Author Topic: How Can AI Be Used for Space Applications?  (Read 74721 times)

Offline InterestedEngineer

  • Senior Member
  • *****
  • Posts: 2493
  • Seattle
  • Liked: 1887
  • Likes Given: 3101
Re: How Can AI Be Used for Space Applications?
« Reply #20 on: 11/15/2022 10:11 pm »
Isn't this exactly the kind of stuff a genetic evolutionary approach could tackle? To try to synthesize a viable biochemical paths working in the Martian environment?
Btw, there've been things like genetically evolved antennas done by NASA:
https://en.wikipedia.org/wiki/Evolved_antenna
(As a side note - as you've been in this field for a long time - do genetic evolution algos belong usually under the AI moniker?)
Yes, evolutionary algorithms are definitely part of "AI". And, when you can do a generation every six seconds or so (give or take), you at least have a chance of getting a useful result from them. The question, though, is whether there's actually a simpler, cheaper way to get that same result (or better). I'm not familiar with the evolved antenna, but I suspect that some other, less fancy search algorithm would work at least as well with less complexity. In other words, evolutionary algorithms always appear to be a solution in search of a problem, and the places that actually use them seem to have viewed using that algorithm as a goal "Yes! We used evolutionary algorithms!" rather than trying to find the best solution to the underlying problem.

I should confess here that part of my negativity comes from a particular group I had the misfortune of interacting with when I worked at Amazon. They kept coming up with "solutions" for hard problems in different areas, but their software almost never worked very well--it was generally way too slow and nowhere near as accurate as they claimed it to be. However, they had friends in high places, and they consistently tried to use politics to force people to adopt their software when they couldn't win on the merits. Obviously, it's not fair to blame everyone working on EA for what one group of people did.

However, the gripes against evolutionary algorithms are common in the field. Go to just about any ML conference and ask some of the experts their opinions on them. I'm far from the only person who's down on them.

As for applying them to biological evolution on Mars, the fact that the algorithms are inspired by evolution doesn't mean they're really all that well suited to solving problems involving the real thing. In this case, the challenge is that a) I don't think we can engineer new organisms all that accurately in the first place and b) I suspect we can't really model the Martian environment all that well either. We might get there someday, of course, but even then I suspect there will be better approaches.


A good 19th century dog breeder would get there faster than biologically illiterate AI programmers

Offline leovinus

  • Full Member
  • ****
  • Posts: 1158
  • Porto, Portugal
  • Liked: 909
  • Likes Given: 1797
Re: How Can AI Be Used for Space Applications?
« Reply #21 on: 11/15/2022 10:23 pm »
Isn't this exactly the kind of stuff a genetic evolutionary approach could tackle? To try to synthesize a viable biochemical paths working in the Martian environment?
Btw, there've been things like genetically evolved antennas done by NASA:
https://en.wikipedia.org/wiki/Evolved_antenna
(As a side note - as you've been in this field for a long time - do genetic evolution algos belong usually under the AI moniker?)
Yes, evolutionary algorithms are definitely part of "AI". And, when you can do a generation every six seconds or so (give or take), you at least have a chance of getting a useful result from them. The question, though, is whether there's actually a simpler, cheaper way to get that same result (or better).

[snip]

Thank you for providing a segway into my favorite throw-back search algorithm "Kangaroo search" :)

First some context. Of course I agree with Greg that Genetic Algorithms are part of AI. I did some work with them in 90s but discarded it because there were better solutions. It is an example of a search algorithm for global optimization, like the less efficient, slower random search and more effective simulated annealing. Even gradient descent as used in neural network optimization is just one "search" approach  to arrive at the "best" solution. And as mentioned earlier, you go to these search algorithms for a global optimization when you do not have an analytic one-shot solution. That it is the important take away - when you have an AI/ML model with non-linear layers and thingies and you >cannot< optimize analytically then (and only then) you grab for global optimization and search approaches.

