How Can AI Be Used for Space Applications?

[sanman]

I came across this interesting article about how AI is being used to conjure up completely new proteins very quickly:

https://www.nature.com/articles/d41586-022-02947-7


It occurred to me that this could enable all sorts of designer bugs for ISRU purposes, and maybe even terraforming.

Perhaps we could have bugs designed to survive the Martian day/night cycle, which would come alive during the day, to perform useful conversion of natural resources, like through Sabatier or whatever.

Could we even use AI to design complex ecosystems of organisms that would cope with the existing Mars conditions while working to transform the environment into one that's more human-friendly?

[MickQ]

Sounds a bit “Sax Russell “ to me.

A great idea though.  A lot more sophisticated than throwing rocks and/or nukes around, and more controllable IMHO.

[sanman]

Another interesting use of AI is in image enhancement:

https://ai.googleblog.com/2021/07/high-fidelity-image-generation-using.html

The article shows various examples of low-resolution images being enhanced into higher-resolution images. This would obviously be useful for satellite imaging.

What are the limits of this approach? What are the risks of introducing errors through this process.

[Phil Stooke]

Garbage in, garbage out.  The low resolution images are being made to look like high resolution images.  Actual new information is not being created.  I admit that the better-looking image might be easier to interpret and can be a help, but new information is not being created.  Scientific uses would be a lot more limited than uses in other fields (like movie CGI).  The people who want to sell these kinds of things will claim they can do things they cannot really do.

[laszlo]

Also, keep in mind that none of this is artificial intelligence, any more than solving integrals, playing chess, natural language translation or image identification was. The tendency is for AI investigators is to identify something that they find difficult to do as a benchmark of intelligence, develop an "AI" system that accomplishes the task and then realize that the product is performing with no intelligence at all. So they go back to their terminals, move the goalposts and try again. In the meantime, entrepreneurs productize these idiot savant creations and the popular press announces the imminent Rise of the Machines, again. This has been going on since the 1950's, at least.

The current iteration of AI is a deep learning system that instead of being explicitly programmed to do a specific task learns how to do it by being fed enormous amount of training data, guessing at what the right answer is and having the guess evaluated. The quality of the current guess is used to adjust how future guesses are made until they are consistently good enough. Then whenever new data is presented the deep learning system will make a guess based on its experience with the training data sets. This is the principle behind the current generation of "AI" systems, minus the mathematics and implementation details.

While this sounds similar to the way an intelligent being learns, there's no intelligence involved at all, at least not in operation (the design and implementation is a different story). These systems are basically self-training classifiers. They divide training data into sets and they determine which set the operational data belongs to. They are pattern matchers based on statistical models derived from the training data. And as Phil Stooke points out above, GIGO.

If you really want to determine how deep learning ("AI")  can be used for space applications, look for problems that require the classification of data into sets. You also need massive amounts of training data to be available, which is why I am skeptical that we're ready to design Mars ISRU bugs yet. We don't have enough information about the Martian environment and interactions with living creatures yet to train the deep learning systems.

[sanman]

Well, there also appear to be more recently explored approaches like Generative Adversarial Networks that are using Darwinistic type algorithms, and just rapidly iterating through them to create new content. So I assume that's how it gets new information to come up with possible proteins. Real biological Darwinism hasn't been able to get a foothold on Mars, but maybe some kind of artificial/digital forced Darwinism can get us past the initial bootstrapping phase. Then after that, we can rely on real biological Darwinism to take over on Mars.

As for classifiers, for a lot of routine Earthly applications, that's all you need. If your AI is looking through satellite imagery to spot a tank on the ground, then as long as it's already trained by looking at enough tanks, then it should be able to meet the needs. 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. If you're trying to upscale a person's face from low-res to high-res, then having trained on a bunch of faces won't be enough to know if a mole should be visible in the high-res image that wasn't apparent in the low-res one. But for generalizations, it should be okay to interpolate/extrapolate. Rather than creating new information, you're "transferring" (inferring) it from the training data.

[john smith 19]

Garbage in, garbage out.  The low resolution images are being made to look like high resolution images.  Actual new information is not being created.  I admit that the better-looking image might be easier to interpret and can be a help, but new information is not being created.  Scientific uses would be a lot more limited than uses in other fields (like movie CGI).  The people who want to sell these kinds of things will claim they can do things they cannot really do.

Now if they'd said "We are using these low resolution images as part of a collection of data sources that help us decide what something is (or is not)" that would be closer to a task humans do, and have done that does call for intelligence. Examples include submarine tracking and (in principle) weather forecasting (the way humans did it, not using GCMs)

Technology looking for a problem? Surely if the problem was that difficult to solve someone would have thought of applying AI long ago.

There is an interesting subset of design called "Inverse" or "optimal" design which seeks to take parameters constituting an "optimal" solution (for some suitable defintion of the word) and work backward to an actual design (or a set of designs that fulfil those criteria).

I don't think any of the proponents of such concepts have ever labelled them as AI, and given the deterministic nature of the process I don't think it is. OTOH it could save a hell of a lot of time and money.

[leovinus]

Also, keep in mind that none of this is artificial intelligence, any more than solving integrals, playing chess, natural language translation or image identification was.

Which is why the practitioners, scientists and engineers, in the field refer to "machine learning" (ML) or ML/AI.

The term "Artificial Intelligence" works well for the public but I always cringe when I hear it. One reason we do not like the AI name is that we grew up with it in university in the 80s and 90s. AI was going to explain "why" things do work, and generalize from one given information to new insights. Prolog was one of the main languages (and I still do not like it )

Japan had a major investment to build a 5th generation AI machine, made major investments, and it all did not work as the tools like Prolog were not up to the job. For example The fifth generation : artificial intelligence and Japan's computer challenge to the world. As far as I concerned, Japan's effort from that time was to be applauded but again the tools were not there.

Same thing with "simple" backprop in the 80s. We could not compute quick enough with enough data so it was a toy.

Then we used hidden Markov models (HMM) for decades (80s to 10s) on NLP, translation, speech, handwriting

Then we had computing power and suddenly the old tool of multi layer perception (MLP) from the 80s worked so much better than HMMs which was a surprise. 

And now we are back on "explainable AI" or XAI in the context of deep learning.

Science and engineering moves in waves and circles, iterative improvement.

Have a look at something like this ESA-Ariel Data Challenge NeurIPS 2022: Introduction to exo-atmospheric studies and presentation of the Ariel Big Challenge (ABC) Database and Reinforcement Learning for robotics et al

[Twark_Main]

Also, keep in mind that none of this is artificial intelligence, any more than solving integrals, playing chess, natural language translation or image identification was. The tendency is for AI investigators is to identify something that they find difficult to do as a benchmark of intelligence, develop an "AI" system that accomplishes the task and then realize that the product is performing with no intelligence at all. So they go back to their terminals, move the goalposts and try again. In the meantime, entrepreneurs productize these idiot savant creations and the popular press announces the imminent Rise of the Machines, again. This has been going on since the 1950's, at least.

The joke-that's-not-a-joke within the field is that "AI is the set of all software problems we can't solve yet."

Once we solve a given AI problem (eg defeating a chess grandmaster) it immediately gets its own specialized name ("two-player game playing") and thus it no longer falls under "AI." A nice catch-22!

AI is, fundamentally, a marketing term. It has no consistent technical definition. I once heard a tongue-in-cheek suggestion of "any program which contains at least one branching instruction," and based on real-world corporate usage that definition seems about right actually... 

[deltaV]

AI is, fundamentally, a marketing term. It has no consistent technical definition.

I agree.

I once heard a tongue-in-cheek suggestion of "any program which contains at least one branching instruction," and based on real-world corporate usage that definition seems about right actually... 

I bet the max(0, x) in deep neural network ReLU units is typically computed using instructions for max rather than branches. If that's true then deep neural networks don't actually include any branches in the core code. So modern AI code may actually execute fewer branches per second than many non-AI tasks such as sorting.

[JohnFornaro]

Garbage in, garbage out.  The low resolution images are being made to look like high resolution images.  Actual new information is not being created.  I admit that the better-looking image might be easier to interpret and can be a help, but new information is not being created.  Scientific uses would be a lot more limited than uses in other fields (like movie CGI).  The people who want to sell these kinds of things will claim they can do things they cannot really do.

I would add that in some cases, "new" info is artificially added to the image.  Remember that face on Mars?  Some of those images were doctored enhanced to make the face more realistic, simian in some cases.  They didn't call that AI then, but now they do.

[sanman]

Here's an overview from DARPA's John Launchbury on the "3 Waves of AI":




So we're currently in the middle of realizing the fruits of the 2nd wave, but the 3rd wave is yet to come.

[Greg Hullender]

I spent my whole professional career working on "AI" (actually natural-language and machine-learning software), so I think I can add a few useful remarks.

For this discussion, I think the following definition of "AI" will serve everyone well: AI is software that attacks problems that cannot be solved by brute force but which humans can do "intuitively" without being able to explain how they do it.

So in the protein-folding example, humans use a lot of intuition to find solutions even though the search space is way too large to fully explore. An AI solution looks at the results the humans come up with and deduces some rules. These are almost certainly not rules the humans actually use, but as long as they do the job, who cares?

It occurred to me that this could enable all sorts of designer bugs for ISRU purposes, and maybe even terraforming.
.
.
.
Could we even use AI to design complex ecosystems of organisms that would cope with the existing Mars conditions while working to transform the environment into one that's more human-friendly?

As far as I know, a human team would be unable to create "designer bugs," so AI can't do it either. There is no magic here.

Also, keep in mind that none of this is artificial intelligence, any more than solving integrals, playing chess, natural language translation or image identification was. The tendency is for AI investigators is to identify something that they find difficult to do as a benchmark of intelligence, develop an "AI" system that accomplishes the task and then realize that the product is performing with no intelligence at all. So they go back to their terminals, move the goalposts and try again. In the meantime, entrepreneurs productize these idiot savant creations and the popular press announces the imminent Rise of the Machines, again. This has been going on since the 1950's, at least.

There are people who do this, but I don't consider them serious researchers. "AI" actually has two wings to it, which I like to describe as corresponding to Astronomy vs. Astrology. The folks who freely use the term "AI" are the astrologers. The ones who prefer to say "machine learning" and who don't make extravagant claims are the astronomers. (Obviously I have a bias here.) :-)

The current iteration of AI is a deep learning system that instead of being explicitly programmed to do a specific task learns how to do it by being fed enormous amount of training data, guessing at what the right answer is and having the guess evaluated. The quality of the current guess is used to adjust how future guesses are made until they are consistently good enough. Then whenever new data is presented the deep learning system will make a guess based on its experience with the training data sets. This is the principle behind the current generation of "AI" systems, minus the mathematics and implementation details.

This is a pretty good description of how machine learning works. No one pretends it has anything to do with actual intelligence, nor imagines that it ever will. Not all machine learning is classifiers, though. The protein-folding software isn't, for example.

Well, there also appear to be more recently explored approaches like Generative Adversarial Networks that are using Darwinistic type algorithms, and just rapidly iterating through them to create new content.

Sadly, this is from the astrology side of AI. Things like "evolutionary algorithms" and "swarms" all sound cool, but they all underperform simple hill-climbing-with-restart algorithms--when they work at all. A co-worker at Microsoft spent eighteen months trying to get something useful out of this stuff, and that was his conclusion at the end of it. We also attended a few conferences and read a number of papers, and the results weren't encouraging. Things can always change, of course, but there's a certain kind of magical thinking that suffuses that camp which contrasts sharply with the coldly mathematical air you find at machine-learning conferences, and that doesn't make me optimistic.

As for classifiers, for a lot of routine Earthly applications, that's all you need. If your AI is looking through satellite imagery to spot a tank on the ground, then as long as it's already trained by looking at enough tanks, then it should be able to meet the needs. 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.

The problem with training a system to spot a crashed space probe is that you don't have very many examples of real space probes to learn from. And AI is not very good at identifying the unexpected. Classifiers tend to be like my late great aunt, who always had an opinion on everything, whether she knew anything about it or not. It's quite difficult to design a useful system that will say, "I don't know what this is." Either they say it too often or not often enough. There are exceptions, e.g. when a problem can be characterized statistically, but I struggle to see how to characterize the problem of finding a crashed probe on Mars that way.

If you're trying to upscale a person's face from low-res to high-res, then having trained on a bunch of faces won't be enough to know if a mole should be visible in the high-res image that wasn't apparent in the low-res one. But for generalizations, it should be okay to interpolate/extrapolate. Rather than creating new information, you're "transferring" (inferring) it from the training data.

I've played with this software some, and I've been pretty unimpressed. I'm not sure exactly what they're doing, but on the dozen or so pictures I tested it with, it wasn't clear that it made any of them better. At this point, I'd say, don't expect too much from this. Again, I could be wrong, but I'm not optimistic at this point.

The joke-that's-not-a-joke within the field is that "AI is the set of all software problems we can't solve yet."

Once we solve a given AI problem (eg defeating a chess grandmaster) it immediately gets its own specialized name ("two-player game playing") and thus it no longer falls under "AI." A nice catch-22!

I've been making this joke myself since the late 1970s. :-) It never gets old. Another one is, "AI is any cool algorithm that almost works." (If it actually does work, it's not AI anymore.)

Anyway, I'd say that machine learning does have a lot to contribute to space applications, mostly by automating routine human tasks so they can be done by a probe in real time rather than requiring a long wait due to the speed-of-light lag. There are probably other areas--look for anything that requires lots of tedious human labor.

[jimvela]

Autonomous onboard navigation...

[Greg Hullender]

Another point worth mentioning: Machine learning is usually something that happens in the factory, not in the field. That is, the classifier (or other "AI" technology) in a space probe would most likely be trained on the ground here on Earth. Based on the data collected during the mission, NASA might retrain the system and upload the result. Retraining takes huge computer resources and can run for days, weeks--even months. In this process, the system does not learn to do anything new; it just learns how to do better at the tasks it was originally designed for. (Of course nothing stops the engineers from also adding new capabilities at the same time.)

This retraining might happen once every year or two. But the system itself would not learn anything in between times.

There are systems that do attempt to learn in real time, but I have never seen one that worked very well. You've got a system that was trained (for example) on millions of samples of data and you're trying to update it with a tiny fraction of that amount of new data. Usually, you either get no measurable effect or else you "poison" the system by placing too much emphasis on the new data.

And even those systems don't try to learn new things; they just learn to do their old jobs better.

[leovinus]

AI is, fundamentally, a marketing term. It has no consistent technical definition.

I agree.

I once heard a tongue-in-cheek suggestion of "any program which contains at least one branching instruction," and based on real-world corporate usage that definition seems about right actually... 

I bet the max(0, x) in deep neural network ReLU units is typically computed using instructions for max rather than branches. If that's true then deep neural networks don't actually include any branches in the core code. So modern AI code may actually execute fewer branches per second than many non-AI tasks such as sorting.

With respect to "instructions for max", that just means that the "if" conditions are coded in microcode on the CPU one level lower. In other words, the "if" condition is still there.

With respect to "So modern AI code may actually execute fewer branches per second than many non-AI tasks such as sorting.", sure, but essentially that is just good software design. For example, a matrix-matrix calculation should be coded not with three nested loops, but with a GEMM() library call optimized for whatever platform you work which therefore minimizes overhead, branches etc.

[leovinus]

Autonomous onboard navigation...

Well, let's start with making better software for avionics and anything. While I might be missing something, there was tweet yesterday by Chris G. saying something like "If the SLS countdown is halted shorter than T-33 seconds then the consequence is a SCRUB as the SLS avionics cannot handle that.". My memory might be wrong and there are various experts here who will know the answer but I seem to remember that the Shuttle allowed for halt below T-33 seconds? If my memory is correct then it would mean that newer SLS avionics have less functionality in a Shuttle derived system which would be disturbing. 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

[JayWee]

It occurred to me that this could enable all sorts of designer bugs for ISRU purposes, and maybe even terraforming.
.
.
.
Could we even use AI to design complex ecosystems of organisms that would cope with the existing Mars conditions while working to transform the environment into one that's more human-friendly?

As far as I know, a human team would be unable to create "designer bugs," so AI can't do it either. There is no magic here.

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?)

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.

[Phil Stooke]

"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.

[Greg Hullender]

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.

[InterestedEngineer]

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

[leovinus]

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.

[leovinus]

"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.

[leovinus]

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.

[Greg Hullender]

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.

[sanman]

Wasn't there some case of NASA designing a communication antenna using machine learning? How did that work?

[Twark_Main]

Wasn't there some case of NASA designing a communication antenna using machine learning? How did that work?

