless of patents (because National Security/patent law). They can make technology a reality, then develop political support with working prototypes. They build things to show how the Government is getting screwed over by some giant defense corp, because of a lack of competition in certain types of contracts.

I think they are becoming less relevant because the corporate-political landscape is becoming more trust-based.

DARPA is effective, but this document paints a rosy picture that is far from the truth. I've been a part of DARPA projects both commerical, university and on the government side.

The real process is:

1. PM gets picked due to knowing someone or being a former employee.

2. The biggest test of a program isn't if it's doable or a good idea, but if its able to be transitioned to another government agency with deeper pockets.

3. Most contracts are lost before you begin writing, as people have insider information about what the PM wants. This is done through just talking with each other (remember that most of the PMs come from the same companies), and not through any other formal process.

4. DARPA has some really cool stuff, but fails to transition it well enough (leading to 2.)

DARPA is not without it's problems, but has a better track record then NSF (NIH has them beat). What is funny is that you quickly realize how much bunk there is in scientific research and how many papers are not replicated.

A bill, the Endless Frontiers Act, has been introduced in both houses of Congress to make a section of the National Science Foundation that works more like DARPA, while massively increasing its funding. The idea is to extend the DARPA model to many more non-military areas, just as suggested in this article.

https://www.sciencemag.org/news/2020/05/us-lawmakers-unveil-...

Certainly a timely article. I don't know whether it's a good thing or not. The DARPA model has certainly been productive, but it isn't suited for every research topic or subdiscipline.

As discussed in the article, currently NSF is the most open to basic science of funding agencies, and gives grantees the most latitude in what they work on. That is a valuable thing to have in the ecosystem.

If Congress expects the majority of publicly-funded research even from the NSF to be on short-term grants for specific visions and technologies, it will rule out working on a lot of important things.

Small anecdote, but good experience with DARPA and SBIR both. One thing I liked - they seemed pretty low overhead / low hassle and outcome / objective focused.

That contrasts VERY strong with most govt contracting which is under allowable cost or cost reimbursement setups where the absolute most important criteria is to bill enough costs to draw contract in the right cost buckets (which can be super annoying if local agencies require super complicated budget mod processes).

In the cost based contracts, the focus really focuses on the accounting for the costs and other compliance related items. Ie, did you send someone to a conference with govt money, how can you prove you didn't use the govt money in this way or that way etc etc. Bam, welcome to personal activity reports with fund codes that no one understands (ie, major universities have an insane number of codes), and all the nightmares that follow including a fair bit of rule pending that even normally ethical folks find themselves being asked to to get through the paperwork.

Seriously, you deliver the product at 50% of cost? You will get a nasty note from the head of your agency saying make sure you draw full contract because agency budget depends on the indirect portion of this award (ie, 30% to overhead) and even the govt agency supervising (who also budgeted based on a cut of full contract) AND other folks for whom leftover money makes life difficult (harder to close contract etc) AND because a lot of govt funding is on the repeat what we did last year model so drawing down everything avoids a cut next year when you may really need it.

You determine it would be cheaper to do x vs y and that required a budget mod? Wait 1-2 months for commission approval if you even bother trying to fight it through (changes < 10% often ignored thankfully).

I don't know how the accounting for SBIRs etc work, but somehow those projects always seemed more results oriented (so a LOT more fun to work on, focus is on getting a solution going).

The irony is palpable:

> I would rather this be read by a few people motivated to take action than by a broad audience who will find it merely interesting. In that vein, if you find yourself wanting to share this on Twitter or Hacker News, consider instead sharing it with one or two friends who will take action on it. Thank you for indulging me!

I'm glad of course it was shared here. As a distillation - I think the author's theme lies with enabling more _researchers_ rather than business people to take moonshots and do foundational knowledge building and discovery that redefines a field and then focus on commercialization from a birds eye view by technically capable visionaries.

This is the SBIR [1] model (also a US Gov requirement to fund small business research for any federal agency with >$100M in funding), the Bell Labs model (which yeilded amazing foundational work like UNIX and the transistor and was a direct result of AT&T's monopoly and excess resources), and perhaps even the YC model (though that one is obviously focused on a shorter horizon and more on commercializing existing tech and more rarely on foundational research).

I've personally thought about this problem a lot and done this at a small scale and would love to expand upon it. https://augmentedlabs.org Would love to hear others experiences and thoughts.

[1] https://en.m.wikipedia.org/wiki/Small_Business_Innovation_Re...

Does DARPA work(anymore)? The first example of how it works was it's "badass program managers", but all of the program managers I've met have been, to put it lightly, idiots. One of the worst was the PM in charge of the entire subject research division. It turned out, they had a graduate thesis that was entirely observing an instrument artifact (showing bad jugement). Not too long after I interacted with them, there was a minor scandal in the community because postdoc who struggled with reproducing the effect and complained about the results was abused and railroaded by the PI in charge of the project. Shortly thereafter the PM didn't seem to be a PM anymore.

