Learning and Sharing
I spent 2.5 hours on the stationary bike. At first I read an article by James Somers, I Should Have Loved Biology, which was sourced from an incredible list of biology essays by Asimov Press. The thesis ("I should have loved biology but I found it to be a lifeless recitation of names") of the article resonates with me. I love finding new gold nuggets like this on the internet. That page, in turn, included a few links to other great pages, so I already know where to look next. Articles like his also remind me – perhaps falsely – of a past, when the internet was very different, weirder, more personal. Computer scientists are overrepresented among people who have a personal website, which makes sense. In many cases, the sites are simple by today's standards – sometimes raw HTML and CSS – but you can still see they were made with love, or at least, with craftsmanship. They are not meant to distract, they are not commercial, and they will never have advertisements. Instead, they are clear, minimal, precise, personal, informative, and unpretentious.
I look at James Somers' website and can only wonder: Who is this guy? Who has written and published so much here? Certainly not like anybody I know, which is probably to my detriment. I would like to know him – the person who put so much of his thoughts out here for everyone to see and gain from.
Yesterday I had read a similar piece: Maps of Matter by Michael Nielsen, where he takes us through an exploration of thought. How cool to follow the inner workings of someone as intelligent as him. I also read Think in Math. Write in Code by Justin Meiners.
Space
Later, still on the bike, I read this article on inflatable space stations, which is about the return of crewed inflatable space stations. This is a technology which, as the author notes, we could have had by now; we were on a good trajectory until the Moon Program broke the budget.
I asked Chat and Gemini lots of questions (mostly in parallel) to sharpen my understanding. It is incredibly annoying that there is still no option (at least on ChatGPT) to copy all the chat’s text to share it with new chats, or with another model provider. Anyway, 1 cm^3 of air on Earth has around 10^19 particles, while 1 cm^3 in interplanetary space has around 10^1 particles! Truly incredible to think of. It also means that there is nothing to inflate anything with. Then how do you inflate the inflatable space stations? You bring the gas up there from down here. And you choose a material that – once it is inflated – hardens, so the system is failure-proof in case of a leak or something like that.
One question I considered was about the testing process for deployments in space. We will build ever more in space, and space is a very hostile environment for deployment. So how will the testing processes develop to accommodate a growing industry and demand? Ideally, you want to do many test iterations before the final deployment. Right now, this is obviously difficult because it is limited by the distance to space and the frequency and cost of launches. If I were to develop, say, a robotic arm that has to do some work in space, I would ideally like to go two blocks further to a facility where we can test it under “real space conditions” here on Earth. How will this play out and who is building solutions for this?
I chatted with Gemini and Chat about this issue. I now understand: Full-out, large, physical space test facilities are not possible in Earth’s gravity field. Instead, there are and will be levels of testing. Firstly, there are digital simulations with millions of iterations. Then there are physical on-earth test facilities: Thermal Vacuum Chambers like the one by IABG in nearby Ottobrunn. After that, we will test in space. Once there is a small but sufficient infrastructure/cluster of activity, this may even be a service that could be in demand.
Another interesting point: What will the ISO-container for space be? This is yet undecided, but once traffic increases it will become more urgent. Another possible opportunity, brought up by Gemini: Solar Sintering. Even as launch costs decrease, for some threshold, it will always be more cost-efficient to produce low-tech materials in space. Consider materials like concrete. Solar Sintering enables you to turn the dust on the Moon into robust ground material just by focusing the sun’s light on it. This is more efficient than other methods for the same purpose, which first capture the light and then turn it into a laser. It is important to harden the lunar surface because the dust particles can travel very fast, especially at launches and landings, and hurt other objects in the vicinity. By hardening the ground, this can be avoided.
With my Mechanistic Interpretability background in mind, Gemini also suggested the following: Train a Neural Network on robotics data on Earth, try to isolate the feature that corresponds to “gravity” (the difference between Earth and space), and test whether subtracting this feature makes the robots able to operate (better) in space. I am not an expert in space, so I can't really judge how serious to take any of these ideas,but the questions were fun to ponder. This is a dilemma, because I want to trust the models, but I am not sure I can, because I don't know enough about the topics.