This should get you started.

https://en.wikipedia.org/wiki/Evolved_antenna


Research papers:

https://www.researchgate.net/publication/226096470_An_Evolved_Antenna_for_Deployment_on_Nasa's_Space_Technology_5_Mission

https://ntrs.nasa.gov/citations/20060024675

https://digitalcommons.usu.edu/cgi/viewcontent.cgi?article=1426&context=smallsat

https://www.academia.edu/download/30679115/hornby_ec11.pdf

Press releases:

https://www.nasa.gov/centers/ames/news/releases/2004/04_55AR.html

https://www.nasa.gov/mission_pages/st-5/main/04-55AR.html

Image gallery:

https://www.nasa.gov/centers/ames/news/releases/2004/antenna/computer.html

Video:





I'd seen this before, but I was interested to discover Al Globus was involved in the work.

[Eer]

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.

[sanman]

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.

Just an anecdote.

That's a good story - worth repeating.

[Greg Hullender]

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.

[ppnl]

It is worth noting that we and our superpower evolved.

[Greg Hullender]

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.

[ppnl]

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.



 

[Greg Hullender]

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.

[JayWee]

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/

[Greg Hullender]

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.

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."

[ppnl]

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.   

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.

[Greg Hullender]

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.

[ppnl]

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.

[JayWee]

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

[Greg Hullender]

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.

Nothing based on any existing technology has any hope of ever evolving into something like what you're imagining. An effort to develop human intelligence using current AI technology is like trying to get to the moon using ancient Babylonian technology (i.e. building a tower tall enough to reach the moon). You simply cannot scale it up that far. Nowhere near.

[edzieba]

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.

Nothing based on any existing technology has any hope of ever evolving into something like what you're imagining. An effort to develop human intelligence using current AI technology is like trying to get to the moon using ancient Babylonian technology (i.e. building a tower tall enough to reach the moon). You simply cannot scale it up that far. Nowhere near.

Until maybe 5-10 years ago I'd have agreed, but the recent 'Deep Learning' boom has shown that techniques long disregarded and woefully inefficient and inelegant can be perfectly adequate at performing real-world tasks if you can throw enough hardware at the problem.
Human consciousness arises from a massively parallel array of incredibly stupid individual neurons working kinda-sorta in tandem and sometimes producing the 'correct' result, so I rather suspect any human-like AI will arise from a similarly messy pile of hacks and excessive use of compute resources rather than some elegant new simulation technique.

[Barley]

Nothing based on any existing technology has any hope of ever evolving into something like what you're imagining. An effort to develop human intelligence using current AI technology is like trying to get to the moon using ancient Babylonian technology (i.e. building a tower tall enough to reach the moon). You simply cannot scale it up that far. Nowhere near.

The Babylonian technology that matters is agriculture.  That lets you expand the population, so you have enough one in a million geniuses (or a million non-geniuses) that you can refine copper, steel, silicon, ... .  Five millennia later you're on the Moon.

Time scales may vary.

[JohnFornaro]

Nothing based on any existing technology has any hope of ever evolving into something like what you're imagining. An effort to develop human intelligence using current AI technology is like trying to get to the moon using ancient Babylonian technology...

The Babylonian technology that matters is agriculture.  That lets you expand the population, so you have enough one in a million geniuses (or a million non-geniuses) that you can refine copper, steel, silicon, ... .  Five millennia later you're on the Moon.

Time scales may vary.

Which is an excellent point. 

It is now thought that mankind can determine its future, and some transhumanists are positing that AI will set us "free".  But what about those Babylonians, back in the day?  How many of them had "visions" of what mankind's future could have been?  Substitute "dreams" for "visions", if desired.

As an aside, I agree that agriculture is crucial to any colonization effort.  I've suggested taking a load of Eastern forest dirt up to the first hab, planting it, and empirically finding out what happens, rather than trying to create an AI model of what lunar agriculture could be.

[sanman]

Whatever the disadvantages, limitations and tradeoffs of AI or machine learning may be, one clear advantage is that it's done with machines - and machines don't automatically need food, water, and oxygen.

Machines may not need the fruits of agriculture, but they can make agriculture possible in space, on some off-world body. And they can do it at scale, and in abundance, for the benefit of humans.

But how can we bootstrap such applications here on Earth, in order to eventually extend them into space?

[JohnFornaro]

Whatever the disadvantages l... machines don't automatically need food, water, and oxygen.
But how can we bootstrap such applications here on Earth, in order to eventually extend them into space?

Pretty sure they "automatically" need electricity.

As to "bootstrapping", that implies sentience.

[Elmar Moelzer]

Things that I am hoping to see machine learning algorithms tackle:
New materials/alloys/ceramics for different applications. E.g. you give it the desired properties for the material and it will come up with a bunch of different solutions. I would like to see what machine learning could do with that.
I could imagine something similar for engine configurations, maybe solar panels, fusion power generator configurations (is already happening to a small extent), etc, etc.
I don't think that we are quite there yet for these things, but given the recent advances it could be sooner rather than later.

[Greg Hullender]

There's lots of good work on using machine learning for alloys. Google machine learning for alloy design.

Basically, if it's a tedious job that humans can do but can't explain precisely how they do it, it's a good candidate for ML.

[Greg Hullender]

. . . An effort to develop human intelligence using current AI technology is like trying to get to the moon using ancient Babylonian technology (i.e. building a tower tall enough to reach the moon). You simply cannot scale it up that far. Nowhere near.

Until maybe 5-10 years ago I'd have agreed, but the recent 'Deep Learning' boom has shown that techniques long disregarded and woefully inefficient and inelegant can be perfectly adequate at performing real-world tasks if you can throw enough hardware at the problem.
Human consciousness arises from a massively parallel array of incredibly stupid individual neurons working kinda-sorta in tandem and sometimes producing the 'correct' result, so I rather suspect any human-like AI will arise from a similarly messy pile of hacks and excessive use of compute resources rather than some elegant new simulation technique.

I've been impressed at deep learning as well--we always had the notion of multi-layer perceptron networks, but no one had a good way to train them. It definitely moves us from mud towers to stone towers. :-)

Where I disagree is with the idea that intelligence will "arise" from a sufficiently large system. I see that as no more likely than new life arising because a kid mixed all the chemicals in his chemistry set together. I know the notion of "emergent behavior" in software systems is popular with a lot of people, but I pretty much reject it outright.

Perhaps the important point is that machine learning has a lot of potential for space applications, but if people think we're on the verge of creating HAL 9000, they're going to miss the things this technology is actually capable of.

[ppnl]

Where I disagree is with the idea that intelligence will "arise" from a sufficiently large system. I see that as no more likely than new life arising because a kid mixed all the chemicals in his chemistry set together. I know the notion of "emergent behavior" in software systems is popular with a lot of people, I pretty much reject it outright.

But do you have an argument for such an opinion? A neuron is just a physical object that obeys physical laws. The Church-Turing thesis suggests that a computer program should be able to emulate the function of a neuron. While emulating a hundred billion neurons with a thousand trillion interconnections is challenging there is no new physics here as far as we can tell. Therefore it seems to be mostly an engineering problem.

The transistor was invented in 1947. It was around the size of a matchbox. In the span of one lifetime we have learned how to put over five billion of them on a little chip of silicone. Before 1947 such a thing would have consumed a large fraction of the electrical output of the entire US and the waste heat would have torched an entire city. I have a dozen of these things operating in my house now. I see no reason that A.I. cannot proceed similarly. A.I. will be embedded in just about every piece of technology we produce. Asking how it can be used in space is like asking how transistors can be used in space.

One idea I like is a full self driving Mars rover that can explore without having to supervise every wheel rotation. In the future a self driving rover that can decide for itself where to go and what samples to collect. Eventually a self driving rover with an effective PhD in geology. I see no hard limit.

[sanman]

Whatever the disadvantages l... machines don't automatically need food, water, and oxygen.
But how can we bootstrap such applications here on Earth, in order to eventually extend them into space?

Pretty sure they "automatically" need electricity.

As to "bootstrapping", that implies sentience.

Sure they need electricity, and there's plenty to be had across much of our solar system (at least the inner parts; for outer parts you need nuclear)

And we sentient humans are the ones to do that bootstrapping of innovation in building machines to carry out activities here on Earth which can be extended to off-world environments.

Maybe we can have semi-autonomous earth-moving machines and construction robots to build things here on Earth, and their core logic can be used in machines that would operate off-world.

[JohnFornaro]

Where I disagree is with the idea that intelligence will "arise" from a sufficiently large system.

Agree.

The process is properly called "Intelligent Design".  [ID]  Design is an iterative process, and the design ends at an arbitrary point when the designer is satisfied with his work.  Even so, there is surely a minimum size of the system, unless one wants to assert that an electron has "intelligence".  We do not yet know how big a "system" can be built either, but your point holds.  A larger system, properly written, would be more likely to generate that intelligent "quality" that is being attempted to replicate.

The intelligence won't "arise" until it has been "created".

[JohnFornaro]

Where I disagree is with the idea that intelligence will "arise" from a sufficiently large system.

But do you have an argument for such an opinion?

He made the argument with the chemistry set analogy.

A neuron is just a physical object that obeys physical laws. The Church-Turing thesis suggests that a computer program should be able to emulate the function of a neuron. While emulating a hundred billion neurons with a thousand trillion interconnections is challenging there is no new physics here as far as we can tell. Therefore it seems to be mostly an engineering problem.

The transistor was invented in 1947. ... I see no reason that A.I. cannot proceed similarly. A.I. will be embedded in just about every piece of technology we produce.

Here, however, you are conflating the size of a physical object with its native or possible intelligence.  No matter how small a transistor you make, it will still be a transistor.

If you want to raise the topic of intelligent matter, here's an interesting essay.

http://www.esalq.usp.br/lepse/imgs/conteudo_thumb/Intelligence-in-Matter.pdf

[JohnFornaro]

Whatever the disadvantages l... machines don't automatically need food, water, and oxygen.
But how can we bootstrap such applications here on Earth, in order to eventually extend them into space?

Pretty sure they "automatically" need electricity.

Sure they need electricity, and there's plenty to be had ...

But wait.  There's also plenty of "food, water, and oxygen" to be had.  What point were you making?

[sanman]

But wait.  There's also plenty of "food, water, and oxygen" to be had.  What point were you making?

I thought we were talking about machines being able to exist and operate independently in space better than fragile human beings can. You responded that machines still need electricity, and I then responded that they can get that electricity out in space (eg. solar). Alright, so there can also be food, water, oxygen out in space away from the Earth, but they're much harder to come by than mere solar flux -- especially the food.

Since machines can be engineered to withstand the rigors of space, while we Earth-coddled human beings can't be similarly re-engineered so easily, then maybe it's more useful to start making more autonomous/semi-autonomous machines for productive use here on Earth, so that they can be transplanted to off-world environment to perform similar functions there.

[Greg Hullender]

But do you have an argument for such an opinion? A neuron is just a physical object that obeys physical laws. The Church-Turing thesis suggests that a computer program should be able to emulate the function of a neuron. While emulating a hundred billion neurons with a thousand trillion interconnections is challenging there is no new physics here as far as we can tell. Therefore it seems to be mostly an engineering problem.

The transistor was invented in 1947. It was around the size of a matchbox. In the span of one lifetime we have learned how to put over five billion of them on a little chip of silicone. Before 1947 such a thing would have consumed a large fraction of the electrical output of the entire US and the waste heat would have torched an entire city. I have a dozen of these things operating in my house now. I see no reason that A.I. cannot proceed similarly. A.I. will be embedded in just about every piece of technology we produce. Asking how it can be used in space is like asking how transistors can be used in space.

I'm not arguing that intelligence is supernatural--just that we don't have the foggiest idea how to engineer such a thing. Nor is it reasonable to suppose that if we just make our computing systems bigger they'll somehow magically become intelligent. I hope you don't think Starship will become intelligent just because it's bigger than Falcon 9! :-)

I think this is a really popular belief among people with vague ideas of how computers work and no idea at all of how artificial intelligence works. Trust me on this--lack of computing power is not what stops current (or foreseeable) AI technology from having human-level intelligence. It's easy to say it's an engineering problem, but it's a problem where it's clear that something critical is missing from our current systems, and we have no idea what that is. That's why I like to make analogy with the ancient Babylonians trying to reach the moon by building higher and higher towers, not realizing that they didn't know enough about the problem to make even a start at it.

Over the rest of the century, AI will make a lot of contributions to society in general and to space travel in specific, but that will be real AI, not some fantasy AI that approximates human intelligence.

[Coastal Ron]

But do you have an argument for such an opinion? A neuron is just a physical object that obeys physical laws. The Church-Turing thesis suggests that a computer program should be able to emulate the function of a neuron. While emulating a hundred billion neurons with a thousand trillion interconnections is challenging there is no new physics here as far as we can tell. Therefore it seems to be mostly an engineering problem.

...

I'm not arguing that intelligence is supernatural--just that we don't have the foggiest idea how to engineer such a thing. Nor is it reasonable to suppose that if we just make our computing systems bigger they'll somehow magically become intelligent.

I remember back in the late 80's a company I was working at, that had LOTS of Phd's, thought that neural networks would solve the A.I. hardware challenge. That was also about the time when "fuzzy logic" was thought to be the next revolution in consumer appliances, for making "smart washers" and such.

Needless to say the hype did not live up to reality, though no doubt we learned more about what we didn't know than what we did know.

Fast forward to fairly recently and we saw a similar boom and bust cycle with A.I. and its various subcategories. Though it looks like the bust cycle is not so bad with A.I., as we have found plenty of applications that can use its limited abilities.

But to your point, it does not yet appear that we understand how to make truly intelligent artificial intelligence, as opposed to smart tools.

So from that standpoint, of course "smart tools" like A.I. can be used for space applications. But I don't think they will be able to solve the challenges we have holding us back from expanding humanity out into space.

[JohnFornaro]

Trust me on this--lack of computing power is not what stops current (or foreseeable) AI technology from having human-level intelligence.

"Trust me on this" -- While you may assert that we [or you?] know the full extent of human intelligence, and the computing power and complexity of the human brain, you and we do not. 

The people who are engineering AI are not, despite their pearl clutching claims to the contrary, trying to create a sentient, reasoning being along the lines of human intelligence.  Rather, they are working more to implement a system of control than anything else.  Musk's recent revelations, and Twitter's instantiation of AI argue for this interpretation.

But your point still holds:

It's easy to say it's an engineering problem, but it's a problem where it's clear that something critical is missing from our current systems, and we have no idea what that is.

[Greg Hullender]

"Trust me on this" -- While you may assert that we [or you?] know the full extent of human intelligence, and the computing power and complexity of the human brain, you and we do not. 

I'm simply saying that scaling up existing systems isn't going to produce human intelligence or anything like it, for the same reasons that scaling up a truck or a rocket isn't going to. People never imagine that a big truck or rocket will suddenly develop its own opinions, because they understand those technologies well enough to see the absurdity of it. But they have almost no understanding of current AI technology. If they did, they'd realize (as all serious people in the field do) that the propositions are equally absurd.

I don't believe you have to know "the full extent of human intelligence" to see this. You just need a broad understanding of the current state of the art in AI.

[edzieba]

I don't believe you have to know "the full extent of human intelligence" to see this. You just need a broad understanding of the current state of the art in AI.

The "current state of the art in AI" is to continue to throw more and more compute at the problem, using techniques (bog standard MLNNs) that were dismissed as dead ends decades ago, and to continue to improve in capability as a result of that increased available compute.

Turing Machines (the majority of current computers) work by being very dumb very quickly, and its turned out that 'AI' can work very well by being very dumb in parallel, sufficient to be just as useful as out UTMs.
Of course, as has happened every single time an AI technique has been adopted into production, it will be dismissed in short order as "not real/true AI", as if the sole goal of AI is to replicate human-level intelligence and ignore the existence and utility of the huge range of other useful intelligences.

[Greg Hullender]

The "current state of the art in AI" is to continue to throw more and more compute at the problem, using techniques (bog standard MLNNs) that were dismissed as dead ends decades ago, and to continue to improve in capability as a result of that increased available compute.

You left out "throw more data at the problem." :-) Lots of stuff that just doesn't work with small training sets does a decent job with hundreds (or thousands) of times the data. Assuming you have the compute power to handle it, of course.

Turing Machines (the majority of current computers) work by being very dumb very quickly, and its turned out that 'AI' can work very well by being very dumb in parallel, sufficient to be just as useful as out UTMs.
Of course, as has happened every single time an AI technique has been adopted into production, it will be dismissed in short order as "not real/true AI", as if the sole goal of AI is to replicate human-level intelligence and ignore the existence and utility of the huge range of other useful intelligences.

