I was initially confused and not sure why you were adding lots of extra parts to the analogy which complicate the process of analogizing by overloading the human prompt window, but then I realized that the analogy was specifically trying to get across the idea of there being loads of specific engineering things to work out as well as general principles (which seems true and useful, but should be captured early on in a tl;dr with the blog post mostly for people who want a deep dive into rocketry history).
I think this was a fair attempt, but missed the costs to easy understanding of loading up large amounts of context (names and dates and numbers and loads of specifics) without making it clear what these were for.
My reading of this post ended up with less understanding of what the analogy the author was trying to use than at the start.
It's fairly common and useful in forming or extending analogies to point out which parts in one domain are similar to parts in the other domain, and why you think that similar consequences would follow. The original rocket alignment post was bad enough in that respect, but at least some correspondences could be inferred and the conclusion deduced (whether agreed with or not).
I have absolutely no idea what parts in this extension post correspond to what in the actual alignment problem, nor why there should be any expectation of similar models, nor what conclusions the analogy is even trying to point toward.
What do the components of the rocket represent in this analogy? Or is it more of a general argument for the need to develop the components before being able to credibly speculate on rocket building itself?
This work was done while at Conjecture.
This post has been written for the first Refine blog post day, at the end of the week of readings, discussions, and exercises about epistemology for doing good conceptual research.
Thanks for comments by Linda Linsefors, Paul Bricman, and Adam Shimi.
The Rocket Alignment Problem presents a fun scenario trying to explain why understanding part of the theoretical basis for a problem might be useful. If we were trying to get to the moon for the first time, understanding Newtonian mechanics would indeed be quite useful! As would astrodynamics in general, which was really developed starting with Herrick in the 1930s. But not sufficient: there were a whole host of other problems that needed to be solved, many of which were at least as difficult as theoretical understanding. At a minimum, we needed to develop rocket fuel, the materials that could survive high temperatures from fuel and reentry, enough understanding of the materials to create models of the rocket themselves and the stress and heat individual parts were under, communication devices so we knew what our rockets were doing, and more. I’ll look at the first two in more detail here then step back and look at the problem as a whole at different points in time. Beyond simply pointing out the rocket alignment problem to show how messy it truly was, pushing deeper into the analogy may let us see what we have developed, and what we expect we need to develop.
Developing rocket fuel powerful enough to lift the rocket but stable enough that it didn’t just immediately explode required quite a bit of chemistry, engineering, and ridiculously dangerous trials. Consideration of this problem directly applied to rocketry, building off of the chemical achievements of the previous century, started in the early 1900s and continued through the development of the rocket program.
A Russian school teacher, Tsiolkovsky, first proposed using liquid fuels, such as liquid hydrogen, paired with liquid oxygen for the oxidizer, recognizing that standard fuels or gunpowder didn’t provide enough energy. He was mostly not noticed, and Goddard’s work on building liquid fuel powered rockets was mostly ignored as well. In the late 20s and 30s, more mainstream efforts pick up across Europe, using both new liquid fuels and oxidizers such as nitrogen tetrooxide or tetranitromethane,, the latter which had a tendency to blow up and take off a few fingers with it. More unstable still were monopropellants, which contained both fuel and the oxidizer in the same molecule and would react with themselves with the right catalyst or temperature, but if they have enough energy to compete with standard propellants, had the unnerving tendency to randomly blow up, which led to them mostly being abandoned. These humble beginnings led to a long and generally productive investigation into fuels that were powerful enough to propel a rocket to space, but not unstable enough to propel various bits of the rocket across the launch pad. This whole process required both theoretical knowledge of chemistry to help guide the search process, and tinkering with fuels and oxidizers to see what worked, with various tangents such as looking at monopropellants not really panning out.
Creating a rocket that could survive this was equally difficult. The range of temperatures that specific parts were expected to be under was extreme, from -250F (-155 C) to 3,000F (1,650C). They had other requirements they needed to meet as well: “It was necessary to design thousands of these tiles that had compound curves, interfaced with thermal barriers and hatches, and had penetrations for instrumentation and structural access.” Developing materials that were lightweight but capable of withstanding these conditions was a hard problem, requiring the development of new materials such as reinforced carbon-carbon, as well as stress testing both individual components and the whole wing under somewhat realistic conditions.
Taking a step back, these problems were downstream of much of the development of science and technology since the start of the industrial revolution. From Newtonian mechanics to advances in understanding orbital mechanics, from advances in chemistry and material science to the development of precision machinery. In the 1930, much of the base was in place, but anything specific to rocketry was in its infancy. We knew about Newtonian mechanics, but nobody had worked out specifically how to get a rocket to the moon. We had some idea of the types of chemicals that would be useful in rockets, but had not worked out what would give the best mix of stability and energy, and had not yet developed hypergolic fuels. Much of the basis in material science was in place, such as understanding strong, light metals like aluminum and titanium, but not the specific technologies that would heat shield the rockets.
At the turn of the century, much of the technological and scientific basis was in place, but not all of it: our understanding of high volatility chemistry was still in its infancy, with the basis of liquid fuels such as liquid oxygen or hydrogen having been developed shortly before they were proposed as rocket fuels. Titanium was known to exist, but there didn’t exist any process for isolating it, the ability to create large amounts of titanium outside of the laboratory was developed in the 1930s. Mass production of aluminum was discovered a decade before, but the reinforced carbon-carbon that coats the most exposed points of the space shuttle was developed for the space program.
Going back further to 1800, perhaps we’d have a vague idea that, if we had to go to the moon, an improved understanding of orbital mechanics, of chemistry, of material science, would be useful. I don’t think our ideas would be all that precise; yelling at people who are working on trying to isolate and create frozen chemicals that they weren’t directly working on getting to the moon would have been a mistake, as would stopping all work of theoretical considerations of astrodynamics.
Leaving the industrial and scientific revolutions behind, if we are anywhere in the world in 1400 and must reach the moon, things are even more uncertain. Perhaps we could look at the nascent development of cannons, but that will not get you to the moon, we are missing so much. The mathematics of 1400 are inadequate, as are Aristotelian physics, chemistry, material sciences, precision engineering; roughly the entire techno-scientific apparatus is missing. The development of science and technology before the scientific and industrial revolutions is too unsystematic, too fragmentary to be able to solve this problem. We need to invent the systematic search for science and technology first.
Just how this maps to alignment depends on what we are missing, and what we already downstream of. We have some of the basis, but are missing so much. I think similar to rocketry, trying to solve theoretical alignment problems are reasonable, so too are engineering problems and technoscience-y questions surrounding neural networks that seem potentially upstream of an alignment solution. And to the extent our science isn’t capable of handling the cluster of problems we face in alignment, we need to create a more powerful science that can handle it.
Sources:
Random stuff from memory
For fuel, the wonderfully entertaining Ignition!
NASA has stuff on what material sciences were necessary: both summaries and initial documents.
For broader questions on material science I’ve read parts of Out of the Crystal Maze and Understanding Materials Science: History, Properties, Applications.