Quote from: JayWee on 11/15/2022 08:13 pmIsn'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.
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?)
Quote from: JayWee on 11/15/2022 08:13 pmIsn'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]
"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.
Quote from: leovinus on 11/15/2022 05:31 pmQuote from: jimvela on 11/15/2022 03:32 pmAutonomous 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.
Quote from: jimvela on 11/15/2022 03:32 pmAutonomous 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
Autonomous onboard navigation...
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.htmlEnjoy.
Quote from: Phil Stooke on 11/15/2022 08:27 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.
Wasn't there some case of NASA designing a communication antenna using machine learning? How did that work?
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.
Haha, unless the genetic algorithm is able to find even better tricks later on.
It is worth noting that we and our superpower evolved.
Quote from: ppnl on 11/22/2022 04:57 pmIt 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.
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.
Quote from: sanman on 11/22/2022 07:34 amHaha, 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.
Quote from: Greg Hullender on 11/22/2022 03:54 pmQuote from: sanman on 11/22/2022 07:34 amHaha, 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/
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.
Quote from: ppnl on 11/22/2022 07:28 pmBut 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.
Well yes but what is the alternative?
Quote from: ppnl on 11/23/2022 12:42 amWell 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.