Somewhere in the early 90s, during the genetic search period, we had a few "beer heavy" discussions when we saw Kangaroo search in a newsgroup. Was is sci.physics or something? Not sure. In a nutshell, there are various ways to find a best solution for your models and dropping kangaroos on Earth to find the highest point is one example solution to one optimization problem. There is a readable discussion here at http://processcontrol4dummies.blogspot.com/2013/06/optimization-technique-kangaroo-analogy.html

Enjoy.
« Last Edit: 11/15/2022 10:27 pm by leovinus »

Offline leovinus

  • Full Member
  • ****
  • Posts: 1158
  • Porto, Portugal
  • Liked: 909
  • Likes Given: 1797
Re: How Can AI Be Used for Space Applications?
« Reply #22 on: 11/15/2022 10:25 pm »
"If it's looking for debris of a crashed space probe on the surface of the Moon, then it would have had to train by looking at enough debris fields. "

Not necessarily.  I many cases you compare your new image with a pre-impact image.  Example: the Chang'e 1 orbiter impact... not present in an Apollo 16 panoramic camera image but present in LRO images.  Almost any future impact site will be findable that way.  Older impacts like (for instance) the Apollo 16 LM ascent stage are much trickier.  But if you think AI can do it better than a human, think again.
In other words, looking for a "change" or "deviation" in a static landscape is almost a binary classification problem i.e. easier than of collection millions of "debris" pictures for fancy classifier. As always, problem formulation is key. Plus a well chosen algorithmic solution.
« Last Edit: 11/15/2022 10:26 pm by leovinus »

Offline leovinus

  • Full Member
  • ****
  • Posts: 1158
  • Porto, Portugal
  • Liked: 909
  • Likes Given: 1797
Re: How Can AI Be Used for Space Applications?
« Reply #23 on: 11/15/2022 10:29 pm »
Autonomous onboard navigation...
...
Adding an "Autonomous" level is additional level of complexity beyond that. Hence, first get the basics right and then we can talk about AI additions ;)
Well, the Mars rovers do have some autonomy. They do absolutely need it - you can't teleoperate them directly from Earth.  I absolutely expect Tesla's FSD-derived autonomy on Mars/Moon one day.
Ah, for autonomous driving on Mars, I would agree a Tesla-like, self-driving, AI/ML derived solution is appropriate. For autonomous avionics (on SLS et al), which is the way I interpreted the original post, it might not be (yet). Which is why I commented.

Offline Greg Hullender

  • Full Member
  • ****
  • Posts: 624
  • Seattle
    • Rocket Stack Rank
  • Liked: 452
  • Likes Given: 349
Re: How Can AI Be Used for Space Applications?
« Reply #24 on: 11/16/2022 04:40 pm »
Somewhere in the early 90s, during the genetic search period, we had a few "beer heavy" discussions when we saw Kangaroo search in a newsgroup. Was is sci.physics or something? Not sure. In a nutshell, there are various ways to find a best solution for your models and dropping kangaroos on Earth to find the highest point is one example solution to one optimization problem. There is a readable discussion here at http://processcontrol4dummies.blogspot.com/2013/06/optimization-technique-kangaroo-analogy.html

Enjoy.
I've seen a version of this before, but this is much more elaborate. I still like the description of simulated annealing, where the kangaroo starts off drunk and gradually sobers up. :-)

"If it's looking for debris of a crashed space probe on the surface of the Moon, then it would have had to train by looking at enough debris fields. "

Not necessarily.  I many cases you compare your new image with a pre-impact image.  Example: the Chang'e 1 orbiter impact... not present in an Apollo 16 panoramic camera image but present in LRO images.  Almost any future impact site will be findable that way.  Older impacts like (for instance) the Apollo 16 LM ascent stage are much trickier.  But if you think AI can do it better than a human, think again.
In other words, looking for a "change" or "deviation" in a static landscape is almost a binary classification problem i.e. easier than of collection millions of "debris" pictures for fancy classifier. As always, problem formulation is key. Plus a well chosen algorithmic solution.