It's worth viewing DARPA from their customer's perspective. What DARPA brings to the DoD in a crisis is a stable of highly competent engineers and scientists who want to push the envelope and are smart enough to not let a crisis go to waste. I'm on the demand side right now and it's been awesome to see all these crack folks swarm down on my requirements.

If you think the DoD is behind the power curve, um, well, I have seen things that don't officially exist show up, obviously already engineered, so fast you can calculate from the shipping label how long it took someone to get off the phone and get it onto a plane.

So, yes, there are absolutely the long game 5-year efforts (which are often really sprouts of 20 or 30 year efforts). But it's a bit like YC: the alumni network is amazing.

I'd like to write a comment but I generally prefer to have read the entire article before I do... see you in a week or so.

Jokes aside, I agree with the author (I think) that DARPA has been a surprisingly effective org in an age of frequent failures of other orgs with similarly lofty goals.

DARPA is effectively what things like the SoftBank Vision Fund should've been (wanted to be?). It would certainly be interesting to see what it would look like to have a privately run clone of DARPA if you injected as much cash as the Vision Fund did. Per the article, about $400M/y is spent on actual R&D by DARPA, where as Vision Fund has injected that amount into single companies many times over.

My more top level view is that DARPA and such are just side channels to sneakily funnel public money into research.

In theory free market should take care of the economy. It's really good at optimizing manufacturing processes and getting stuff as cheaply as possible into hands of as many people as possible.

However there's one thing essential to the economy that free market sucks at. It's research. And that's not because companies are bad at doing research. It's because companies are bad at funding research. Research is inherently risky and no sane capitalist will invest in anything beyond tinker level research because he will loose. And he's not supposed to loose.

In theory you could fund research overtly with country budget, but no one is going to support that. Why spend money on eggheads playing with useless stuff if people are hungry and streets are dirty?

What nobody opposes is giving more money to the military. And what anyone can't oppose is military spending money in whichever way they please. So they can spend it on research. Most will be wasted, very few will have actual military potential. The rest can be graciously dumped into the economy for companies to tinker, optimize making of, manufacture, market and sell.

IMHO military is the core of USA success (or even success of global capitalism), not through might, but through ability to get plenty of public money and ability to allocate it into things that would never get funded in any other way.

When you say that ‘DARPA’ “works”, what precisely do you mean by that, and how do you measure success? It seems to me that it primarily works by throwing large amounts of money at academics and then claiming credit for anything they invent, whether or not it contributes to whatever DARPA is trying to accomplish.

If it were possible, a closer reading of history might provide additional lessons. For instance, ARPA took a hit in the early '90s under Bush. Some silver linings, but that's a transition which might illuminate process tradeoffs.

There was (at least back then) an ongoing meta discussion about how to do better. So it might be useful to explore not just the organizational designs which were realized, but also the space of things considered. For instance, before Bush hit, there was discussion of tiny "fix that!" grants. Like there's one person who is an outlier in understanding how to do X, and their book just isn't getting finished. So rather than society waiting years on diffusion and reinvention (which is what ended up happening), it might be worth paying someone to sit outside their office, and stand on their desk, and be a forcing factor on making the book happen.

At least back then, with a failed attempt at a commerce ARPA clone, it was thought important to have a clear metric to prioritize projects. "What's better for DoD?", rather than the far less tractable "what's better for the commercial economy?".

Perhaps I missed it skimming, but a major issue has been the death valley between research and commercial impact. And attempts to address that making things even worse (ie, researchers encouraged to think commercialization, sacrifice impact by holding things close, and then commercialization generally fails, so there's no offsetting benefit). And there's the unfortunate pipeline from research to patent to unsuccessful startup to dominant company having yet more anticompetitive ammo. I wonder if it might be fruitful to broaden focus to the research pipeline rooted in ARPA? Because a successful clone would presumably again face this difficulty. And there might be some other design point that is less ARPA-like, but does impact better.

How well ARPA works comes and goes. It's not a stable equilibrium. So instead of asking how to create a successful clone, perhaps one might ask how to create something viable in the vicinity of success, and separately, how to increase time spent less distant from success?

The difference between an old-school autonomous PM, rolodex and checkbook in hand, showing up on your doorstep and saying "I've heard you interested in doing X - what would you need?", and say NSF exploratory grants of "groups with the following characteristics, may submit grants addressing the following issues, with a deadline of mumble, and the following logistics"... is really really big.

Darpa is incredible because it's run by scientists and not managers. A lot of small tech is also run like that, it's so refreshing.

There's an organization closely resembling DARPA but with exclusive focus on computing. Wish more gets written on it →

https://en.wikipedia.org/wiki/Intelligence_Advanced_Research...

It seems that this would be a much better way of funding research in a general sense, since that seems to be the general process of their problem solving. The only thing is that to apply this to other fields, the barrier for getting people that are equally motivated and intelligent about analyzing if the research is 'going' somewhere is high

I would say the key difference between DARPA funding and other research functing is that DARPA PMs have a crystal clear map of the leading edge in a field that's ready for advancement and can target it perfectly with both money and real competition.

I think it comes down to the fact that the military has a lot of money and the projects have goals that are somewhat directed but also long term. Don't think it's more complex than that really.

it works because there are no google bros