The rise of ‘ARPA-everything’ and what it means for science (2021) | Hacker News Reader

The rise of ‘ARPA-everything’ and what it means for science (2021)

nature.com

53 pointstwo_almonds21 comments

As someone with ARPA-related experience, I can say there are some trade-offs (possibly even flaws) in the execution of "the recipe" as practiced in the modern day. However, it's probably also the part of government that most closely resembles the start-up world. In fact, when I was there, we often intentionally modeled certain things after ideas borrowed from that side of the private sector. If you believe in maxims like "being okay with failure", "failing fast", and "going for moonshots", then you might think it's worth the cost (and all the untrimmed fat).

That said, I'd probably like to see high-level visibility on some accounting of cost versus success in, say, the past 15 years. From what I've seen, there haven't really been the kind of big-impact wins you usually see touted from further back in history. Instead it's mostly been PR, shrug-worthy output, and outright graft (e.g., a golden parachute for a certain Solyndra exec).

Agreed. Worked on an ARPA-E funded project and it was reported as a success even though the primary deliverable was never completed and was probably not viable anyway. Everyone involved meant well, but I sensed the PIs viewed it more as a vehicle for their ICs' careers than as a mission-driven enterprise.

This is a wicked incentive problem and I don't know how to solve it.

For those not familiar with the jargon, can you expand what PI and IC means? Private Investors and Individual Contributors?

sorry, Principal Investigator and Individual Contributor

But, did the project significantly advance the knowledge in the field, even if the deliverable couldn't be completed?

In that case, I would call it a success anyway.

It produced some modest results which slightly advanced the field. Not groundbreaking. The applied work did demonstrate the viability of a few robotics applications and the infeasibility/insufficient ROI of a few others. I would say that was the most valuable output.

It did not begin to approach the goals in the project description, which I've since come to believe are detached from reality. A moonshot may have seemed collossally ambitious in 1961 but we already had most of the basic capabilities necessary to succeed. In this case, however, I don't think we're anywhere close, and even if we were, it would be a suboptimal solution to the problem it's meant to solve. (It would however be politically and economically convenient).

Curious to what degree the "low-hanging fruit" dynamic comes into play when comparing the current era to the past.

It keeps coming up in many other fields.

For those less informed about its history, is there any transparent accounting of ARPA's cost-vs-success during its initial decades that you recommend reading? I am only marginally familiar with some of the 'big-impact wins' that are frequently mentioned, so I would love to read a comprehensive analysis of its accomplishments.

I’ve absolutely loved working with ARPA-E in my job and I cannot imagine why we would create ARPA-C. Anyone in ARPA-E will tell you they are fundamentally working on climate. They also do some energy security stuff but I they would all be very dissatisfied if it got paired down to that.

ARPA-H sounds like a great idea, I hope it’s well executed!

Well climate != clean energy.

It would be great if there were people looking at climate change in truly innovative ways. But if it's a lot of the same-old, wrapped in a new ribbon it's going to be a disappointment.

And honestly there is a huge disconnect between the dire warnings we're getting and the actions we're taking. If it truly is an existential crisis, better wind turbines and solar panels that are 2x more efficient are great, but they're not gonna move the needle, and the real work should be a bout surviving in a hotter world.

I think ARPA-E is actually already working on climate apart from energy. I have seen their researchers talking about everything from low-land-usage/climate-resilient agriculture methods to carbon capture.

A cynics take awaits you, however, if you consider scientific research an organism of sorts then this makes complete sense. since most legislated spending at the congressional and senate level is only ever expediently approved if it is in the interest of the national defense, then the simplest way to encourage and continue meaningful advances in the body scientific is to simply blanket them all as "defense" projects.

so, your climate analysis grant may face grim prospects until the cosmic DARPA modifier is applied, and in doing so, becomes easily funded.

> if you consider scientific research an organism of sorts

Somewhat off-topic, but at a macro level, society itself and the systems it produces are fundamentally the same thing as the individual cells that make up our bodies (the specifics of what those systems act on and the context in which they act in are obviously different). Basically every societal structure can be reduced to the behaviors of an organic system at some level of complexity.

This article is about the creation of new ARPAs without the D, which are explicitly not about defense. They are rather are meant to take DARPA’s model and apply it to other areas (climate, energy, health). Organizationally speaking, they’re unrelated, being under different federal agencies depending on the area. They may borrow some of DARPA’s glamor but they aren’t considered defense projects. So your comment doesn’t seem applicable.

I'm a bit surprised there hasn't been much of a pitch for an ARPA-AI, maybe because AI touches on a bit of everything or too many other stakeholders in gov are doing AI stuff? But AI seems like an area where an organization just focused on that would be rather interesting and it would likely have broad ramifications for the other ARPA's being considered (e.g. advancements in medical with protein folding and drug discovery, advancements in energy with automated grid management and figuring out optimal mixes for minerals for batteries)

i think that's what they have DARPA for.

Actually Norbert Wiener thought a lot about intelligent machines, he called that Cybernetics - but there was a problem, he didn't want to accept any funding from the military:

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

"After the war, Wiener became increasingly concerned with what he believed was political interference with scientific research, and the militarization of science. His article "A Scientist Rebels" from the January 1947 issue of The Atlantic Monthly[19] urged scientists to consider the ethical implications of their work. After the war, he refused to accept any government funding or to work on military projects. The way Wiener's beliefs concerning nuclear weapons and the Cold War contrasted with those of von Neumann is the major theme of the book John Von Neumann and Norbert Wiener.[20]"

John McCarthy invented the term Artificial Intelligence, in order to avoid any association with Cybernetics - I think he didn't want any association with Norbert Wiener. I suspect that this might have something to do with funding...

see http://jmc.stanford.edu/artificial-intelligence/reviews/bloo...

Sure but DARPA does all sorts of work, not just AI related stuff

certainly, TCP/IP was a DARPA project, lots of other things were also financed by DARPA. https://www.darpa.mil/about-us/timeline/tcp-ip

TCP/IP was a major change in networking, because all telephone companies were used to have reliability at the link layer - that's how you do a telephone network. But TCP/IP is much more scalable, because it has a reliability layer that is not dependent on the link layer.

I don't know why, but I always thought it was cool that the Lambda the Ultimate Papers were sponsored by ARPA. I suppose to some extent because they're basic research not specifically applicable to any defense project, and yet like a fair amount of CS/PL research, they're very practical as well.

As someone with passing familiarity with the U.K. political landscape, I can confidently predict that ARIA will not amount to much, and that the exact same politicians constantly saying government should be more entrepreneurial will apply their usual mixture of ridiculous oversight, neglect and punishing failure that produces the current outcomes and cycle of self-fulfilling prophecy will continue.

It's strange that the article ignores the Mansfield Amendment and the transformation of ARPA to DARPA.

Reply on news.ycombinator.com