This problem is an interesting one. You've got regular pictures from Mars (or the moon--or wherever) and you're looking for changes. But simply because of the angle of the sun, you will get changes in the images, so you need an algorithm to find unexpected changes. Toward that end, I think the best approach might be to try to construct a 3D map of the surface, based on the images and the known sun angles. But that's quite a challenge, given that it involves assumptions about roughness and reflectivity--not just elevation. It's definitely an optimization problem--even without trying to find real changes--but I'm not sure AI figures into it. A few quick searches didn't turn up any papers on the topic, but I'm probably using the wrong terms or something.

Offline sanman

  • Senior Member
  • *****
  • Posts: 6057
  • Liked: 1348
  • Likes Given: 8
Re: How Can AI Be Used for Space Applications?
« Reply #25 on: 11/20/2022 10:31 pm »
Wasn't there some case of NASA designing a communication antenna using machine learning? How did that work?

Offline Twark_Main

  • Senior Member
  • *****
  • Posts: 3793
  • Technically, we ALL live in space
  • Liked: 1961
  • Likes Given: 1231
« Last Edit: 11/21/2022 02:09 am by Twark_Main »

Online Eer

  • Full Member
  • ****
  • Posts: 642
  • Liked: 476
  • Likes Given: 959
Re: How Can AI Be Used for Space Applications?
« Reply #27 on: 11/21/2022 01:57 am »
Others here have provided good answers. One aspect of machine learning, whether via neural networks or genetic algorithms, is that for either, you need a high fidelity simulation of reality in which to test / train or trial via tournament competitions in order to be able to rank / compare which alternative to proceed with (or to adjust weights in the neural net). My information is old - from around 1990 - but worked with both.

One style of solution my colleague had success with genetic algorithms was design of parameters for a fuzzy controller.  The genetic algorithm did a better job of finding precise results when the resolution of the controller was more coarse than humans provided - they tried to get closer and closer until the controller could no longer step in small enough steps to get better.  The genetic algorithms moved the control outside the closer range, and was then able to take larger steps to get closer.  It was a solution human engineers hadn't considered and was measurably better.  Chances are that once the genetic algorithm taught the humans a better approach, they no longer needed the algorithm.

Just an anecdote.
From "The Rhetoric of Interstellar Flight", by Paul Gilster, March 10, 2011: We’ll build a future in space one dogged step at a time, and when asked how long humanity will struggle before reaching the stars, we’ll respond, “As long as it takes.”

Offline sanman

  • Senior Member
  • *****
  • Posts: 6057
  • Liked: 1348
  • Likes Given: 8
Re: How Can AI Be Used for Space Applications?
« Reply #28 on: 11/22/2022 07:34 am »
Others here have provided good answers. One aspect of machine learning, whether via neural networks or genetic algorithms, is that for either, you need a high fidelity simulation of reality in which to test / train or trial via tournament competitions in order to be able to rank / compare which alternative to proceed with (or to adjust weights in the neural net). My information is old - from around 1990 - but worked with both.

One style of solution my colleague had success with genetic algorithms was design of parameters for a fuzzy controller.  The genetic algorithm did a better job of finding precise results when the resolution of the controller was more coarse than humans provided - they tried to get closer and closer until the controller could no longer step in small enough steps to get better.  The genetic algorithms moved the control outside the closer range, and was then able to take larger steps to get closer.  It was a solution human engineers hadn't considered and was measurably better.  Chances are that once the genetic algorithm taught the humans a better approach, they no longer needed the algorithm.


Haha, unless the genetic algorithm is able to find even better tricks later on.

Quote
Just an anecdote.

That's a good story - worth repeating.

