> There are tons of data from the Voyager trips collected 31 years ago that are still unanalyzed, and getting funding to examine them is very, very difficult.
the PIs i dealt with on space missions were usually on to the next piece of hardware they were going to fling into space before they had the data from the last one.
and i think the lab i was at had some voyager (or pioneer?) data laying around that one of the emeritus scientists was talking about getting transferred off of tape so he could finally look at it.
Unsupervised learning is a thing. It could tell us "these data points are similar", we just still need a human to try and quantify why it thinks they are similar.
It seems you have a lot of faith in AI, but don't seem to really grasp what AI is/does/can do.
In your link, you neglected the part where NASA first had to learn what rocks made good samples before they could "teach" their AI to look for similarly interesting samples.
In this case, you have a lot of data strewn about many different mediums (some now obsolete). Yes there's images, but what would the AI look for? In the other data collected, what would it analyze... what would be the goal for it to achieve? This is why we still have human scientists... and yes, AI can help, but it cannot replace.
As another HN'er said, we're a long way away from "AI, discover new science!".
I am a Data Scientist with Master's in Neural Networks so I do understand the challenges in preparing data for modeling. I am merely suggesting the option of exploring the problem of finding patterns in data through ML techniques. This does not take away the work involved in preparing clean data sets for training. Here are some Astronomers who are doing this kind of work. https://www.wired.com/2017/03/astronomers-deploy-ai-unravel-... I agree I don't understand the nature of data NASA has but it does not hurt getting some help from companies that are searching for problems to solve.
> I agree I don't understand the nature of data NASA
It's not the nature of the data that poses an issue... it's the lack of direction. AI/ML/NN (whatever buzz word of choice) simply cannot do the things you seem to believe it capable of.
> it does not hurt getting some help from companies that are searching for problems to solve.
What makes you think Google or Microsoft have any interest in working on NASA's research? These are very, very different domains.
Even if they were interested, what problems would they be solving exactly? This is the lack of direction that is the core problem with you assertion AI/ML/NN would be of any help here on it's own.
To add to the above, it's the difference between telling a model:
"Model, show me spectral signatures which are similar."
And "Model, is there anything interesting about the spectral signatures collected?"
In the former, you are potentially able to (or your NN layers can) identify representative features for your intended result.
In the latter, you're asking machine learning to synthesize the sum of human scientific knowledge, then extract interesting facts from the data using it. E.g. "Hmm, the vortex patterns on this Jupiter storm are incongruous with our fluid dynamics models" or "The reflectivity of the surface seems to indicate a different composition than we expected."
AI isn't currently capable of formulating questions, asking them, answering them, and ranking the results on significance.
And in lieu of that, the only use is "Help a scientist answer a question they posed." Which seems to be the initial problem that started this thread! Not enough scientists can get grants to pose all the questions that should be asked of old data.
for what it's worth, they were doing a much better job of getting data from their more recent missions (maybe 2000 on?) up on websites.
the data from these research instruments is not easily analyzed using public information, though. overviews of how the instruments are _supposed_ to work are published in a book on the entire mission (for instance, for SDO: https://link.springer.com/book/10.1007/978-1-4614-3673-7 ) but as far as i know the weirdness instruments develop over their life isn't written down anywhere public.
and the data processing code for these things is sufficiently complicated that even with all of the documentation and access to the PIs and hardware engineers, it takes at least a few engineer*years to get working.
Here's one thing that has bugged me for a while. Why is it said that the great red spot is around 500 years old? How do we know that it hasn't been around for much longer?
Do we have before/after evidence that at one point it didn't exist, and then around 500 years ago it existed?
To extend, a series of disjoint astronomical observations of the Earth would likely show a hurricane/cyclone/typhoon somewhere, but it wouldn't necessarily follow that they were all the same one.
22 comments
[ 4.6 ms ] story [ 62.1 ms ] threadthe PIs i dealt with on space missions were usually on to the next piece of hardware they were going to fling into space before they had the data from the last one.
and i think the lab i was at had some voyager (or pioneer?) data laying around that one of the emeritus scientists was talking about getting transferred off of tape so he could finally look at it.
AI isn't just some magic that "figures out" anything and everything.
In your link, you neglected the part where NASA first had to learn what rocks made good samples before they could "teach" their AI to look for similarly interesting samples.
In this case, you have a lot of data strewn about many different mediums (some now obsolete). Yes there's images, but what would the AI look for? In the other data collected, what would it analyze... what would be the goal for it to achieve? This is why we still have human scientists... and yes, AI can help, but it cannot replace.
As another HN'er said, we're a long way away from "AI, discover new science!".
It's not the nature of the data that poses an issue... it's the lack of direction. AI/ML/NN (whatever buzz word of choice) simply cannot do the things you seem to believe it capable of.
> it does not hurt getting some help from companies that are searching for problems to solve.
What makes you think Google or Microsoft have any interest in working on NASA's research? These are very, very different domains.
Even if they were interested, what problems would they be solving exactly? This is the lack of direction that is the core problem with you assertion AI/ML/NN would be of any help here on it's own.
"Model, show me spectral signatures which are similar."
And "Model, is there anything interesting about the spectral signatures collected?"
In the former, you are potentially able to (or your NN layers can) identify representative features for your intended result.
In the latter, you're asking machine learning to synthesize the sum of human scientific knowledge, then extract interesting facts from the data using it. E.g. "Hmm, the vortex patterns on this Jupiter storm are incongruous with our fluid dynamics models" or "The reflectivity of the surface seems to indicate a different composition than we expected."
AI isn't currently capable of formulating questions, asking them, answering them, and ranking the results on significance.
And in lieu of that, the only use is "Help a scientist answer a question they posed." Which seems to be the initial problem that started this thread! Not enough scientists can get grants to pose all the questions that should be asked of old data.
the data from these research instruments is not easily analyzed using public information, though. overviews of how the instruments are _supposed_ to work are published in a book on the entire mission (for instance, for SDO: https://link.springer.com/book/10.1007/978-1-4614-3673-7 ) but as far as i know the weirdness instruments develop over their life isn't written down anywhere public.
and the data processing code for these things is sufficiently complicated that even with all of the documentation and access to the PIs and hardware engineers, it takes at least a few engineer*years to get working.
Edit: turns out you can download at least some of them from archive.org! https://archive.org/details/VoyagerstotheOuterPlanetsVol5
Do we have before/after evidence that at one point it didn't exist, and then around 500 years ago it existed?
While there were earlier observations of large spots on Jupiter, it's not 100% certain they were actually the same storm.
https://en.wikipedia.org/wiki/Great_Red_Spot
To extend, a series of disjoint astronomical observations of the Earth would likely show a hurricane/cyclone/typhoon somewhere, but it wouldn't necessarily follow that they were all the same one.