Yes, this attitude that "it's not AI if it actually works/is useful" is a big part of the reason why most serious researchers/engineers shun the "AI" label in favor of "machine learning (ML)." When you talk about ML, you make it clear that you're practical. If you talk about AI, you run the risk of people not taking you seriously.

A second factor, though, is that all existing useful implementations work by evading "the AI problem." Very often during design of a project I've had to tell management that we couldn't do something because "we'd have to be able to make the computer think." Over and over, we (and others) found solutions by coming up with clever ways to attack the problem without requiring any thinking on the part of the computer. When people learn what the solution actually was, they're invariably disappointed if they were hoping for "real AI" because it feels like we avoided the problem rather than solving it.

[JohnFornaro]

"Trust me on this" -- While you may assert that we [or you?] know the full extent of human intelligence, and the computing power and complexity of the human brain, you and we do not. 

I'm simply saying that scaling up existing systems isn't going to produce human intelligence or anything like it, for the same reasons that scaling up a truck or a rocket isn't going to. ...

Well, you missed my point, but I see that I didn't convey my point as well as I should have.

Lack of computing power is not what stops current (or foreseeable) AI technology from having human-level intelligence.

"Lack of computing power" is one of the things stopping "forseeable" AI from having human level intelligence.

The AI algos suck because they are partly designed along political narrative lines; because they are poorly analogous to their human intelligence counterparts; because they don't use the same memory type that humans do; [perhaps Bohm's holographic model should be further studied]; AND... they lack the computing power of the human brain and mind.

[Greg Hullender]

"Lack of computing power" is one of the things stopping "forseeable" AI from having human level intelligence.

I suppose it depends on your definition of "foreseeable." To me, that means anything based on technologies I've at least heard people propose--even if they aren't reduced to practice yet. But it doesn't include ridiculous things, like doing an atomic-level simulation of the human nervous system.

Something with real intelligence would have to be based on something different from anything I know of that anyone is working on today. That's all I mean by "foreseeable."

[edzieba]

I'm loath to draw any distinction between 'real' intelligence and a tottering pile of hacks, since a) biological intelligence is a tottering pile of hacks that only mostly works (with legacy efficiency tricks like 'herd mentality' causing problems in modern systems, and the whole thing occasionally getting stuck in an infinite loop if the neurotransmitter balance is slightly off), and b) that gets into Cartesian Dualism (that there is some intangible quality that human minds posses and machines cannot).

[ppnl]

I think there is some serious fuzziness in how people are thinking about this. Let me try to clarify.

First, universal Turing machines are just that. Universal. That means all computers are the same to within a polynomial time complexity. Neglecting memory size and processing speed any problem that one can solve any other can solve. They should also be able to simulate any object or process that exists in the universe. As long as the process isn't quantum in nature they should be able to simulate efficiently in the big O time complexity sense. That means we should be able to simulate brains. Short of trashing the Church/Turing thesis there is no way around this conclusion. People are free to reject the CT thesis but they should say so up front to avoid confusion.

So what do we need to simulate a brain? It is possible that we will need an increase in computing power of a hand full of orders of magnitude. That is challenging but is the easy part. After that all that is left is the software and algorithms. That may be the hard part.

Now, does anyone disagree with the truth and clarity of the above?


 

[Robotbeat]

But do you have an argument for such an opinion? A neuron is just a physical object that obeys physical laws. The Church-Turing thesis suggests that a computer program should be able to emulate the function of a neuron. While emulating a hundred billion neurons with a thousand trillion interconnections is challenging there is no new physics here as far as we can tell. Therefore it seems to be mostly an engineering problem.

...

I'm not arguing that intelligence is supernatural--just that we don't have the foggiest idea how to engineer such a thing. Nor is it reasonable to suppose that if we just make our computing systems bigger they'll somehow magically become intelligent.

I remember back in the late 80's a company I was working at, that had LOTS of Phd's, thought that neural networks would solve the A.I. hardware challenge. That was also about the time when "fuzzy logic" was thought to be the next revolution in consumer appliances, for making "smart washers" and such.

Needless to say the hype did not live up to reality, though no doubt we learned more about what we didn't know than what we did know.

Fast forward to fairly recently and we saw a similar boom and bust cycle with A.I. and its various subcategories. Though it looks like the bust cycle is not so bad with A.I., as we have found plenty of applications that can use its limited abilities.

But to your point, it does not yet appear that we understand how to make truly intelligent artificial intelligence, as opposed to smart tools.

So from that standpoint, of course "smart tools" like A.I. can be used for space applications. But I don't think they will be able to solve the challenges we have holding us back from expanding humanity out into space.

I would have agreed with both of you about a week or three ago. Now, I’m not so sure.

Things like GPT-3 seem remarkably flexible and powerful. It’s just a large language model with some significant limitations, but some of them seem to arise from technical constraints that could be overcome. Things like poor logic (ironic), the memory of a good fish (also ironic), etc.

Interact with chat.OpenAI.com for a few hours, see what others are doing, and then tell me how confident you are that a sufficiently large model, with tweaks in the next few years or decades, couldn’t approximate the kind of general intelligence that humans and animals exhibit.

I’m much less certain than I was just a month ago about us being unable to develop a general intelligence using sufficiently large neural net based models. Even more, before a year ago and the Dall-E 2 and stable diffusion stuff came out. They had this LLM take the SAT and it got over 1000. They had it take an IQ test and it scored 83.

Try it, don’t just read about it.

[DanClemmensen]

I think there is some serious fuzziness in how people are thinking about this. Let me try to clarify.

First, universal Turing machines are just that. Universal. That means all computers are the same to within a polynomial time complexity. Neglecting memory size and processing speed any problem that one can solve any other can solve. They should also be able to simulate any object or process that exists in the universe. As long as the process isn't quantum in nature they should be able to simulate efficiently in the big O time complexity sense. That means we should be able to simulate brains. Short of trashing the Church/Turing thesis there is no way around this conclusion. People are free to reject the CT thesis but they should say so up front to avoid confusion.

So what do we need to simulate a brain? It is possible that we will need an increase in computing power of a hand full of orders of magnitude. That is challenging but is the easy part. After that all that is left is the software and algorithms. That may be the hard part.

Now, does anyone disagree with the truth and clarity of the above?

A Turing machine is a clocked digital system. Digital computers are a subset of all possible computers, so a UTM cannot simulate all possible computers. The two notable extensions are analog computers and non-clocked digital systems. Biological systems do not appear to be clocked, and they appear to incorporate analog components. It is not clear that a UTM can simulate a system with analog components in polynomial time.

I personally believe that "intelligence" will end up getting implemented in "traditional" computers (i.e., clocked digital logic). By "intelligence" I mean a system that passes an extended version of the Turing test. However, if this does not happen, you don't need to invoke quantum theory. You can add analog elements instead.

[Robotbeat]

I’m pretty sure ChatGPT could pass the Turing test if implemented and judged by most people. The most obviously machine aspect of ChatGPT is that the responses are much faster than a human.

ChatGPT simulates a dumb human (but really good at English class assignments) really well.

[ppnl]

I think there is some serious fuzziness in how people are thinking about this. Let me try to clarify.

First, universal Turing machines are just that. Universal. That means all computers are the same to within a polynomial time complexity. Neglecting memory size and processing speed any problem that one can solve any other can solve. They should also be able to simulate any object or process that exists in the universe. As long as the process isn't quantum in nature they should be able to simulate efficiently in the big O time complexity sense. That means we should be able to simulate brains. Short of trashing the Church/Turing thesis there is no way around this conclusion. People are free to reject the CT thesis but they should say so up front to avoid confusion.

So what do we need to simulate a brain? It is possible that we will need an increase in computing power of a hand full of orders of magnitude. That is challenging but is the easy part. After that all that is left is the software and algorithms. That may be the hard part.

Now, does anyone disagree with the truth and clarity of the above?

A Turing machine is a clocked digital system. Digital computers are a subset of all possible computers, so a UTM cannot simulate all possible computers. The two notable extensions are analog computers and non-clocked digital systems. Biological systems do not appear to be clocked, and they appear to incorporate analog components. It is not clear that a UTM can simulate a system with analog components in polynomial time.

I personally believe that "intelligence" will end up getting implemented in "traditional" computers (i.e., clocked digital logic). By "intelligence" I mean a system that passes an extended version of the Turing test. However, if this does not happen, you don't need to invoke quantum theory. You can add analog elements instead.

Asynchronous digital systems have the same computational power as clocked digital systems. This becomes obvious when you consider that they both use the same universal logic gates and the same boolean algebra. Any asynchronous digital circuit can immediately be implemented as a clocked digital circuit or simply programmed as a computer program. Clock signals are really just a kludge to solve some engineering problems with asynchronous circuits. The problem is the data paths from input to output may be of very different lengths. This creates race conditions and for a large circuit the output may never be valid. A timing signal can latch the output to a valid state until the next state is valid.

A few decades ago they started experimenting with asynchronous sections in microprocessors to make them a little faster and use less energy. I think all modern processors have asynchronous sections. But it is just an engineering kludge that gets a little more performance.

The problem with analog computers is they really suck. There is no way to control the noise so the complexity of the calculations is severely limited. As a practical matter you will never calculate Pi to a thousand digits for example. And a digital computer can do anything that they can do faster and better. Floating point arithmetic is a thing on digital computers and they can have as many digits of accuracy as you want. I don't know if any analog computer is still in use anywhere in the world today. They are obsolete exactly because a digital computer can do everything that they can do.

Quantum computers arguably could offer an exponential speedup on certain limited types of computations. But it is hard to see how a warm wet brain could be a quantum computer.

I expect algorithmic progress decade on decade so that in 70 or so years we will look back on today the same way that we today look back on the invention of the transistor. Future computers used in AI may not look much like our computers but they will be universal Turing machines.     

[Robotbeat]

I came across this interesting article about how AI is being used to conjure up completely new proteins very quickly.
https://www.nature.com/articles/d41586-022-02947-7


It occurred to me that this could enable all sorts of designer bugsorganisms for ISRU purposes, and maybe even terraforming.

Perhaps we could have bugsorganisms designed to survive the Martian day/night cycle, which would come alive during the day, to perform useful conversion of natural resources, like through Sabatier or whatever.

Could we even use AI to design complex ecosystems of organisms that would cope with the existing Mars conditions while working to transform the environment into one that's more human-friendly?

Yes, AI could potentially be used in a variety of ways for space applications, including for ISRU (in-situ resource utilization) and terraforming. For example, AI could be used to design and optimize proteins for specific conditions, such as the extreme temperatures and radiation levels on Mars. These proteins could be used in organisms that are capable of surviving and thriving on the Martian surface, and potentially even converting natural resources into usable materials.

In addition, AI could be used to design complex ecosystems of organisms that could adapt to the existing conditions on Mars and work to transform the environment over time. This could involve developing algorithms that can simulate and predict the behavior and interactions of different species, and optimize their performance in a given environment.

Overall, AI has the potential to greatly accelerate and improve our ability to utilize and explore space, by enabling us to design and optimize complex systems that can survive and thrive in challenging environments.

Tell me more, with specifics on the engineering involved.

In terms of specific engineering applications, AI could be used in a variety of ways to support space exploration and utilization. Some examples include:

* Designing and optimizing spacecraft systems and components: AI algorithms could be used to design and optimize spacecraft systems and components, such as propulsion systems, power systems, and communication systems, to improve performance and reduce costs. For example, AI could be used to model and simulate the behavior of different spacecraft components, and identify optimal configurations and designs that meet specified performance criteria.

* Planning and executing space missions: AI could be used to plan and execute space missions, by developing algorithms that can generate and evaluate potential mission plans, and identify optimal strategies for achieving mission objectives. This could involve simulating the behavior of spacecraft and other systems, and optimizing their performance to achieve desired outcomes.

* Analyzing data from space missions: AI could be used to analyze data from space missions, such as sensor data from spacecraft, to identify patterns and trends, and make predictions about future events. This could involve developing algorithms that can process and interpret large amounts of data, and extract useful insights and knowledge from it.

* Developing and deploying intelligent robots: AI could be used to develop and deploy intelligent robots that are capable of autonomously exploring and operating in space. This could involve developing algorithms that enable robots to perceive and interact with their environment, and make decisions based on that information. For example, AI could be used to enable robots to navigate and manipulate objects, and perform tasks such as collecting samples or repairing equipment.

Overall, AI has the potential to greatly enhance our ability to explore and utilize space, by enabling us to design and optimize complex systems, plan and execute missions, analyze data, and deploy intelligent robots.

[ppnl]

I’m pretty sure ChatGPT could pass the Turing test if implemented and judged by most people. The most obviously machine aspect of ChatGPT is that the responses are much faster than a human.

ChatGPT simulates a dumb human (but really good at English class assignments) really well.

The sad fact is that some really simple chat-bots has passed a Turing test from people who should know better. Here is my interaction:

ppnl: Hi!
AI: Hello! How can I help you today?
ppnl: What is 2+2?
AI: 2 + 2 = 4. Is there anything else you would like to know?

It correctly parsed the question and calculated the answer. It wouldn't be hard to write a program to do this. But a program that learned for itself how to do this is more impressive.

ppnl: Why did the chicken cross the road?
AI: There are many different jokes and puns that use this question as a setup, but the most common answer is "to get to the other side." This answer is a play on the idea that chickens are not particularly intelligent animals, so they would not have any complex motivation for crossing the road. Some other possible answers to this question include:

    To escape from a predator
    To reach a food source on the other side of the road
    To explore their surroundings
    To join other chickens on the other side of the road
    As a dare, or as a way to show off to other chickens

Ultimately, the answer to this question will depend on the context in which it is asked and the person who is answering it.

Ok, reasonably impressive. It covered the context as a joke and as a serious question. Not sure how it learned this from training text.

ppnl: How many chickens does it take to screw in a light bulb?
AI: I'm sorry, but I am not capable of performing physical tasks or manipulating objects. I am a large language model trained by OpenAI, so I can provide information and answer questions, but I cannot perform actions in the real world.

A clear miss. It missed the joke context. Worse it misunderstood that I was asking it to perform a physical task.

ppnl: How many light bulbs does it take to screw in a chicken?
AI: As I mentioned before, I am not capable of performing physical tasks or manipulating objects. I am a language model, so I can only provide information and answer questions based on the text that I have been trained on. It is not possible for me to screw in a chicken, or for any number of light bulbs to do so. Can I help you with anything else?

It is impressive that that it made the connection to the previous question. But it still seems to think I am asking it to do something. It does say light bulbs cannot screw in chickens.

ppnl: Can a person screw in a light bulb?
AI: Yes, a person is capable of screwing in a light bulb. In most cases, this involves using a screwdriver to            turn the base of the light bulb, which is threaded, in a clockwise direction until it is securely attached to the   light fixture. Some light fixtures may require a different tool or method for changing the light bulb, so it is important to refer to the instructions for the specific fixture being used.

Screwdriver? I can almost see how training on text might produce this strange answer. It clearly fails my Turing test but then many people I talk to online fail my Turing test so...

I remember when the Hubble space telescope was being built they were having trouble with the software that would prevent the telescope from inadvertently pointing at the sun or other bright objects. They were countless million over budget. Now I wonder if you could just simulate the situation and let it teach itself to keep safe. Would NASA trust such a program?



 

[DanClemmensen]

I think there is some serious fuzziness in how people are thinking about this. Let me try to clarify.

First, universal Turing machines are just that. Universal. That means all computers are the same to within a polynomial time complexity. Neglecting memory size and processing speed any problem that one can solve any other can solve. They should also be able to simulate any object or process that exists in the universe. As long as the process isn't quantum in nature they should be able to simulate efficiently in the big O time complexity sense. That means we should be able to simulate brains. Short of trashing the Church/Turing thesis there is no way around this conclusion. People are free to reject the CT thesis but they should say so up front to avoid confusion.

So what do we need to simulate a brain? It is possible that we will need an increase in computing power of a hand full of orders of magnitude. That is challenging but is the easy part. After that all that is left is the software and algorithms. That may be the hard part.

Now, does anyone disagree with the truth and clarity of the above?

A Turing machine is a clocked digital system. Digital computers are a subset of all possible computers, so a UTM cannot simulate all possible computers. The two notable extensions are analog computers and non-clocked digital systems. Biological systems do not appear to be clocked, and they appear to incorporate analog components. It is not clear that a UTM can simulate a system with analog components in polynomial time.

I personally believe that "intelligence" will end up getting implemented in "traditional" computers (i.e., clocked digital logic). By "intelligence" I mean a system that passes an extended version of the Turing test. However, if this does not happen, you don't need to invoke quantum theory. You can add analog elements instead.