Offline Greg Hullender

  • Full Member
  • ****
  • Posts: 624
  • Seattle
    • Rocket Stack Rank
  • Liked: 452
  • Likes Given: 349
Re: How Can AI Be Used for Space Applications?
« Reply #29 on: 11/22/2022 03:54 pm »
Haha, unless the genetic algorithm is able to find even better tricks later on.
Despite the name, a genetic algorithm can't really find "new tricks" in the way you're probably thinking. The programmer essentially gives it a set of things it can try to combine in different orders, and it gets to try different combinations. But what it cannot do is "think outside the box." It can't decide to try a sequence that the programmer didn't permit. It can't try to include things the programmer didn't include in the allowed set. And it definitely can't come up with a whole new approach.

Of course you can imagine that the programmer gave it complete freedom to try anything. (I.e. the things it's combining are low-level software instructions.) But the likelihood of success is greater the more restrictions you can put on it. (An algorithm generating arbitrary machine instructions will likely run for billions of years without producing anything useful.) So the more you restrict it, the more likely it is to find something, but the less likely it is to find something original.

When things like this work at all, they typically find a few good solutions relatively early on, but the rate of discovery tapers off petty fast, and the later discoveries are usually tiny improvements that aren't really worth the effort. You never see a case where, after the initial set of results, it goes dry for a few weeks and then makes a breakthrough discovery. Once it goes dry, your best bet is to shake up the whole algorithm; that is, give it more/different things to try, reshuffle the restrictions, and maybe even change how you score the results. That's much more likely to produce different, useful results.

And, of course, best of all is for the human programmer to try to think outside the box to find a whole different way to attack the problem. That's our superpower, after all.

Offline ppnl

  • Full Member
  • ***
  • Posts: 312
  • Liked: 209
  • Likes Given: 19
Re: How Can AI Be Used for Space Applications?
« Reply #30 on: 11/22/2022 04:57 pm »
It is worth noting that we and our superpower evolved.

Offline Greg Hullender

  • Full Member
  • ****
  • Posts: 624
  • Seattle
    • Rocket Stack Rank
  • Liked: 452
  • Likes Given: 349
Re: How Can AI Be Used for Space Applications?
« Reply #31 on: 11/22/2022 05:08 pm »
It is worth noting that we and our superpower evolved.
Indeed. Via real evolution, which had quadrillions of individuals evolving in parallel for billions of years. (Or millions for millions of years, if you just want to count hominids.) No existing GA system has anything remotely comparable.

Offline ppnl

  • Full Member
  • ***
  • Posts: 312
  • Liked: 209
  • Likes Given: 19
Re: How Can AI Be Used for Space Applications?
« Reply #32 on: 11/22/2022 07:28 pm »
It is worth noting that we and our superpower evolved.
Indeed. Via real evolution, which had quadrillions of individuals evolving in parallel for billions of years. (Or millions for millions of years, if you just want to count hominids.) No existing GA system has anything remotely comparable.

True but real evolution is a vastly inefficient and slow process. Humans have generations that are decades long. Evolution is pulling them in multiple competing directions and there is a vast amount of irrelevant noise in a natural environment.

But I think it is clear that GAs will never directly lead to human level intelligence. For that you will probably need neural nets. GAs might be useful for developing those neural nets.



 

Offline Greg Hullender

  • Full Member
  • ****
  • Posts: 624
  • Seattle
    • Rocket Stack Rank
  • Liked: 452
  • Likes Given: 349
Re: How Can AI Be Used for Space Applications?
« Reply #33 on: 11/22/2022 08:19 pm »
But I think it is clear that GAs will never directly lead to human level intelligence. For that you will probably need neural nets. GAs might be useful for developing those neural nets.
I think you're being misled by the term "neural nets." Neural nets are an excellent tool, but I suspect you imagine they do something rather different from what they really do. They're probably best described as "loosely inspired by an obsolete understanding of how neurons in the brain work." What they really are is a method to try to find a function that describes a particular set of inputs.

