I was pleased to speak with Will Thomas on the Acquisition Talk podcast. He is a senior science policy analyst at the American Institute of Physics, and is a historian of science and technology. His book is Rational Action: The Sciences of Policy in Britain and America, 1940-1960. There is a ton of interesting facts and useful analysis in the history of how the military learned how to learn, with clear application for today’s debates on innovation.
During the discussion we touch on a wide-range of issues, including the origins of operations research, whether a market in defense can actually exist, the separation of R&D from production — and whether software considerations have changed the logic, the uses of technology readiness levels, similarities and differences between healthcare and defense, and the experience of Donald Trump’s uncle, John, who was head of the British Branch of MIT’s radiation lab during WWII.
The talk features an analysis of the debate in RAND between systems analysts like ES Quade and luminary economists like Kenneth Arrow and Armen Alchian, who favored a sequential decision-making in R&D due to the prevalence of uncertainty. I tried to pick apart some distinctions between Arrow and Alchian, characterizing the former as more of an optimizer using an allocation paradigm and the latter as more evolutionary using an exchange paradigm. Will responds that I was over-interpreting Arrow, and that the goal of both was to advocate for policies of government support to exploratory development without locking in technical configurations prematurely.
Podcast annotations.
One of the interesting things is that in both Germany and the UK and the US, you had a lot of very novel projects that got started during the war. So you had German rocketry, jet engines, and that sort of thing is well known, in the US of course there was the atomic bomb project, and in Britain they had the bouncing bomb. These were technological novelties, and the people who were really concerned about coordinating the efforts of the research laboratories with military needs… regarded them as not being responsive to the actual needs of the war…
During World War II, I talked about bombs attached to parachutes, another one was Project Habakkuk… It was to be an aircraft carrier made out of ice, or a specific kind of ice called pykrete named after its inventor Dr. Geoffrey Pike, and they actually did a lot of R&D on this up in a lake in Canada before deciding it wouldn’t work out. So they were really exploring these very horizon type ideas in an environment where things were supposed to be directed toward the immediate needs of the war.
It seems important for the government to tolerate funding projects which look speculative, even in the most dire of times. It’s not entirely clear to me when a prospective project looks more like science than engineering. For example, I wouldn’t blame most decision-makers in 1942 for looking at the Manhattan Project as more science than engineering. The way Werner Heisenberg executed the atomic project for Germany certainly looked like pure science. The Americans went after it as a major industrial engineering project.
I think it’s pretty hard to contain scientific breakthroughs, or develop them independently of the wider academic community. Scientific results tend to get out fairly quickly, such as Otto Hahn’s results on fission. But there’s plenty of room for doing much better engineering with the science we have, and exploring that space will provide new questions to feed back into science, which ultimately government will have to support whether by the DOD or otherwise.
Ultimately I don’t know whether the marginal dollar should go toward research or development. However, for the applied development funding that is available in those accounts, more should go towards austere prototypes that are open-ended in the same way that we tolerate for basic research.
Here is Will introducing us to how some of the early economists who worked at RAND thought about problems of weapons choice after the war:
UCLA economist Armen Alchian and Stanford University economist Kenneth Arrow had a series of memoranda in which they said ‘we need to support development projects,’ this is somewhere between basic research and an actual acquisition program, and ‘we need to support a wide variety of different projects, and as we gather information over the course of these different projects, only then can we decide which ones we’re going to kill.’ They didn’t create a formula or anything for deciding when you kill a project or anything like that, but what they were saying is that they government has a huge role to play in supporting very speculative R&D projects that you didn’t think were going to work out…
What both of them were really arguing against was premature commitment, so you know as well as I that Alchian’s concern was that systems analysts were doing mathematical design exercises that would narrow in on, if not an optimal, then a preferred configuration for a particular weapon system. He was worried that military contractors were starting to use these systems analyses to try to lock the military into an acquisition program before it was really wise to do so.
Will really nicely explains the controversy in the 1950s about systems analysis and program choice. Listen to the podcast for more, including the story of the RAND bomber study which generated introspection and eventual change for the original systems analysts.
Here was a part I really liked:
When you are trained to think like a modern day operations researcher you can see the analogies between a spectrograph that aims 2,400 fibers at different stars [and has to decide when each fiber has collected enough data to move on] and an R&D director who is funding 200 different projects and has to make decisions about when to cut-off a particular one — except for the fact that stars don’t talk back, stars don’t make it look like their right around the corner from having a workable prototype, so that does introduce additional complications…
I asked Will about the question of centralization of defense decision-making, and whether that monopsony power on the buying side created efficiencies:
You’ve really hit upon one of the central themes in the history of economic thought. There’s no coincidence that it reminds me of the socialist calculation controversy in the 1930s, the idea that if you’re going to have a centralized economy then you’re going to have to know everything there is to know about that economy in order to distribute resources within it in anything like an efficient way.
Then you have people, most famously Hayek, who say ‘well, it’s simply impossible, you actually have to have things be decentralized in order for efficiency to be possible at all.’ And again going back to Arrow’s greatest hits, this is very much the model of Arrow’s general equilibrium proof, a decentralized representation of an economy that can nevertheless work its way toward some kind of optimized equilibrium.
I would push back a bit to say that Arrow’s general equilibrium proof seemed to bolster the view of socialist planners because even though the buyers and sellers in the model were decentralized, for it to work there had to be perfect and symmetrical knowledge about all factors of importance including of homogenous products, input prices, production methods, etc. When the (absurd) assumptions were not upheld, this view of markets led to a “market failure” mentality.
But the socialist calculation debate is very much in the same vein as the debates over weapons choice in an organization as large and diverse as the Department of Defense. There enters the familiar problem that knowledge is dispersed across participants, who hold separate and even contradictory pieces of information. The major unresolved problem for defense is how those dispersed units are best able to coordinate. As Will pointed out, the competitive developments of the 1950s had their drawbacks:
If we look at the situation of defense in the 1950s, you have the services and they each wanted as much territory as they could control vis-a-vis the other services, and so you had weapon systems that each one was working to justify… and you don’t have any central means for adjudicating claims, you end up with what Eisenhower complained about which is the Military-Industrial Complex, which grows and grows because you don’t have any means of resolving between — you in a sense invest in everything because you don’t have a means of saying ‘no’ to anybody.
I’d like to thank Will for joining me on the Acquisition Talk podcast. Be sure to check out his website which includes links to his book Rational Action and nearly a dozen fascinating articles. Read his paper on Donald Trump’s uncle, “A profile of John Trump, Donald’s oft-mentioned scientist uncle.” His Twitter handle is @GWilliamThomas and he occasionally blogs at EtherWave. Will also recommends reading David Edgerton, among others, on the history of science and technology.
Leave a Reply