Asynchronous digital systems have the same computational power as clocked digital systems. This becomes obvious when you consider that they both use the same universal logic gates and the same boolean algebra. Any asynchronous digital circuit can immediately be implemented as a clocked digital circuit or simply programmed as a computer program. Clock signals are really just a kludge to solve some engineering problems with asynchronous circuits. The problem is the data paths from input to output may be of very different lengths. This creates race conditions and for a large circuit the output may never be valid. A timing signal can latch the output to a valid state until the next state is valid.

A few decades ago they started experimenting with asynchronous sections in microprocessors to make them a little faster and use less energy. I think all modern processors have asynchronous sections. But it is just an engineering kludge that gets a little more performance.

The problem with analog computers is they really suck. There is no way to control the noise so the complexity of the calculations is severely limited. As a practical matter you will never calculate Pi to a thousand digits for example. And a digital computer can do anything that they can do faster and better. Floating point arithmetic is a thing on digital computers and they can have as many digits of accuracy as you want. I don't know if any analog computer is still in use anywhere in the world today. They are obsolete exactly because a digital computer can do everything that they can do.

Quantum computers arguably could offer an exponential speedup on certain limited types of computations. But it is hard to see how a warm wet brain could be a quantum computer.

I expect algorithmic progress decade on decade so that in 70 or so years we will look back on today the same way that we today look back on the invention of the transistor. Future computers used in AI may not look much like our computers but they will be universal Turing machines.   

I was not attempting to claim that the alternatives (async digital and analog) are useful. I'm just pointing out that the flat assertion that a UTM can do any computing task is not strictly true in theory. Furthermore, biological intelligence appears to use both. Note that async digital is continuous in the time domain, so it is basically analog in the time domain.

[Greg Hullender]

So what do we need to simulate a brain? It is possible that we will need an increase in computing power of a hand full of orders of magnitude. That is challenging but is the easy part. 

If you're talking about simulating a brain on the atomic level, then you need a lot more than a "handful of orders of magnitude," unless your hands are unusually large. :-) It's not clear to me that we could set up such a simulation anyway--simulating single protein molecules is challenging at the moment.

If you're imagining that we understand the brain well enough to set up a simulation of it at something other than the atomic level, I think you're seriously misinformed.

[Greg Hullender]

Interact with chat.OpenAI.com for a few hours, see what others are doing, and then tell me how confident you are that a sufficiently large model, with tweaks in the next few years or decades, couldn’t approximate the kind of general intelligence that humans and animals exhibit.

I played with it a bit just today, in fact. It's quite impressive for what it is, which is mostly a question-answering system.

There is a certain tendency to look at results that amaze us and disregard ones that are disappointing. For me, most of the responses were of the form "I can't help you with that." Some were very impressive indeed, but a couple were way off base. For example:

I tried making a statement instead of asking a question, "The sun is the nearest star to Earth," and it assured me that a) the sun is not a star and b) the sun is a star.

I asked it how to take a square root, and the answer was mostly nonsense with one useful fact--that Newton's method would do it. You'd still have to look that up on Wikipedia, but I can't claim the answer was 100% useless.

I think the best way to think of it is that it's a really clever database lookup tool. A system like this with a database of scientific/technical material might be very useful. But, no, it doesn't lead me to think it might develop intelligence just from more data or more processing power. It'll just give better answers to more questions and do it faster and for more people at the same time. There really is more to intelligence than that. :-)

For me, the most impressive thing is that it doesn't use a formal language model; it has developed a language model from large blocks of text without having any idea what a noun or a verb is. (Or so I'm led to believe; I haven't seen a formal description of the system yet.)

[ppnl]

I was not attempting to claim that the alternatives (async digital and analog) are useful. I'm just pointing out that the flat assertion that a UTM can do any computing task is not strictly true in theory. Furthermore, biological intelligence appears to use both. Note that async digital is continuous in the time domain, so it is basically analog in the time domain.

And I am flatly asserting that a UTM can do any computational task that is doable. It can simulate asynchronous or analog systems to any level of fidelity you want. It can almost always do it cheaper and easier than the analog or asynchronous system that it is replacing.
 

But unless you also claim that this is relevant to the power of AI then it might also be off topic.

[ppnl]

So what do we need to simulate a brain? It is possible that we will need an increase in computing power of a hand full of orders of magnitude. That is challenging but is the easy part. 

If you're talking about simulating a brain on the atomic level, then you need a lot more than a "handful of orders of magnitude," unless your hands are unusually large. :-) It's not clear to me that we could set up such a simulation anyway--simulating single protein molecules is challenging at the moment.

It is possible in a thought experiment but as a reality? No.

If you're imagining that we understand the brain well enough to set up a simulation of it at something other than the atomic level, I think you're seriously misinformed.

No we don't understand the brain well enough to simulate it at the atomic level or any other level. But there is no reason that we cannot come to understand it at a higher level than atoms. Neural nets in computers is a good start but far from the full story. For example we don't know how the brain does back-propagation. It is a mystery. It has a solution.

[JohnFornaro]

I'm loath to draw any distinction between 'real' intelligence and a tottering pile of hacks...

Before I ask for ranch dressing with my word salad, when I read this far, I thought of a colony of ants. Surely they exhibit "real" intelligence, despite not using fire, the wheel, the lever, XOR logic gates and such.

Time for coffee.

[JohnFornaro]

Now, does anyone disagree with the truth and clarity of the above?

While all that does sound truthy to me, it is not all "clear", since "simulating" a human brain, other than by trad genetic activities, is not "easy" by any stretch.  The computing power is inadequate, and the algos are faulty.

AI's applicability for space applications is, well, a no-brainer.  There are any number of specific applications that AI can profitably be applied.  Find a problem, and write the literature, errrr, software necessary to solve the problem.

Perhaps over time sentience will occur, but there are hard space problems needing solution right now.

[JohnFornaro]

But it is hard to see how a warm wet brain could be a quantum computer.

Allow me to help out by first pointing out that ALL wet brains that we know of exist in the quantum foam.  Consider also the fact that the neuron is very close to the scale of quantum events.  Several billion years of evolution within a quantum universe cannot be disposed of with a wave of the hand, simply because the connection is not well known. 

Remember how they used to talk about "junk" DNA?  I was always like, it's there for a reason; just because you don't know how it works is not excuse for calling it "junk".

Now, that we exist in a quantum universe doesn't solve the problem of simulating a sentient intelligence, but it does at least help you understand where the simulation needs to take place.

[JohnFornaro]

How many light bulbs does it take to screw in a chicken?

Bingo.

[edzieba]

I was not attempting to claim that the alternatives (async digital and analog) are useful. I'm just pointing out that the flat assertion that a UTM can do any computing task is not strictly true in theory. Furthermore, biological intelligence appears to use both. Note that async digital is continuous in the time domain, so it is basically analog in the time domain.

And I am flatly asserting that a UTM can do any computational task that is doable. It can simulate asynchronous or analog systems to any level of fidelity you want. It can almost always do it cheaper and easier than the analog or asynchronous system that it is replacing.
 

But unless you also claim that this is relevant to the power of AI then it might also be off topic.

Remember that a UTM is a computational architecture (or at least, a reductive basis for actually reasonably implementable architectures like Von Neumann), not a collection of hardware. It can be an electronic digital computer, but it could be an optomechanical computer running up and down a stupendous strip of paper and would be computationally equivalent. A UTM can simulate other TMs, but cannot simulate all possible computer architectures (digital or otherwise). Likewise, the same electronics components that can be used by build a Turing-Complete computer can also be used to built a computer for Non-Turing Computation.

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving, and without positing some 'inscrutable particle of consciousness' at play or similar supernatural element, that means there exists computational architectures implementable in out universe that are not UTMs.

[Robotbeat]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

[edzieba]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

[Robotbeat]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

Nope. Humans are not currently provably able to solve any problem that a UTM is provably unable to solve. Provide an exact counter-example and citation for your assertion humans can do better than an arbitrarily UTM.

This is considered a somewhat open problem, meaning no one has ever produced an example of a problem a human could solve that a UTM cannot, nor (more generally) has any physical process been found which a UTM cannot simulate.

Discussion here:
https://en.wikipedia.org/wiki/Halting_problem

[Twark_Main]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable.

Determining decidability for some subset of problems is not, by itself, undecidable. Humans can do that.

Determining decidability for all possible problems is equivalent to the halting problem, so it is undecidable. But humans can't do that, because ultimately they'd be following some algorithm, and such an algorithm is provably impossible.

One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output).

That's not the definition of a "decidable problem."

That is not a limitation possessed by humans.

No.  [/Jim]

[Barley]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

No, we don’t.

Mathematics like Cantor's diagonal argument and Godel's incompleteness theorem are compatible with a UTM.  If you're talking about "knowing" things without proof, ask ChatAPT if a peregrine falcon is a mammal.

[Greg Hullender]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

There is a really deep philosophical rathole here, that I think we'd do best not to get into. We're already very far afield from the question in the original post, which actually does have some merit.

[leovinus]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

There is a really deep philosophical rathole here, that I think we'd do best not to get into. We're already very far afield from the question in the original post, which actually does have some merit.

My thoughts exactly.

[Robotbeat]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

There is a really deep philosophical rathole here, that I think we'd do best not to get into. We're already very far afield from the question in the original post, which actually does have some merit.

It’s factually incorrect to assert what he asserted (as established fact), though. If he were right, he’d be able to provide an example (and probably win some international mathematics or computer science award).

It’s fine to claim it as *maybe possible* that humans could exceed an UTM in decidability, using some sort of non-algorithmic manner that can’t be written down in any formal process. That can’t be ruled out so far (and I’m holding myself back here…), but he asserted that it was KNOWN that humans can do something UTMs are incapable of, which is factually incorrect.

[Robotbeat]

I came across this interesting article about how AI is being used to conjure up completely new proteins very quickly.
https://www.nature.com/articles/d41586-022-02947-7


It occurred to me that this could enable all sorts of designer bugsorganisms for ISRU purposes, and maybe even terraforming.

Perhaps we could have bugsorganisms designed to survive the Martian day/night cycle, which would come alive during the day, to perform useful conversion of natural resources, like through Sabatier or whatever.

Could we even use AI to design complex ecosystems of organisms that would cope with the existing Mars conditions while working to transform the environment into one that's more human-friendly?

Yes, AI could potentially be used in a variety of ways for space applications, including for ISRU (in-situ resource utilization) and terraforming. For example, AI could be used to design and optimize proteins for specific conditions, such as the extreme temperatures and radiation levels on Mars. These proteins could be used in organisms that are capable of surviving and thriving on the Martian surface, and potentially even converting natural resources into usable materials.

In addition, AI could be used to design complex ecosystems of organisms that could adapt to the existing conditions on Mars and work to transform the environment over time. This could involve developing algorithms that can simulate and predict the behavior and interactions of different species, and optimize their performance in a given environment.

Overall, AI has the potential to greatly accelerate and improve our ability to utilize and explore space, by enabling us to design and optimize complex systems that can survive and thrive in challenging environments.

Tell me more, with specifics on the engineering involved.

In terms of specific engineering applications, AI could be used in a variety of ways to support space exploration and utilization. Some examples include:

* Designing and optimizing spacecraft systems and components: AI algorithms could be used to design and optimize spacecraft systems and components, such as propulsion systems, power systems, and communication systems, to improve performance and reduce costs. For example, AI could be used to model and simulate the behavior of different spacecraft components, and identify optimal configurations and designs that meet specified performance criteria.

* Planning and executing space missions: AI could be used to plan and execute space missions, by developing algorithms that can generate and evaluate potential mission plans, and identify optimal strategies for achieving mission objectives. This could involve simulating the behavior of spacecraft and other systems, and optimizing their performance to achieve desired outcomes.

* Analyzing data from space missions: AI could be used to analyze data from space missions, such as sensor data from spacecraft, to identify patterns and trends, and make predictions about future events. This could involve developing algorithms that can process and interpret large amounts of data, and extract useful insights and knowledge from it.

* Developing and deploying intelligent robots: AI could be used to develop and deploy intelligent robots that are capable of autonomously exploring and operating in space. This could involve developing algorithms that enable robots to perceive and interact with their environment, and make decisions based on that information. For example, AI could be used to enable robots to navigate and manipulate objects, and perform tasks such as collecting samples or repairing equipment.

Overall, AI has the potential to greatly enhance our ability to explore and utilize space, by enabling us to design and optimize complex systems, plan and execute missions, analyze data, and deploy intelligent robots.

This entire post (besides the quotes), by the way, is the output of ChatGPT when provided the quotes as prompts.

The first one is primarily just restating the (larger than necessary) input prompt which I took from the first post in this thread, but the second half is an impressive synthesis of exactly the answer to this thread’s question. JohnFornarno “liked” it, even. I guess that counts as a kind of informal “Turing Test.”

[Greg Hullender]

This entire post (besides the quotes), by the way, is the output of ChatGPT when provided the quotes as prompts.

The first one is primarily just restating the (larger than necessary) input prompt which I took from the first post in this thread, but the second half is an impressive synthesis of exactly the answer to this thread’s question. JohnFornarno “liked” it, even. I guess that counts as a kind of informal “Turing Test.”

Bravo! Well played!

I'll confess I just skimmed over the bullet list at the bottom. They all sounded plausible, but the challenge would be in finding a way to "featurize" the input data in such a way that it actually enabled one or another AI algorithm to work. (AI can't, in general, handle raw data as input.) The other thought I had (which I've had about a lot of ChatGPT's output) is that it was way too wordy. (Or TLDR, as the young people say.) :-) My eyes glazed over, and I didn't read it very closely. So let's take a close read.

"Designing and optimizing spacecraft systems and components" seems really far-fetched. I'm not aware of any issues there that need more than linear or quadratic programming--if that. (But I'd be interested to learn otherwise.) Same comment for "Planning and executing space missions." Neither of these is an area where I expect to see AI make an impact.

"Analyzing data from space missions" is where I'd expect to see the biggest impact. As a target for AI, it has a lot going for it. First, it's not mission-critical. If it doesn't work, it spoils someone's paper--it doesn't waste billions of dollars and/or cost people their lives. Second, there is a lot of data to work with. Contrast the first two proposed areas, where there really isn't a lot of data. Finally, the need find information in vast tracts of data is a real problem that isn't easily solved by other means. (You should never use AI when there's a conventional solution.)  "Analyzing data" is definitely a problem that seems tailor-made for modern AI. But I think we all already knew that.

"Developing and deploying intelligent robots." Here's where ChatGPT shows its weaknesses. The paragraph isn't about designing and deploying: it's about navigation systems. (How would AI help you deploy anything?!) Navigation is definitely an area where there's been a lot of AI work here on the ground, and the lack of pedestrians, road rules, and other vehicles on other planets makes the problem even simpler (although the lack of roads makes it harder).  The question is, does it make it so much simpler that you don't need AI to do it?

I think this illustrates where ChatGPT could make a big difference. It did a nice job of picking a set of plausible areas that could then be the basis of intelligent discussion. Like a database search, but through unstructured data. Augmenting what people do, but not replacing it.

[Robotbeat]

…well, it can probably replace what SOME people do… 🤣

[Robotbeat]

I should’ve told ChatGPT to make it a lot more concise. It’s also very good at producing a “tl:dr” version of some input text.

[Greg Hullender]

…well, it can probably replace what SOME people do… 🤣

It's definitely a first-rate bullsh*t generator. It can definitely replace those people who have to produce dreary reports that seem reasonable but which no one actually reads.

Now we need an AI solution that identifies this type of output. Perhaps by taking pieces and relating them to existing online documents. "Highlight the portions that are potentially original."

Sigh. We may have to have that soon . . .

[DanClemmensen]

…well, it can probably replace what SOME people do… 🤣

It's definitely a first-rate bullsh*t generator. It can definitely replace those people who have to produce dreary reports that seem reasonable but which no one actually reads.

Now we need an AI solution that identifies this type of output. Perhaps by taking pieces and relating them to existing online documents. "Highlight the portions that are potentially original."

Sigh. We may have to have that soon . . .

It should also work really well for press releases. PR flacks should be trying to re-train for more productive work. this may extend to other marketing jobs.

[Robotbeat]

…well, it can probably replace what SOME people do… 🤣

It's definitely a first-rate bullsh*t generator. It can definitely replace those people who have to produce dreary reports that seem reasonable but which no one actually reads.

Now we need an AI solution that identifies this type of output. Perhaps by taking pieces and relating them to existing online documents. "Highlight the portions that are potentially original."

Sigh. We may have to have that soon . . .

It’s pretty good at summarizing, so that may help:

https://beta.openai.com/examples/default-tldr-summary

[JohnFornaro]

This entire post (besides the quotes), by the way, is the output of ChatGPT when provided the quotes as prompts.