To clarify what I mean by that, consider an expression like y = f(x). If we know that f(x) = x^2 and I give you x and ask you for y, that's function evaluation, and the answer is x^2. If I give you y and ask you for x, that's function inversion, and the answer is sqrt(y) (or the negative of that). But neural nets are designed to solve the problem when I give you lots of (x,y) pairs and ask you to find f. In general, x and y are both vectors. (Neural nets are not the only way to try to do this, by the way.)

I think you can see how this could be very useful, but it should also be clear that it has no tangible connection to human intelligence.

Offline JayWee

  • Full Member
  • ****
  • Posts: 1014
  • Liked: 1021
  • Likes Given: 1986
Re: How Can AI Be Used for Space Applications?
« Reply #34 on: 11/22/2022 10:47 pm »
Haha, unless the genetic algorithm is able to find even better tricks later on.
Despite the name, a genetic algorithm can't really find "new tricks" in the way you're probably thinking.
Check this story out: https://hackaday.com/2018/11/12/how-to-evolve-a-radio/

Offline Greg Hullender

  • Full Member
  • ****
  • Posts: 624
  • Seattle
    • Rocket Stack Rank
  • Liked: 452
  • Likes Given: 349
Re: How Can AI Be Used for Space Applications?
« Reply #35 on: 11/23/2022 12:40 am »
Haha, unless the genetic algorithm is able to find even better tricks later on.
Despite the name, a genetic algorithm can't really find "new tricks" in the way you're probably thinking.
Check this story out: https://hackaday.com/2018/11/12/how-to-evolve-a-radio/
Interesting story, although I don't think those are the sort of "new tricks" sanman had in mind. This is much like the earlier story, where the GA discovered there was an advantage to starting with really large swings of the dial. In this case, "new tricks" would correspond to it trying entirely new kinds of electronics.

The abstract of the paper itself is also of some interest.
Quote from: The Evolved Radio and its Implications for Modelling the Evolution of Novel Sensors
Sensor evolution research typically uses evolutionary algorithms (EAs) to generate sensors that near optimally satisfy large numbers of constraints. This is qualitatively different from the phylogenetic process found in nature that has resulted, for example, in the mammalian auditory ossicles evolving from the jaw bones of amphibians and reptiles, that in turn had previously acted as gill arches in fish. This paper describes an evolvable hardware experiment that resulted in a network of transistors sensing and utilising the radio waves emanating from nearby PCs. We argue that this evolved ‘radio’ is only the second device ever whose sensors were constructed in a way that in key aspects is analogous to that found in nature. We highlight the advantages and disadvantages of this approach and show why it is practically impossible to implement a similar process in simulation.

This claims that the approach in the paper (which is actually over twenty years old) was just the second time anyone successfully used genetic algorithms worthy of the name for anything useful. I note, though, that I don't find a lot of citations to it since it came out, and most that I do find are in biology papers, not computer science. It certainly doesn't seem to have revolutionized the field. I suspect most of the value in it comes from providing different ways to think about real evolution, but not to finding solutions to actual technical problems.

As I said earlier, there's definitely something romantic about EA and other biologically-inspired systems, but in the real world of limited budgets and limited time, they almost always (if not always) fall short of less romantic but more practical approaches. Ones that usually go by "machine learning" and not "artificial intelligence."
« Last Edit: 11/23/2022 12:41 am by Greg Hullender »

Offline ppnl

  • Full Member
  • ***
  • Posts: 312
  • Liked: 209
  • Likes Given: 19
Re: How Can AI Be Used for Space Applications?
« Reply #36 on: 11/23/2022 12:42 am »
But I think it is clear that GAs will never directly lead to human level intelligence. For that you will probably need neural nets. GAs might be useful for developing those neural nets.
I think you're being misled by the term "neural nets." Neural nets are an excellent tool, but I suspect you imagine they do something rather different from what they really do. They're probably best described as "loosely inspired by an obsolete understanding of how neurons in the brain work." What they really are is a method to try to find a function that describes a particular set of inputs.

