While people may think that this is wasting money, I see that this these teams gain unieque knowledge and experience on how to resolve issues on "machines far far away", and if we are ever to explore/move humans to a new planet, you want THOSE exact teams as 'helpdesk' that use anything and everything to bring things back to life and operation.
This is assuming generational knowledge can be sufficently passed down to be useful in future missions. I'd assume this is quite a niche/specialized skillset they developed for this particular machine. The utility of the knowledge has a limited timeframe which may not justify education of the next generation of scientists.
But regardless recording it would likely be helpful as most problems in engineerings seem to recur over time in different and new contexts.
We build on the knowledge of those before us. The people who invented the original programming languages are mostly gone and most of us probably don't know how to use the stuff created but future works were based off the findings of those before them and we all benefit from that.
Hubble has been around enough now that the next generation of stellar and space engineers are getting though university and into NASA and other space ventures.
Kids need to see the previous generation has set extraordinary goals to inspire them to set their own goals. What could be more inspiring than a middle school student seeing Hubble launch, being inspired to study and then after college get to participate in what inspired them in the first place.
I think its reasonable to assume AI will be able to fix these kinds of problems in the future. They could run simulations of all the possible failures and solutions ahead of time or during operation eventually
I think the creativity and imagination required to solve this kind of problems is not really within current AI reach yet. The ability to take existing ideas and combine them in unique ways to come up with something new is very powerful, and not yet easily replicable by AI.
Genetic Algorithms are pretty creative. I think a bigger problem is that we have no general way to turn “what we want” into a mathematically formalised goal that an AI can work towards — we can only manage specific cases like “how many components does this signal generator circuit use” or “make a new face based on these examples of faces”.
The search space is so huge and the amount of training data we have is so little that AI, as we currently understand will be useless for this kind of stuff.