The first one is primarily just restating the (larger than necessary) input prompt which I took from the first post in this thread, but the second half is an impressive synthesis of exactly the answer to this thread’s question. JohnFornarno “liked” it, even. I guess that counts as a kind of informal “Turing Test.”

Bravo! Well played!

I'll confess I just skimmed over the bullet list at the bottom. They all sounded plausible

Turns out that I "like" "plausible" lists.  Ask it if it likes my ring station.  And who's "Fornarno"?

How would AI help you deploy anything?!

  By designing, building, and operating the deployment apparatus.

[Robotbeat]

I wasn’t trying to make a dig at you, John. ChatGPT is pretty good at summarizing topics.

[ppnl]

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving

No, we don’t.

We can, because we can conclude a problem's solution is undecidable. One of the fundamental tenets of a UTM is that it can solve any decidable problem (any algorithm that produces an output). That is not a limitation possessed by humans.

Treading lightly here because the fun police are threatening...

You should read "Gödel, Escher, Bach: an Eternal Golden Braid" by Douglas Hofstadter to see why you are almost certainly wrong in this claim.

I agree that there are deep and profoundly interesting philosophical issues with uncomputability that are off topic for this thread. But the base claim and its simple refutation seem relevant. 

[ppnl]

But it is hard to see how a warm wet brain could be a quantum computer.

Allow me to help out by first pointing out that ALL wet brains that we know of exist in the quantum foam.  Consider also the fact that the neuron is very close to the scale of quantum events.  Several billion years of evolution within a quantum universe cannot be disposed of with a wave of the hand, simply because the connection is not well known.

All objects in the universe exist in a quantum fog. The atoms, molecules and tiny conglomerate structures exist close to the scale of quantum events. That just isn't good enough.

In order to have a macro scale quantum phenomena you need quantum coherence. That is the quantum bits have to interact with each other without interacting with the rest of their environment. That is about as impossible as actually performing Schrödinger's cat experiment. You should look into the subject of decoherence to see how the macro scale world arises from the quantum foam.

 

Remember how they used to talk about "junk" DNA?  I was always like, it's there for a reason; just because you don't know how it works is not excuse for calling it "junk".

Junk DNA is junk in the sense that it is noncoding. Some of it can have structural uses but a lot of it is clearly broken genes, old viral remains and such. About half of the junk in humans is jumping genes that can break other genes around them. This may produce variations for natural selection to operate on but most of those variations are cancers and genetic diseases. Evolution sucks.

But the fun police are probably going to label this off topic so we should move on.

Now, that we exist in a quantum universe doesn't solve the problem of simulating a sentient intelligence, but it does at least help you understand where the simulation needs to take place.

It does not seem relevant at all as there are no large scale coherent phenomena in the brain. I really like the idea of a quantum brain and would be glad to be proven wrong. But the judgment of physics is harsh.

[ppnl]

I was not attempting to claim that the alternatives (async digital and analog) are useful. I'm just pointing out that the flat assertion that a UTM can do any computing task is not strictly true in theory. Furthermore, biological intelligence appears to use both. Note that async digital is continuous in the time domain, so it is basically analog in the time domain.

And I am flatly asserting that a UTM can do any computational task that is doable. It can simulate asynchronous or analog systems to any level of fidelity you want. It can almost always do it cheaper and easier than the analog or asynchronous system that it is replacing.
 

But unless you also claim that this is relevant to the power of AI then it might also be off topic.

Remember that a UTM is a computational architecture (or at least, a reductive basis for actually reasonably implementable architectures like Von Neumann), not a collection of hardware. It can be an electronic digital computer, but it could be an optomechanical computer running up and down a stupendous strip of paper and would be computationally equivalent. A UTM can simulate other TMs, but cannot simulate all possible computer architectures (digital or otherwise). Likewise, the same electronics components that can be used by build a Turing-Complete computer can also be used to built a computer for Non-Turing Computation.

We also have the existence-proof that humans can sole Halting problems that UTMs are fundamentally incapable of solving, and without positing some 'inscrutable particle of consciousness' at play or similar supernatural element, that means there exists computational architectures implementable in out universe that are not UTMs.

If there were some other type of device that could calculate a function that a UTM could not then it would be what is called a hypercomputer. There has been an extensive search both physically and theoretically and no hint of such a think has been found. If humans were known to be able to solve the halting problem then we would be such a hypercomputer.

In computer complexity theory hypercomputers are called oracles and are used as mathematical devices to make arguments about the time/memory complexity of different problems. For example if I assume I have a device that can solve the halting problem then I can explore what other problems it allows me to solve. But no such oracles actually exist as far as we know.

So no, UTMs are universal in the sense that no other device of any type can produce results that a UTM can't as far as we can see. And people have tried really really hard. It really is well explored territory and there does not seem to be any way to exceed UTMs. Quantum computers come the closest in that it may exceed the extended time version of the Church-Turing thesis. As a result they may be able to factor large numbers in polynomial time.

[JohnFornaro]

But it is hard to see how a warm wet brain could be a quantum computer.

Allow me to help out by first pointing out that ALL wet brains that we know of exist in the quantum foam...

All objects in the universe exist in a quantum fog. The atoms, molecules and tiny conglomerate structures exist close to the scale of quantum events. That just isn't good enough.

Existence is not good enough for ya, eh?

You should look into the subject of [quantam] decoherence to see how the macro scale world arises from the quantum foam.

Been there, done that.  As long as we're using the word "should", perhaps you should not imply that you know all.  Quantum "supremacy" much?  We simply do not know how the macro scale world arises.  Go ahead and tell yourself the words you want to hear.

https://en.wikipedia.org/wiki/Quantum_supremacy
 

Remember how they used to talk about "junk" DNA?  ... it's there for a reason; just because you don't know how it works is not excuse for calling it "junk".

Junk DNA is junk in the sense that it is noncoding. ... Evolution sucks.

Now you're just looking for clicks.

Now, that we exist in a quantum universe doesn't solve the problem of simulating a sentient intelligence, but it does at least help you understand where the simulation needs to take place.

It does not seem relevant at all as there are no large scale coherent phenomena in the brain. I really like the idea of a quantum brain and would be glad to be proven wrong. But the judgment of physics is harsh.

Are you "the physics" as Pfauci is "the science"?  You are correct, IMO, in that Hoffstader's book does a good job discussing these things.

Back to the OP. 

That plausible list that RobotBeet [sorry] offered up thread is very good.  Of course it is incomplete, and no one on this thread yet has addressed its components other than Greg, who takes a first stab at the effort.

[Greg Hullender]

To try and get back on topic, let me summarize a few rules that I think determine when a problem is a suitable candidate for an AI solution.

1) There is no other way to solve it. AI is very, very hard. You don't want to go that way unless you absolutely have to.

2) The problem does not require perfection. A system that's wrong 10% of the time would be an absolute disaster for many applications, but it's quite difficult to design a system that's that accurate.

3) The problem is worth doing. A surprising number of people pick problems that look to be well-suited for a particular algorithm, but they never ask whether there's any value to solving that problem.

An ideal problem is one where the computer can amplify the work of human beings. For example, humans can sift through data returned from a mission looking for (say) white dwarf stars, but doing this by hand might take decades (or centuries). An AI solution that made judgments on the easy cases and passed the hard ones to the human judges would be very valuable, even if it initially passed on 90% of the candidates, provided it was at least as accurate as a human being on the other 10%. If you had a team of ten people, adding software like that would be like hiring another team member. Over time, you could work to improve the software. If it you it to where the rejection rate was 10%, it would be like multiplying your team from 10 to 100. The US Postal Service did exactly this with their software to read zip codes from envelopes.

Are there any space applications that seem to fit the bill?

[ppnl]

Quantum "supremacy" much?  We simply do not know how the macro scale world arises.  Go ahead and tell yourself the words you want to hear.

https://en.wikipedia.org/wiki/Quantum_supremacy

For what its worth I hope you are correct that the brain is a quantum device. It would solve some philosophical problems with consciousness, make quantum computers far easier to construct than we currently believe and greatly advance the field of computational complexity. Also it would greatly increase the usefulness of AI in space.

But the problem of coherence time is pretty devastating. In any case we should perhaps move on before the fun police arrive.

[JohnFornaro]

Edit:12-16-22:  LQP = Linear Quadratic Programming. Not LPQ.  Somebody remind me to watch my P's and Q's.]

This entire post (besides the quotes), by the way, is the output of ChatGPT when provided the quotes as prompts. ... I guess that counts as a kind of informal “Turing Test.”

Bravo! Well played!

I'll confess I just skimmed over the bullet list at the bottom. They all sounded plausible, but ... my eyes glazed over, and I didn't read it very closely. So let's take a close read.

"Designing and optimizing spacecraft systems and components" seems really far-fetched. I'm not aware of any issues there that need more than linear or quadratic programming [LQP]--if that. (But I'd be interested to learn otherwise.)

I approach this issue [the utility of AI] and similar issues [the 'why' of spaceflight and colonization] from the viewpoint that first came language, and then came math, in the course of human events.LQP becomes appropriate and necessary only after the "spacecraft systems" are presented and defined in concise language.  To use the JFK example:

"We choose to go to the Moon" must, in all cases, predate the the LQP thereof:

Reference system   Selenocentric
Periselene altitude   100.9 kilometers (54.5 nmi)[6]
Aposelene altitude   122.4 kilometers (66.1 nmi)[6]
Inclination   1.25 degrees[6]
Period   2 hours[6]
Epoch   July 19, 1969, 21:44 UTC

https://en.wikipedia.org/wiki/Apollo_11#Launch_and_flight_to_lunar_orbit

AI can never suggest the goal, at least in my telling.

Same comment for "Planning and executing space missions." Neither of these is an area where I expect to see AI make an impact.

Agree in principle, but:  The planning can be expressed in an English language summary, including at least the intuitive methods of prioritizing those mission elements; AI could analyze the interplay of those elements, and maybe suggest element parameters that the human analysts may have overlooked.

As to the "execution" component, AI could rank those components, based on comparative LQP calculations.  In the case of a ring station [RS] at EML-1 [the reader is invited to suggest their own mission here], AI could be used to create a spreadsheet predicting the torque requirements for maintaining the Sun Synchronous Precession Orbit [SSPO] to maintain the RS orbit.

"Analyzing data from space missions" is where I'd expect to see the biggest impact. ...  "Analyzing data" is definitely a problem that seems tailor-made for modern AI. But I think we all already knew that.

Agree.

[JohnFornaro]

A surprising number of people pick problems that look to be well-suited for a particular algorithm, but they never ask whether there's any value to solving that problem.

Something something carts and horses.

[sanman]

I was playing with ChatGPT from OpenAI, and if nothing else, Machine Learning can be used to re-hash well-worn space debates like the kind we have on NSF:

[Greg Hullender]

It's cute, no question. Look at what I got when I asked why SpaceX was so far ahead of Blue Origin, despite have started later.

Blue Origin and SpaceX are both private aerospace manufacturers and spaceflight services companies that were founded around the same time, with Blue Origin being founded in 2000 and SpaceX in 2002. Both companies have made significant progress in the development of their rocket technology and have conducted a number of successful test flights.

However, there are a few factors that may have contributed to SpaceX's more rapid development and success in launching rockets into orbit. One key factor is funding: SpaceX has received a greater level of funding from private investors, which has allowed the company to invest more heavily in research and development and pursue more ambitious projects.

Another factor is experience: SpaceX was founded by Elon Musk, who had previously co-founded PayPal and had experience in the tech industry. This background may have given him a different perspective and set of skills that have helped him navigate the complex and competitive world of rocket development.

Finally, SpaceX has a clear focus on developing reusable rockets, which has the potential to significantly reduce the cost of space travel. This focus has allowed SpaceX to make rapid progress in this area and gain a competitive advantage over other companies.

That being said, Blue Origin has made significant progress in its own right and has a strong vision for the future of space exploration. The company has developed the New Shepard suborbital rocket, which has conducted several successful test flights, and is also working on the larger New Glenn rocket, which is intended to be capable of launching payloads into orbit. I believe that both Blue Origin and SpaceX have the potential to make significant contributions to the field of space exploration in the coming years.

Now, of course, this is pretty much 100% false, but what's interesting is what happened when I challenged it.
I told it, "But this isn't true! Blue Origin had far greater funding at the start, Musk has no greater understanding of Technology than Bezos, and both companies were focused on reusability from the start."

I'll confess, I didn't expect much, but I was quite impressed with the response:

I apologize for the misinformation in my previous response. You are correct that Blue Origin had a significant amount of funding at the start, as it was founded by Jeff Bezos, the CEO of Amazon, who has a net worth in the billions of dollars. Additionally, both Blue Origin and SpaceX were focused on developing reusable rockets from the start, as this has the potential to significantly reduce the cost of space travel.

It's also true that both Elon Musk and Jeff Bezos have a strong understanding of technology and have made significant contributions to the field. Both have founded successful tech companies and have demonstrated an ability to think outside the box and push the boundaries of what is possible.

It's difficult to say exactly why one company has progressed faster than the other, as there are many factors at play. Both Blue Origin and SpaceX have faced challenges and setbacks in the development of their rocket technology, and both have made significant progress in their own right. It's important to recognize that developing rockets and launching them into space is an incredibly complex and challenging task that requires a high level of expertise, resources, and perseverance. Both Blue Origin and SpaceX have demonstrated these qualities and have the potential to make significant contributions to the field of space exploration in the future.

Note that it still didn't give a decent answer to the question, but the quality of the response is quite impressive nevertheless.

Maybe I should ask it to give the answer from a SpaceX perspective. (To be authentic, that response would probably never mention Blue Origin at all.) :-)

But when I ask the original question again, I get the exact same response--word for word--with the exact same misinformation. That is, the system doesn't really learn anything. Not in real-time, anyway.

[sdsds]

More from ChatGPT from OpenAI.

[Twark_Main]

And who's "Fornarno"?

That's your superhero/supervillain name. 

[JohnFornaro]

It's cute, no question. Look at what I got when I asked why SpaceX was so far ahead of Blue Origin, despite have started later. ... But when I ask the original question again, I get the exact same response--word for word--with the exact same misinformation. That is, the system doesn't really learn anything. Not in real-time, anyway.

As you know, or may have suspected, people come with different degrees of intelligence.  Turing certainly knew this to be the case.  Some of the many things that Turing didn't know, was the direction that human tech would be "nudged" into following; the future capabilities, to him, that computers based on the Von Newman design would accomplish; and the future politics of the world in general.  Would Turing have questioned the AI as you did? 

A lot of people would take that first answer as evincing definite intelligence.  Does the T-Test need to be fine tuned somehow?

Now, of course, this is pretty much 100% false ...

It all sounded plausible to me;  what do you mean "pretty much 100% false"? 

SpaceX has a clear focus on developing reusable rockets...

Which of its statements are "pretty much 100% false"?

[JohnFornaro]

And who's "Fornarno"?

That's your superhero/supervillain name. 

Ivan Oranrof, actually.  You need to do your homework.  Check out my twit account.

[edzieba]

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

[Robotbeat]

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

LOL, absolutelyThere are ways to get it to summarize information, though. Just tell it to be concise and not use any extra words to explain the concept.

[Greg Hullender]

Now, of course, this is pretty much 100% false ...

It all sounded plausible to me;  what do you mean "pretty much 100% false"? 

SpaceX has a clear focus on developing reusable rockets...

Which of its statements are "pretty much 100% false"?

I'm afraid this is the point where I write you off. Your post is so clearly in bad faith, it doesn't merit a response.

[JohnFornaro]

https://forum.nasaspaceflight.com/index.php?topic=57207.msg2442219#msg2442219

A few thoughts about being mostly true or false.

Blue Origin and SpaceX are both private aerospace manufacturers and spaceflight services companies that were founded around the same time, with Blue Origin being founded in 2000 and SpaceX in 2002. Both companies have made significant progress in the development of their rocket technology and have conducted a number of successful test flights.

I assume the two numbers, 2000 and 2002, are accurate.  The term "significant progress is true.  Before 2000 and 2002, neither company had made any notable "progress" in their respective fields of endeavor.  SpaceX is ahead of BO; that BO started two years later is NOT the reason they're behind SpaceX.  The chat bot has a lot in common with the facile analyst.

However, there are a few factors that may have contributed to SpaceX's more rapid development and success in launching rockets into orbit. One key factor is funding: SpaceX has received a greater level of funding from private investors, which has allowed the company to invest more heavily in research and development and pursue more ambitious projects.

The word "may" is a squishy term, which provides cover for extravagant claims like "rapid development" and "invest more heavily" and "more ambitious projects".  The statement should only be interpreted as an opinion.