Well yes but what is the alternative? If you believe that the universe is computable then anything that happens is a function with inputs and outputs. Cause and effect. A brain neuron produces output based on inputs. A function. To learn it has to produce a different output for the given inputs. A function search.

To learn a neuron must get feedback from the output somehow. An ANN uses hardwired backpropagation external to the ANN. A brain cannot do this because all it has is neurons.

Backpropagation seems to be vastly better than whatever the brain does in some ways. For example alphago only has the number of neurons of an insect yet it beats grandmasters. It's like playing chess with an ant and having it wipe the floor with you. It is kind of scary.

In other ways backpropagation is vastly inefficient. It requires a vast amount of training data and massive power to learn. The brain could not deal with that amount of training data or power needed to do calculations.

The brain probably uses networks of neurons that feedback information from output back to input. Thus the backpropagation function is contained in the network itself. Essentially a neural net that looks at itself to adjust its on function. The problem is that these neural circuits seem to be very unstable in ANNs.

The brain probably learns with less training data because evolution has structured the brain to rapidly learn the type of things we need to learn. Those structures were created by genetic algorithms and evolution.   

Quote
To clarify what I mean by that, consider an expression like y = f(x). If we know that f(x) = x^2 and I give you x and ask you for y, that's function evaluation, and the answer is x^2. If I give you y and ask you for x, that's function inversion, and the answer is sqrt(y) (or the negative of that). But neural nets are designed to solve the problem when I give you lots of (x,y) pairs and ask you to find f. In general, x and y are both vectors. (Neural nets are not the only way to try to do this, by the way.)

I think you can see how this could be very useful, but it should also be clear that it has no tangible connection to human intelligence.

No. An atom would seem to have little in common with an airplane. But an airplane is ultimately composed of atoms thus any understanding of airplanes must eventually be grounded in an understanding of atoms. Brain cells compute functions and the function they compute can be altered on the fly by some feedback mechanism. It is a function search. Any understanding of the brain must be grounded in the function search happening in these neurons.

Offline Greg Hullender

  • Full Member
  • ****
  • Posts: 624
  • Seattle
    • Rocket Stack Rank
  • Liked: 452
  • Likes Given: 349
Re: How Can AI Be Used for Space Applications?
« Reply #37 on: 11/23/2022 12:45 am »
Well yes but what is the alternative?
Remember that we're talking about space applications here. For that purpose, there are lots of alternatives. I'm afraid a deep discussion over how the brain really works is off-topic.

Offline ppnl

  • Full Member
  • ***
  • Posts: 312
  • Liked: 209
  • Likes Given: 19
Re: How Can AI Be Used for Space Applications?
« Reply #38 on: 11/23/2022 12:56 am »
Well yes but what is the alternative?
Remember that we're talking about space applications here. For that purpose, there are lots of alternatives. I'm afraid a deep discussion over how the brain really works is off-topic.

Yeah, probably off topic a bit. But we would have to agree on what could be achieved with AI in order to decide on how it could be used in space. Maybe someday AIs will be discussing how we can be used in space applications. I live in fear of their conclusion.

Offline JayWee

  • Full Member
  • ****
  • Posts: 1014
  • Liked: 1021
  • Likes Given: 1986
Re: How Can AI Be Used for Space Applications?
« Reply #39 on: 11/23/2022 01:05 am »
Well yes but what is the alternative?
Remember that we're talking about space applications here. For that purpose, there are lots of alternatives. I'm afraid a deep discussion over how the brain really works is off-topic.
Just one little "offtopic thing" I've seen very recently - https://www.csail.mit.edu/news/solving-brain-dynamics-gives-rise-flexible-machine-learning-models


Tags:
 

Advertisement NovaTech
Advertisement Northrop Grumman
Advertisement
Advertisement Margaritaville Beach Resort South Padre Island
Advertisement Brady Kenniston
Advertisement NextSpaceflight
Advertisement Nathan Barker Photography
0