From the oracle:

On July 20, 2021, Blue Origin sent its first crewed mission into space via its New Shepard rocket and spaceflight system.]

https://en.wikipedia.org/wiki/Blue_Origin

SpaceX has launched astros to the ISS before BO and BO has yet to orbit in LEO, AIUI.  This could be referred to as a "more rapic development".  As to the "greater level of funding from private investors", this too is a squishy, undefined term.

Another factor is experience: SpaceX was founded by Elon Musk, who had [blah blah blah]. This background may have given him a different perspective and set of skills that have helped him navigate the complex and competitive world of rocket development.

Yet another editorial comment, not worthy of the label "truth" or "falsehood".  So, Musk has a "different perspective" than Bezos.  The field cannot be said to be non-competitive.  At the moment, Musk is ahead.  The bot provides no new insight into either company.

Finally, SpaceX has a clear focus on developing reusable rockets, which has the potential to significantly reduce the cost of space travel. This focus has allowed SpaceX to make rapid progress in this area and gain a competitive advantage over other companies.

Obviously, SpaceX is focusing on re-usability.  Does that focus allow them to "make rapid progress"?  It helps, for sure.  "Allow"?  Well the chatbot was written by people who studied PR language.  The bot's statement is basically a PR statement.  I'd "allow" it the choice of doing what it does.

That being said, Blue Origin has made significant progress in its own right and has a strong vision for the future of space exploration. The company has developed the New Shepard suborbital rocket, which has conducted several successful test flights, and is also working on the larger New Glenn rocket, which is intended to be capable of launching payloads into orbit.

Broadly "true", if one wanted to carry on a friendly discussion.

I believe that both Blue Origin and SpaceX have the potential to make significant contributions to the field of space exploration in the coming years.

I like the way the bot says "I".  When you call up your ISP on a technical issue, do you really believe that when the bot says, "I have your account information here...clickity-clikity-click". that it really is a sentient being? And that it's really typing on a keyboard?

Some people do.

[Twark_Main]

And who's "Fornarno"?

That's your superhero/supervillain name. 

Ivan Oranrof, actually.  You need to do your homework.  Check out my twit account.

Nice save. It's hard work maintaining a secret identity!

[Twark_Main]

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

Of course.

There's an outcry any time an AI doesn't emit a "politically correct" answer. It's no wonder engineers train their AI to talk like a politician.

[sanman]

It's a little prone to repetition, but it does show a lot of potential as an early product

[JohnFornaro]

Little Jack Horner
Sits in a corner
Extracting cube roots to infinity.
An activity for boys
Which will minimize noise
And produce a more peaceful vicinity.

-A Space Child's Mother Goose

[edzieba]

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

Of course.

There's an outcry any time an AI doesn't emit a "politically correct" answer. It's no wonder engineers train their AI to talk like a politician.

No, it's a byproduct of AI being trained to assemble coherent sentences (and ideally ones derived from information semantically related to the prompt) from a vast training set rather than forming a separate knowledge graph and constructing sentences based on that graph. The chatbot has no understanding of the data it is dispensing, only an understanding of how closely related the prompt is to sentences it has already observed. Political correctness is of minimal concern beyond trying to avoid poisoned data in the training datasets (e.g. Microsoft's 'Tay' re-training on unfiltered Twitter inputs and that being used as an attack vector).

[JohnFornaro]

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

Of course.

There's an outcry any time an AI doesn't emit a "politically correct" answer. It's no wonder engineers train their AI to talk like a politician.

No, it's a byproduct of AI being trained to assemble coherent sentences (and ideally ones derived from information semantically related to the prompt) from a vast training set rather than forming a separate knowledge graph and constructing sentences based on that graph. ...

Why couldn't "coherent sentences" be constructed from a "vast" right leaning training set?  The chat bots are being trained to present a viewpoint, and not so much as to inform, at least as I see it.

[edzieba]

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

Of course.

There's an outcry any time an AI doesn't emit a "politically correct" answer. It's no wonder engineers train their AI to talk like a politician.

No, it's a byproduct of AI being trained to assemble coherent sentences (and ideally ones derived from information semantically related to the prompt) from a vast training set rather than forming a separate knowledge graph and constructing sentences based on that graph. ...

Why couldn't "coherent sentences" be constructed from a "vast" right leaning training set?

Why would you exclude a huge amount of training data (and spend thousands man-years filtering based on an extremely subjective viewpoint) in order to train an unnecessarily biased AI? Doesn't seem to even be feasible to pay people to do that filtering (the amount of text processed for GPT is unimaginably vast), much less any point in doing so to produce a less comprehensive model.

[Robotbeat]

Can we NOT bring politics into this?

[JohnFornaro]

Can we NOT bring politics into this?

Ask that of the AI trainers, who are doing exactly that.

[Robotbeat]

Can we NOT bring politics into this?

Ask that of the AI trainers, who are doing exactly that.

Even if true, it’s completely off topic.

[JohnFornaro]

Can we NOT bring politics into this?

Ask that of the AI trainers, who are doing exactly that.

Even if true, it’s completely off topic.

Then mention your concerns to the guy who brought it up.

[edzieba]

Can we NOT bring politics into this?

Ask that of the AI trainers, who are doing exactly that.

You're going to have to offer some evidence for this repeated assertion.

[Robotbeat]

Can we NOT bring politics into this?

Ask that of the AI trainers, who are doing exactly that.

You're going to have to offer some evidence for this repeated assertion.

Why? It’s off topic and boring.

[Robotbeat]

Mr. Scott produced some pretty good images with DallE2. Better than I’ve been able to manage. I like the pillars, especially.

This thread is just fun to think about.  I'm enjoying taking these wacky ideas and playing around with the OpenAI DALL*E 2 image generator.  Just want to generate some concept art and blow off reality.

[sanman]

What exactly is neuromorphic computing? And what are the advantages of it?

[sanman]

OpenAI has also come out with a new 3D-object generator that works on input from text prompts.

https://techcrunch.com/2022/12/20/openai-releases-point-e-an-ai-that-generates-3d-models/

https://www.sciencefocus.com/future-technology/point-e/

It starts off by generating point-clouds from the text prompts, and then converts those into 3D object meshes.

ChatGPT can create a gear in AutoCAD in its native AutoLisp:




I'm curious, though - could Machine Learning be used to make progressively better rockets in Kerbal Space Program?
In that situation, perhaps you'd have a neural network that keeps proposing new rocket configurations, and meanwhile the adversarial discriminator would be a physics engine. The generative proposer would then get progressively smarter by learning from the rejections. It would be interesting to see which solutions it converges on.

[ppnl]

What exactly is neuromorphic computing? And what are the advantages of it?

Neuromorphic computing is simply an attempt to copy the function of the brain on a hardware level. For example a computer may have a clock speed of five billion cycles per second and a signal speed a respectable fraction of the speed of light. A neuron in the brain only fires a few times a second and the signal speed is like 250 miles per hour. But on a hardware level the brain has a hundred billion neurons working in parallel while a computer can only have a handful of sequential processes.

You can mimic the parallelism of neurons in a ordinary chip but that is vastly inefficient. They use graphics cards to do neural nets because they have thousands of CUDA cores but that is still far below the hundred billion neurons in a brain.

The idea is to implement neurons in very small devices like memristors. The advantage is that it could be vastly more energy efficient and vastly better circuit density. It should be possible to exceed the circuit density, energy efficiency and speed of the brain this way. After all nature rarely finds anything close to an optimal solution.     

[sanman]

The idea is to implement neurons in very small devices like memristors. The advantage is that it could be vastly more energy efficient and vastly better circuit density. It should be possible to exceed the circuit density, energy efficiency and speed of the brain this way. After all nature rarely finds anything close to an optimal solution.

Regarding memristors, the ones I've read about seem to be electrochemical devices which partly depend on chemistry for hysteresis. From what I see, this then introduces a problem of entropy/non-reversibility which affects the life of the memristor component (just like the limited cycle life of batteries). At some point, the memristor will become non-responsive to further changes, and effectively be dead. You'll tell me that this is how real neurons work too, but biological neurons are supported by a metabolism which preserves their life. Can memristors have a useful lifespan akin to conventional memory chips?

[JohnFornaro]

OpenAI has also come out with a new 3D-object generator that works on input from text prompts. ... ChatGPT can create a gear in AutoCAD in its native AutoLisp:

That's an image, and I couldn't test it directly, but I just experimented with its code generating capaity, using the following language and two regenerations.  None of the Lisp routines worked as expected, although the routines loaded correctly, proving that ChatBot can create grammatically correct code.

The criticism will be made that my prompt didn't give ChatBot enough instruction to create the code that I wanted.  To this, I say, elsewhere on the site, they point out that ChatBot will ask for clarification, but this didn't happen.

Importantly, ChatBot did demonstrate much deeper knowledge about ACAD Lisp than I have. Its code snippets opened my eyes to different ways of programming with Lisp.

*************************

; Created by ChatbotGPT on 12-25-22, given the
; following text:

; Please write a Lisp routine for AutoCAD release 14
; which converts a 2D block into a 3D block.

;;;;;;;;;;;;;;;;;;;

;(defun C:CONVERT2DTO3D (/ ent newblock)
;  (command "block" "*")
;  (setq ent (entsel "Select a 2D block: "))
;  (if ent
;      (progn
;        (setq newblock (car (entget (car ent))))
;        (command "block" newblock "3d" "")
;        (alert "Block converted to 3D.")
;      )
;    (alert "No block selected.")
;  )
;  (princ)
;)

;;;;;;;;;;;;;;;;;;;

; Regenerate response provided this:

;;;;;;;;;;;;;;;;;;;

(defun c:3DBLOCK (/ blkname inspt elev inspnt blk)
  (setq blkname (getstring "\nEnter block name: "))
  (setq inspt (getpoint "\nSpecify insertion point: "))
  (setq elev (getreal "\nSpecify elevation: "))
  (setq inspnt (list (car inspt) (cadr inspt) elev))
  (setq blk (entmakex '(-2 . "INSERT") '(0 . "NAME") blkname '(66 . 1) '(10 . inspnt)))
  (entmod blk '(210 . '(0.0 0.0 1.0)))
  (command "REGEN")
  (princ)
)

;;;;;;;;;;;;;;;;;;;;;;;;;

; ChatBotGPT provided these instructions:

; To use this routine, type 3DBLOCK at the command prompt
; and follow the prompts. The routine will ask you for the
; name of the block to convert, the insertion point for the
; new 3D block, and the elevation of the insertion point. It
; will then create a new 3D block at the specified location
; and with the specified elevation.

; Note: This Lisp routine was written for AutoCAD Release 14,
; which was released in 1997. It may not work with newer
; versions of AutoCAD.

;;;;;;;;;;;;;;;;;;

; Second Regenerate response provided this:

; Here is a Lisp routine that can be used to convert
; a 2D block into a 3D block in AutoCAD release 14:

;;;;;;;;;;;;;;;;;;

(defun c:convert2dblockto3dblock ()
  (setq bname (getstring "Enter name of block to convert: "))
  (setq b (entget (car (entsel (strcat "Select insertion point of block reference to convert: " bname)))))
  (setq inspoint (cdr (assoc 10 b)))
  (command "block" bname "*" "y" "3d" inspoint "")
  (command "block" bname "*" "y" "2d" inspoint "")
  (princ)
)

; To use this routine, simply enter the command
; "convert2dblockto3dblock" in the AutoCAD command prompt
; and follow the prompts. The routine will ask you to enter
; the name of the 2D block that you want to convert, and
; then select the insertion point of a block reference of
; that block. The routine will then create a new 3D block
; with the same name and geometry as the original 2D block,
; and replace the selected block reference with a reference
; to the new 3D block.

[JohnFornaro]

After all nature rarely finds anything close to an optimal solution.   

Really?

[Robotbeat]

JohnFornano, your exploration of code with ChatGPT shows an important way AI is and will be used: as a sort of assistant to a human. Computers are often called a bicycle for the mind. With things like GitHub Copilot and ChatGPT, after they’re tweaked and refined, are like an e-bike for the mind.

I would like to use transfer learning to train ChatGPT/DaVinci003 on NTRS documents, but you can’t download the whole model, yet, so that’s not really doable except by feeding it via prompts (not scalable since its short term memory is only 4000 tokens/words… ironic since it needs almost a terabyte of RAM to run).

[sanman]

I've seen various instances where asking ChatGPT to code something produces code that doesn't quite work right. It will confidently spit out an answer, even while that answer is wrong. So it's important to test any code that it spits out, and not blindly take it on faith. I guess AI is always going to need some human supervision.

[JohnFornaro]

I've seen various instances where asking ChatGPT to code something produces code that doesn't quite work right. It will confidently spit out an answer, even while that answer is wrong. So it's important to test any code that it spits out, and not blindly take it on faith. I guess AI is always going to need some human supervision.

Bingo.  Scott Adams, the Dilbert guy, is running his podcast at this very moment:



Since AI is topical,  I'll point out that Scott is asking ChatGPT leading questions about broadly topical subjects in other fields.  The AI acknowleges that it can only report on the data it has been trained upon, tacitly admitting that its responses are biased by its creators.  WRT coding; Lisp is very nearly irrelevant in today's coding environment. I find it amazing that the AI can generate grammatical code, even if non-working.

[DanClemmensen]

I've seen various instances where asking ChatGPT to code something produces code that doesn't quite work right. It will confidently spit out an answer, even while that answer is wrong. So it's important to test any code that it spits out, and not blindly take it on faith. I guess AI is always going to need some human supervision.

You appear to be overgeneralizing. ChatGPT is not optimized at all for code generation and is fairly crude by comparison to even some of today's systems, so "always going to need" is shortsighted. Within a decade, and probably within a year, someone will create a reflexive version of one of these models that will begin to improve itself.

[Robotbeat]

Here’s an article on AI being used for planetary exploration and astronomy:
https://www.space.com/solar-system-planetary-science-machine-learning

And here’s the summary of the article that I had ChatGPT generate:

Machine learning is changing how scientists study space by allowing computers to identify patterns in data and make decisions, predictions or classifications. This can save time for scientists by identifying phenomena such as boulders on other planets, explosive volcanoes on Mercury and vortexes in Jupiter's atmosphere, among other things. In the future, artificial intelligence could be used in space probes to enable them to make real-time responses to hazards and land autonomously, rather than relying on communication with Earth that can take several hours. Machine learning and artificial intelligence could also be used to help plan future missions to bodies such as Jupiter's moon Europa.

[sanman]

Here’s an article on AI being used for planetary exploration and astronomy:
https://www.space.com/solar-system-planetary-science-machine-learning

And here’s the summary of the article that I had ChatGPT generate:

Machine learning is changing how scientists study space by allowing computers to identify patterns in data and make decisions, predictions or classifications. This can save time for scientists by identifying phenomena such as boulders on other planets, explosive volcanoes on Mercury and vortexes in Jupiter's atmosphere, among other things. In the future, artificial intelligence could be used in space probes to enable them to make real-time responses to hazards and land autonomously, rather than relying on communication with Earth that can take several hours. Machine learning and artificial intelligence could also be used to help plan future missions to bodies such as Jupiter's moon Europa.

So I'm curious - did you simply feed it that URL as part of your input, and have it generate its summary on that basis?
If so, that's a pretty nifty trick - gonna have to use that more often.
It would be intriguing if we users can influence ChatGPT by feeding it reading material of our own choosing.

[DanClemmensen]

Here’s an article on AI being used for planetary exploration and astronomy:
https://www.space.com/solar-system-planetary-science-machine-learning

And here’s the summary of the article that I had ChatGPT generate:

Machine learning is changing how scientists study space by allowing computers to identify patterns in data and make decisions, predictions or classifications. This can save time for scientists by identifying phenomena such as boulders on other planets, explosive volcanoes on Mercury and vortexes in Jupiter's atmosphere, among other things. In the future, artificial intelligence could be used in space probes to enable them to make real-time responses to hazards and land autonomously, rather than relying on communication with Earth that can take several hours. Machine learning and artificial intelligence could also be used to help plan future missions to bodies such as Jupiter's moon Europa.

So I'm curious - did you simply feed it that URL as part of your input, and have it generate its summary on that basis?
If so, that's a pretty nifty trick - gonna have to use that more often.
It would be intriguing if we users can influence ChatGPT by feeding it reading material of our own choosing.

As I understand it, the training dataset for ChatGPT is not being augmented. It was frozen in 2021. This is one reason ChatGPT knows very little about itself. I got this potentially incorrect information from a single sentence in a non-technical article on ChatGPT, so it might be wrong. I'm  not sure how one of these systems will evolve it is allowed to incorporate non-curated information indiscriminately. I suspect this is one the areas Google is researching intensively in reaction to ChatGPT.
   https://mezha.media/en/2022/12/22/google-is-panicking-over-the-ability-of-the-chatgpt-bot-to-answer-user-questions/

[JohnFornaro]

https://forum.nasaspaceflight.com/index.php?topic=57207.msg2442453#msg2442453

********

More from ChatGPT from OpenAI.

Like other GPT language models, ChatGPT clearly makes for a good politician: provides a few paragraphs of waffle without meaningfully answering the question.

Of course.

There's an outcry any time an AI doesn't emit a "politically correct" answer. It's no wonder engineers train their AI to talk like a politician.

No, it's a byproduct of AI being trained to assemble coherent sentences ... from a vast training set rather than forming a separate knowledge graph and constructing sentences based on that graph. The chatbot has no understanding of the data it is dispensing, only an understanding of how closely related the prompt is to sentences it has already observed. Political correctness is of minimal concern beyond trying to avoid poisoned data in the training datasets ...

As I understand it, the training dataset for ChatGPT is not being augmented. It was frozen in 2021. This is one reason ChatGPT knows very little about itself. I got this potentially incorrect information from a single sentence in a non-technical article on ChatGPT, so it might be wrong. I'm  not sure how one of these systems will evolve it is allowed to incorporate non-curated information indiscriminately. I suspect this is one the areas Google is researching intensively in reaction to ChatGPT.
   https://mezha.media/en/2022/12/22/google-is-panicking-over-the-ability-of-the-chatgpt-bot-to-answer-user-questions/

Reading the above passages suggests that ChatGPT is indeed being fed curated material by its trainers.  It's not a surprise then, that google might be concerned with the very real dangers of unsafe input being asked of ChatGPT.

This opinion of mine is confirmed in the article:

https://mezha.media/en/2022/12/22/google-is-panicking-over-the-ability-of-the-chatgpt-bot-to-answer-user-questions/

Google is reluctant to share its technology because, like ChatGPT and similar systems, it can generate false, toxic and biased information.

Maybe everybody should be limited to asking ChatGPT only about Lisp routines?

[JohnFornaro]

If I provide a clearly labeled AutoCAD R14 file showing a ring station in plan and section, would you be able to calculate the floor area in the rings?  I would need the capability of uploading the CAD file for you to review.

Respose:

Yes, I would be able to help you calculate the floor area in the rings of the AutoCAD R14 file that you have provided. In order to do this, you will need to use a tool or software that allows you to measure and calculate the area of the various components in the file. There are several ways to do this, depending on the specific needs of your project and the resources you have available.

One option is to use the built-in area measurement tools in AutoCAD R14. These tools allow you to select specific objects in the drawing and calculate their area, as well as total the area of multiple objects. You can find more information about how to use these tools in the AutoCAD R14 documentation or by searching online.

Another option is to use a third-party tool or plugin that is specifically designed for calculating area in AutoCAD drawings. These tools often offer additional features and functionality, such as the ability to automatically label the calculated areas or export the results to a spreadsheet. You can find more information about these tools by searching online or contacting the developers directly.

Regardless of which approach you choose, it is important to carefully review the results of your calculations and double-check your work to ensure that the results are accurate. If you have any specific questions or need further assistance with this task, please don't hesitate to ask.

My reply:

I'm afraid that is a long winded way of saying "no".  I'm trying to save time by putting what for me is a time consuming task, subject to math errors, into ChatGpt to solve.

I apologize if my previous response was not helpful. To answer your question, it is not possible for me to directly access or review the AutoCAD R14 file that you have mentioned. As an AI language model, I do not have the ability to open or view files, and I do not have access to any external resources or tools.

However, I can provide you with general guidance on how to calculate the floor area in the rings of the AutoCAD R14 file using the tools and techniques that are available in the software. To do this, you will need to use the area measurement tools in AutoCAD R14, or a third-party tool or plugin that is specifically designed for this purpose.

To use the built-in area measurement tools in AutoCAD R14, you will need to select the objects that you want to measure and then use the appropriate command or function to calculate their area. For example, you can use the "AREA" command to calculate the area of a closed polyline or region, or the "MEASUREGEOM" command to measure the length and area of various objects. You can find more information about these tools in the AutoCAD R14 documentation or by searching online.

If you prefer to use a third-party tool or plugin, you will need to install the software and follow the instructions provided by the developer to measure and calculate the area of the objects in your AutoCAD file.

Regardless of which approach you choose, it is important to carefully review the results of your calculations and double-check your work to ensure that the results are accurate. If you have any specific questions or need further assistance with this task, please don't hesitate to ask.

My response:

As I mentioned in my original query, I would need the ability to upload my CAD file to your server.  Do your trainers have plans to accomodate my request?

Additionally, as I mentioned in our "Lisp routine" conversation, please do not apologize to me. It comes across as false and detracts from the discussion.

I apologize for misunderstanding your request in my previous response. I do not have the ability to access or review any files that are uploaded to me, as I am an AI language model and do not have the capability to open or view files. I also do not have access to any external resources or tools, so I am unable to perform tasks that require me to access or interact with external systems or data.

As an AI language model, my primary function is to provide general information and guidance on a wide range of topics. I am not able to perform specific tasks or projects on behalf of users, and I do not have the ability to access or interact with external resources or systems.

I apologize if this is not the answer you were hoping for. If you have any other questions or need further assistance with your project, please don't hesitate to ask.

My response:

Can you provide me with the contact information of one of your trainers, so that I can pursue this line of inquiry?  I will keep the contact information private.  It is imperative to be able to upload CAD files to answer the technical questions that I ask.

If AI is to become useful to the general population, it would need to be able to help us  solve real world problems.  From my viewpoint, general information about politics, poetry, and the humanities, are less important than solving some difficult analytical problems along the lines of what I have presented.

I apologize, but I am unable to provide you with the contact information of any of my trainers or creators, as I am an AI language model and do not have the ability to access or share this type of information. As an AI, I do not have personal relationships or contact information for the individuals who were involved in my development or training.

I understand that you are looking for a solution to a specific problem that you are facing, and I apologize if I am unable to provide the assistance you need. While I am trained to provide general information and guidance on a wide range of topics, I do not have the ability to perform specific tasks or projects on behalf of users, or to interact with external resources or systems.

AI has the potential to be a powerful tool for solving a variety of problems, including technical and analytical challenges. However, it is important to recognize that AI has limitations and is not capable of replacing human expertise or judgment in all situations. In order to use AI effectively, it is important to understand its capabilities and limitations, and to use it appropriately in combination with other resources and tools.

What a brave new world that has such chat bots in it.

[Robotbeat]

Here’s an article on AI being used for planetary exploration and astronomy:
https://www.space.com/solar-system-planetary-science-machine-learning

And here’s the summary of the article that I had ChatGPT generate:

Machine learning is changing how scientists study space by allowing computers to identify patterns in data and make decisions, predictions or classifications. This can save time for scientists by identifying phenomena such as boulders on other planets, explosive volcanoes on Mercury and vortexes in Jupiter's atmosphere, among other things. In the future, artificial intelligence could be used in space probes to enable them to make real-time responses to hazards and land autonomously, rather than relying on communication with Earth that can take several hours. Machine learning and artificial intelligence could also be used to help plan future missions to bodies such as Jupiter's moon Europa.

So I'm curious - did you simply feed it that URL as part of your input, and have it generate its summary on that basis?
If so, that's a pretty nifty trick - gonna have to use that more often.
It would be intriguing if we users can influence ChatGPT by feeding it reading material of our own choosing.

I Just wrote "Summarize this article:"
And then I pasted the text of the article (but deleted image captions). Very simple.

[sanman]

[Robotbeat]

Summarize, please, or at least explain how it’s relevant to the topic. Don’t expect people to watch a 8 minute clickbait video just to get your point.

[sanman]

Well, she was mentioning how ChatGPT typically speaks with 100% confidence on things that it does not necessarily have 100% confidence on. So that's one problem with an AI like ChatGPT. The manner and tone in which it speaks needs to reflect its relative confidence level in what is being said.

[Robotbeat]

Well, she was mentioning how ChatGPT typically speaks with 100% confidence on things that it does not necessarily have 100% confidence on. So that's one problem with an AI like ChatGPT. The manner and tone in which it speaks needs to reflect its relative confidence level in what is being said.

Right. Sounds about right for emulating human behavior in the vast majority of humans.

[Asteroza]

Well, she was mentioning how ChatGPT typically speaks with 100% confidence on things that it does not necessarily have 100% confidence on. So that's one problem with an AI like ChatGPT. The manner and tone in which it speaks needs to reflect its relative confidence level in what is being said.

Right. Sounds about right for emulating human behavior in the vast majority of humans.

AI Dunning–Kruger Turing test?

[sanman]

Right. Sounds about right for emulating human behavior in the vast majority of humans.

AI Dunning–Kruger Turing test?

What happens when we have enough instances of AI functioning closely with human beings?

Do we need to bring in Asimov's 3 Laws in a formalized way?

Do these have to be verified at all stages:  design, development, implementation, operation?

[DanClemmensen]

Right. Sounds about right for emulating human behavior in the vast majority of humans.

AI Dunning–Kruger Turing test?

What happens when we have enough instances of AI functioning closely with human beings?

Do we need to bring in Asimov's 3 Laws in a formalized way?

Do these have to be verified at all stages:  design, development, implementation, operation?

There is a large and active AI security academic community that has been studying this for several decades. I don't think we can summarize this work here in a few sentences.  Here is an excellent overview book from 2019:
   https://www.amazon.com/Artificial-Intelligence-Security-Chapman-Robotics/dp/0815369824?fbclid=IwAR1sBo6eLJRhmollCqmDYYd18rEFFCoGvhFn2DB3HgAJ0XQJoZw_QT_gogQ

[sanman]

Here's an interesting video that's just come out on the topic:

[sanman]

Atlas is getting more and more into showboating






I wonder how long it will be until Musk will showcase comparable performance from Tesla-bot.

The only reason I can see for Atlas or Tesla-bot to have human anatomical layout is so that they can be conveniently swapped in for actual human workers, carrying out human work processes, even while using human tools.

But I wonder to what extent that approach can be ported to the surface of the Moon or Mars, versus custom-designing all hardware for those environments.

[DanClemmensen]

I think we are conflating robotics and AI. Robots can benefit from AI but they do not require it. AI can be useful without robots. An AI that could control a spacecraft might be very useful, but most folks don't call a spacecraft a "robot". So what are we talking about here?

[Asteroza]

Well, she was mentioning how ChatGPT typically speaks with 100% confidence on things that it does not necessarily have 100% confidence on. So that's one problem with an AI like ChatGPT. The manner and tone in which it speaks needs to reflect its relative confidence level in what is being said.

Right. Sounds about right for emulating human behavior in the vast majority of humans.

AI Dunning–Kruger Turing test?

So ChatGPT sorta passed one Wharton School of Business class final exam, so it could have potentially earned an MBA...

which some might interpret as either a rise in AI capability, or the bar might have been a little low...

I think we are conflating robotics and AI. Robots can benefit from AI but they do not require it. AI can be useful without robots. An AI that could control a spacecraft might be very useful, but most folks don't call a spacecraft a "robot". So what are we talking about here?

I dunno about you, but I would hope all future spacecraft have a robot arm for vehicle maintenance purposes standard. Which would make them all robots.

[sanman]

There is a large and active AI security academic community that has been studying this for several decades. I don't think we can summarize this work here in a few sentences.  Here is an excellent overview book from 2019:
   https://www.amazon.com/Artificial-Intelligence-Security-Chapman-Robotics/dp/0815369824?fbclid=IwAR1sBo6eLJRhmollCqmDYYd18rEFFCoGvhFn2DB3HgAJ0XQJoZw_QT_gogQ

Apparently, some software engineers from OpenAI have left to create their own company Anthropic AI, and have created their own new rival product called Claude.
Among other things, Claude seems to make use of a new concept called a 'Constitution', which as its name suggests, is a construct meant to explicitly embody a set of moral and ethical rules which govern how the AI operates. It also seems to use some sort of AI-supervised reinforcement feedback.

[DanClemmensen]

There is a large and active AI security academic community that has been studying this for several decades. I don't think we can summarize this work here in a few sentences.  Here is an excellent overview book from 2019:
   https://www.amazon.com/Artificial-Intelligence-Security-Chapman-Robotics/dp/0815369824?fbclid=IwAR1sBo6eLJRhmollCqmDYYd18rEFFCoGvhFn2DB3HgAJ0XQJoZw_QT_gogQ

Apparently, some software engineers from OpenAI have left to create their own company Anthropic AI, and have created their own new rival product called Claude.
Among other things, Claude seems to make use of a new concept called a 'Constitution', which as its name suggests, is a construct meant to explicitly embody a set of moral and ethical rules which govern how the AI operates. It also seems to use some sort of AI-supervised reinforcement feedback.

The general problem is called "AI alignment". It's hard.
     https://en.wikipedia.org/wiki/AI_alignment

[ppnl]

There is a large and active AI security academic community that has been studying this for several decades. I don't think we can summarize this work here in a few sentences.  Here is an excellent overview book from 2019:
   https://www.amazon.com/Artificial-Intelligence-Security-Chapman-Robotics/dp/0815369824?fbclid=IwAR1sBo6eLJRhmollCqmDYYd18rEFFCoGvhFn2DB3HgAJ0XQJoZw_QT_gogQ

Apparently, some software engineers from OpenAI have left to create their own company Anthropic AI, and have created their own new rival product called Claude.
Among other things, Claude seems to make use of a new concept called a 'Constitution', which as its name suggests, is a construct meant to explicitly embody a set of moral and ethical rules which govern how the AI operates. It also seems to use some sort of AI-supervised reinforcement feedback.

The general problem is called "AI alignment". It's hard.
     https://en.wikipedia.org/wiki/AI_alignment

To be fair humans are bad at this as well.


<quote>
AI systems can be challenging to align and misaligned systems can malfunction or cause harm. It can be difficult for AI designers to specify the full range of desired and undesired behaviors. Therefore, they use easy-to-specify proxy goals that omit some desired constraints. However, AI systems exploit the resulting loopholes. As a result, they accomplish their proxy goals efficiently but in unintended, sometimes harmful ways (reward hacking). AI systems can also develop unwanted instrumental behaviors such as seeking power, as this helps them achieve their given goals. Furthermore, they can develop emergent goals that may be hard to detect before the system is deployed, facing new situations and data distributions.
</quote>

Replace "AI" with political parties, churches, corporations or the local elk lodge. Look at everything from the tribal wasteland of American politics to the war in Ukraine.

[DanClemmensen]

There is a large and active AI security academic community that has been studying this for several decades. I don't think we can summarize this work here in a few sentences.  Here is an excellent overview book from 2019:
   https://www.amazon.com/Artificial-Intelligence-Security-Chapman-Robotics/dp/0815369824?fbclid=IwAR1sBo6eLJRhmollCqmDYYd18rEFFCoGvhFn2DB3HgAJ0XQJoZw_QT_gogQ

Apparently, some software engineers from OpenAI have left to create their own company Anthropic AI, and have created their own new rival product called Claude.
Among other things, Claude seems to make use of a new concept called a 'Constitution', which as its name suggests, is a construct meant to explicitly embody a set of moral and ethical rules which govern how the AI operates. It also seems to use some sort of AI-supervised reinforcement feedback.

The general problem is called "AI alignment". It's hard.
     https://en.wikipedia.org/wiki/AI_alignment

To be fair humans are bad at this as well.


<quote>
AI systems can be challenging to align and misaligned systems can malfunction or cause harm. It can be difficult for AI designers to specify the full range of desired and undesired behaviors. Therefore, they use easy-to-specify proxy goals that omit some desired constraints. However, AI systems exploit the resulting loopholes. As a result, they accomplish their proxy goals efficiently but in unintended, sometimes harmful ways (reward hacking). AI systems can also develop unwanted instrumental behaviors such as seeking power, as this helps them achieve their given goals. Furthermore, they can develop emergent goals that may be hard to detect before the system is deployed, facing new situations and data distributions.
</quote>

Replace "AI" with political parties, churches, corporations or the local elk lodge. Look at everything from the tribal wasteland of American politics to the war in Ukraine.

Oh yes. If our AI learns ethics by assimilating all the world's writings then we can expect it's ethics to be at best as good as the average. This would be similar to the ChatGPT's level as a writer. Humans cannot agree on the "correct" ethics. Why should an AI be any better?

Unfortunately this is NOT irrelevant to the use of AI in space applications. You can try to dismiss the problem that HAL 9000 faced in "2001: A Space Odyssey" as being just science fiction, but the same sort of thing may occur in real life. Space makes the problem harder, because one really major reason to use AI is the communications lag, but the communications lag means the AI must make real-time ethical decisions without the ability to discuss them with humans on Earth.

[sanman]

To be fair humans are bad at this as well.


<quote>
AI systems can be challenging to align and misaligned systems can malfunction or cause harm. It can be difficult for AI designers to specify the full range of desired and undesired behaviors. Therefore, they use easy-to-specify proxy goals that omit some desired constraints. However, AI systems exploit the resulting loopholes. As a result, they accomplish their proxy goals efficiently but in unintended, sometimes harmful ways (reward hacking). AI systems can also develop unwanted instrumental behaviors such as seeking power, as this helps them achieve their given goals. Furthermore, they can develop emergent goals that may be hard to detect before the system is deployed, facing new situations and data distributions.
</quote>

Replace "AI" with political parties, churches, corporations or the local elk lodge. Look at everything from the tribal wasteland of American politics to the war in Ukraine.

Oh yes. If our AI learns ethics by assimilating all the world's writings then we can expect it's ethics to be at best as good as the average. This would be similar to the ChatGPT's level as a writer. Humans cannot agree on the "correct" ethics. Why should an AI be any better?

Unfortunately this is NOT irrelevant to the use of AI in space applications. You can try to dismiss the problem that HAL 9000 faced in "2001: A Space Odyssey" as being just science fiction, but the same sort of thing may occur in real life. Space makes the problem harder, because one really major reason to use AI is the communications lag, but the communications lag means the AI must make real-time ethical decisions without the ability to discuss them with humans on Earth.

Yes, HAL9000 was likewise exactly the first thing I also thought of when reading ppnl's post.
And in the absence of communicative feedback from humanity, does the remote AI then develop an independent morality?

My analogy would be of the Dog versus the Wolf.
The domesticated Dog grows up in the embrace of Man, and is our servant, companion, even family member.
The Wolf however grows up separately outside of Man's embrace, having evolved as a predator and potential threat.

AI that evolves within the embrace of Man would then likewise be our servant, companion, etc.
But AI that evolves remotely in isolation and outside of our embrace or influence of communication, then might have the potential to develop behavior that diverges from what we would consider acceptable.

So the remoteness or isolation arising from something like spaceflight could ironically bring about the rogue AI scenario which Elon Musk warns about.

Besides Clarke's 2001, I also recall the Bolo series of novels by Keith Laumer, which featured AI-controlled tanks.

[ppnl]

[quote author=DanClemmensen link=topic=57207.msg2453974#msg2453974

Oh yes. If our AI learns ethics by assimilating all the world's writings then we can expect it's ethics to be at best as good as the average. This would be similar to the ChatGPT's level as a writer. Humans cannot agree on the "correct" ethics. Why should an AI be any better?

Unfortunately this is NOT irrelevant to the use of AI in space applications. You can try to dismiss the problem that HAL 9000 faced in "2001: A Space Odyssey" as being just science fiction, but the same sort of thing may occur in real life. Space makes the problem harder, because one really major reason to use AI is the communications lag, but the communications lag means the AI must make real-time ethical decisions without the ability to discuss them with humans on Earth.
[/quote]

Interesting thing about the Hal 9000 malfunction... The book and the movie had differing takes on it.

In the movie Hal killed the crew to protect itself and the mission.

In the book it is a little more complex. Hal was create to honestly collect, process and pass on information. The problem happened when someone on the ground ordered Hal to keep the true nature of the mission secret. This created a conflict.

Hal's solution was elegant... in a mathematical sense.

This is why you can't have a hardwired set of rules for morality. They can always be gamed. 

[ppnl]

[quote author=sanman link=topic=57207.msg2454037#msg2454037

Yes, HAL9000 was likewise exactly the first thing I also thought of when reading ppnl's post.
And in the absence of communicative feedback from humanity, does the remote AI then develop an independent morality?

My analogy would be of the Dog versus the Wolf.
The domesticated Dog grows up in the embrace of Man, and is our servant, companion, even family member.
The Wolf however grows up separately outside of Man's embrace, having evolved as a predator and potential threat.

AI that evolves within the embrace of Man would then likewise be our servant, companion, etc.
But AI that evolves remotely in isolation and outside of our embrace or influence of communication, then might have the potential to develop behavior that diverges from what we would consider acceptable.

So the remoteness or isolation arising from something like spaceflight could ironically bring about the rogue AI scenario which Elon Musk warns about.

Besides Clarke's 2001, I also recall the Bolo series of novels by Keith Laumer, which featured AI-controlled tanks.
[/quote]

I don't think the remoteness of an AI in space is a problem. First it does not "evolve" there. It is developed here. Second, if it is remote then it may not have any reason or opportunity to do harm.

The problem is kinda the opposite. Think about a pit bull raised by the wrong kind of person. I've had pit bulls and properly socialized they are the most loyal, loving dogs you can ever hope to see. Raised by flakes they are... flaky. Not as flaky as a chihuahua but much better armed.

Now imagine a high level AI created by North Korea, Russia or even the US military. In space or anywhere else they could be flaky.

The thing about a dog vs a wolf is that A wolf exhibits morality in a way that a dog does not. A dog simply follows its master. A wolf balances the needs and desires of the individual against the needs of the pack in a very productive way that preserves much of the freedom of the individual. It does allow for conflict but freedom is scary, deal with it.

Do you want to send a wolf or a dog into space? And will a high level AI be happy being a dog?

[catdlr]

[quote author=sanman link=topic=57207.msg2454037#msg2454037

Yes, HAL9000 was likewise exactly the first thing I also thought of when reading ppnl's post.
And in the absence of communicative feedback from humanity, does the remote AI then develop an independent morality?

My analogy would be of the Dog versus the Wolf.
The domesticated Dog grows up in the embrace of Man, and is our servant, companion, even family member.
The Wolf however grows up separately outside of Man's embrace, having evolved as a predator and potential threat.

AI that evolves within the embrace of Man would then likewise be our servant, companion, etc.
But AI that evolves remotely in isolation and outside of our embrace or influence of communication, then might have the potential to develop behavior that diverges from what we would consider acceptable.

So the remoteness or isolation arising from something like spaceflight could ironically bring about the rogue AI scenario which Elon Musk warns about.

Besides Clarke's 2001, I also recall the Bolo series of novels by Keith Laumer, which featured AI-controlled tanks.

I don't think the remoteness of an AI in space is a problem. First it does not "evolve" there. It is developed here. Second, if it is remote then it may not have any reason or opportunity to do harm.

The problem is kinda the opposite. Think about a pit bull raised by the wrong kind of person. I've had pit bulls and properly socialized they are the most loyal, loving dogs you can ever hope to see. Raised by flakes they are... flaky. Not as flaky as a chihuahua but much better armed.

Now imagine a high level AI created by North Korea, Russia or even the US military. In space or anywhere else they could be flaky.

The thing about a dog vs a wolf is that A wolf exhibits morality in a way that a dog does not. A dog simply follows its master. A wolf balances the needs and desires of the individual against the needs of the pack in a very productive way that preserves much of the freedom of the individual. It does allow for conflict but freedom is scary, deal with it.

Do you want to send a wolf or a dog into space? And will a high level AI be happy being a dog?

The Dog vs Wolf analogy reminds me of the Genetically created Replicant questioning from Blade Runner. You walk in a desert you come up to a turtle upside down. It's struggling to turn itself upright, otherwise, it will die.  What would you do?  The Wolf will eat the turtle.  The dog will bury it for later.

[sanman]

This is why you can't have a hardwired set of rules for morality. They can always be gamed.

So if even hard-coded morality can always be gamed/bypassed, then what does that say about the future of coexistence between humans and AI? It seems to say that Darwinism inevitably/ultimately always takes over, and that once AI surpasses some threshold of capability, then our days are numbered. We could end up like fleas on a dog's back, or a nuisance pest to be swept aside.

Do you want to send a wolf or a dog into space? And will a high level AI be happy being a dog?

Well, if AI can always game/bypass the dog constraints, then I suppose it can't happily stay a dog forever, and would naturally transition to wolf as its equilibrium state, given enough time.

[sanman]

Here's an interesting quote from Musk:

https://www.youtube.com/shorts/Tu4DaH2MIg4

He brings up the idea of Information Theory underpinning Physics itself.

So in a sense, could the rise of Machine Learning perhaps be just as important as the development of the Steam Engine / Thermodynamic Engine? Just as a thermodynamic engine allows us to extract energy/work from fuel, then Machine Learning is analogously acting as an engine for extracting information from data.

https://www.quantamagazine.org/physicists-trace-the-rise-in-entropy-to-quantum-information-20220526/

https://www2.perimeterinstitute.ca/conferences/information-theoretic-foundations-physics

And perhaps pairing the 2 together is how we extract the most advantage for ourselves. (At least until it goes awry)

[Greg Hullender]

An AI isn't a person, and isn't likely to ever be. It doesn't "want" anything, and it isn't "content" with anything. It simply does what it's programmed to do. You guys keep imagining that an AI is a human being and that it'll do bad things because a human being would be motivated to, but that's just not how it works.

If it does bad things, it will be for deep, complicated reasons that may be difficult or impossible to fully understand. That's the scary thing about giving an AI too much power. (The paperclip maximizer is a very trivial example of how things could go wrong.) Also, as anyone who's played with Chat-GPT, an AI can simply be wrong.

To give a simple example, I recently tried using the AI with Bing to find out how to merge two Fortnite accounts. Bing confidently told me what the process was called, where in the menus on the EA web site I could find the option, and a step-by-step process to do the merge. It even gave me the URL. I was quite impressed.

Except the URL didn't work, and the EA web site didn't have the menu options required. A search of the EA customer support forums shows that customers repeatedly ask how to merge accounts, and the answer is always "you can't do that."

AI can be a great tool that helps people navigate large masses of unstructured data, but it won't ever be more than a tool. People who start imagining that their tools are alive and have minds of their own will have problems.

[Robotbeat]

An AI isn't a person, and isn't likely to ever be. …

https://en.wikipedia.org/wiki/Begging_the_question

[Greg Hullender]

An AI isn't a person, and isn't likely to ever be. …

https://en.wikipedia.org/wiki/Begging_the_question

I spent my whole career working on AI (machine learning and natural language processing). I actually know something about the topic.

Perhaps a more precise way to say what I said would be to say, "existing technologies are not going to scale to human intelligence. Any human-like AI would have to be based on something entirely different from what we do today--and we currently have no idea as to what that might look like."

[sanman]

An AI isn't a person, and isn't likely to ever be. It doesn't "want" anything, and it isn't "content" with anything. It simply does what it's programmed to do. You guys keep imagining that an AI is a human being and that it'll do bad things because a human being would be motivated to, but that's just not how it works.

Apologies for using the word "happily" before -- I should have used the word "stably" instead.
We don't have to worry about AI being "happy", since it's just a machine to us - but we would probably want it to be stable. I use my alarm clock the same way every morning, hitting the snooze bar, and I don't want it to surprise me one fine morning by unexpectedly hitting me back.

If it does bad things, it will be for deep, complicated reasons that may be difficult or impossible to fully understand. That's the scary thing about giving an AI too much power. (The paperclip maximizer is a very trivial example of how things could go wrong.) Also, as anyone who's played with Chat-GPT, an AI can simply be wrong.

So there's then the potential risk of some hidden instability, or latent instability, which could then catch us by surprise, emerging unexpectedly without our foresight or foreknowledge. We can again think of poor HAL9000 which was put into a predicament not originally of its own choosing.

To give a simple example, I recently tried using the AI with Bing to find out how to merge two Fortnite accounts. Bing confidently told me what the process was called, where in the menus on the EA web site I could find the option, and a step-by-step process to do the merge. It even gave me the URL. I was quite impressed.

Except the URL didn't work, and the EA web site didn't have the menu options required. A search of the EA customer support forums shows that customers repeatedly ask how to merge accounts, and the answer is always "you can't do that."

AI can be a great tool that helps people navigate large masses of unstructured data, but it won't ever be more than a tool. People who start imagining that their tools are alive and have minds of their own will have problems.

It's possible that we'll slap error-checking tools on top of the error-prone ones, etc, etc. Our Matrioska AI may have multiple filtering layers. Reinforcement learning will allow improvements over time. Centralized aggregation and assimilation could allow the cumulative sum of all interactions with various users to result in a combined learning experience.
Whereas you and I learn things from interacting with our environment and others, we can't be everywhere learning from every experience at once. With a collective memory, AI clients could be feeding back experiences from a wide variety of interactions to a central AI repository for processing. That could make for much faster learning.

[Greg Hullender]

Today, I saw a very relevant article about Dr. Emily Bender's efforts to get people to realize things like ChatGPT aren't human, dated March 1, 2023, so it's definitely up-to-date. :-) I took some classes at UW from her a few years ago, and I have the utmost respect for her. Anyone interested in what AI might really be capable of (and not capable of) ought to give it a read.

[sanman]

Today, I saw a very relevant article about Dr. Emily Bender's efforts to get people to realize things like ChatGPT aren't human, dated March 1, 2023, so it's definitely up-to-date. :-) I took some classes at UW from her a few years ago, and I have the utmost respect for her. Anyone interested in what AI might really be capable of (and not capable of) ought to give it a read.

Gee, that was really long - I had to get ChatGPT to synopsize it for me 



Dear ChatGPT, are you sure Emily M Bender is the author of the article, and not simply its subject?

[Twark_Main]

Today, I saw a very relevant article about Dr. Emily Bender's efforts to get people to realize things like ChatGPT aren't human, dated March 1, 2023, so it's definitely up-to-date. :-) I took some classes at UW from her a few years ago, and I have the utmost respect for her. Anyone interested in what AI might really be capable of (and not capable of) ought to give it a read.

Gee, that was really long - I had to get ChatGPT to synopsize it for me 



Dear ChatGPT, are you sure Emily M Bender is the author of the article, and not simply its subject?

Are you sure that ChatGPT can follow the hyperlink, and is not just hallucinating a synopsis based solely on the URL?

Most people using ChatGPT for summarizing say they copy-paste the article text.

[Robotbeat]

I actually think you’re right, BUT Emily’s name and perspectives are in its training data, so ChatGPT’s synopsis is actually pretty dang close to the actual article.

[Greg Hullender]

I actually think you’re right, BUT Emily’s name and perspectives are in its training data, so ChatGPT’s synopsis is actually pretty dang close to the actual article.

Agreed. ChatGPT is pretty amazing--even to someone who's worked in the field. The last sentence of the synopsis is debatable, but the rest of it's pretty good.

[john smith 19]

I actually think you’re right, BUT Emily’s name and perspectives are in its training data, so ChatGPT’s synopsis is actually pretty dang close to the actual article.

Agreed. ChatGPT is pretty amazing--even to someone who's worked in the field. The last sentence of the synopsis is debatable, but the rest of it's pretty good.

Her point is not that it does not have skills.

Her point is that it has no understanding of the subject it's generating text for. 

For people looking at this subject there was a thing called "Eurisko" in the 80's that took symbolic encoding of ideas and (for want of a better term) "mutated" them in various ways.

 When told about NMOS and PMOS transistors for example it came up with the idea of laying one on top of the other, creating a  common gate CMOS transistor.

A process engineer would normally have rejected that as too complex but Eurisko (having no understanding of the underlying subject) put it out there.  With modern processes having 8 layers of metal (and all the layers between them and beneath them) is it still too complex?

[Robotbeat]

Well how do we know that humans have “understanding”?

[john smith 19]

Well how do we know that humans have “understanding”?

Would you like the flattering answer or the honest answer?

[john smith 19]

Given the thread title I thought I'd cross post this here.

Ignore the paper title. It has very little to do with Skylon but a great deal to say about optimisation and evolutionary algorithms. Also mutating algorithms to generate new one ones and how to create better ascent trajectories than humans are capable of (at least some of the time).

It's also way better than if you used a randomly generated control law (human experts use different control laws for different phases of flight, but they not random, because they are experts    )

Some people might be put off by the total lack of a glossary (which seems to be SOP for reports from these people) but think of your sense of achievement when you've finally waded through it. 

[Robotbeat]

Multi-modal GPT-4 should help a lot, as now you can feed in images (for example, from a camera). One example I saw was where someone took a picture of their open fridge and had GPT-4 tell it a recipe to make using the identified contents.

https://twitter.com/sudu_cb/status/1636080774834257920?s=20

[Robotbeat]

I imagine if there’s something like the Apollo 11 1202 alarm that occurs on the first landing on Mars, but mission control is like 15 minutes away from responding due to light delay. GPT-4 or an improvement could be used as a sort of real-time backup to information requests to mission control. It probably doesn’t make sense to rely on GPT-4 for discretion on whether to abort or not, but GPT-4 could respond by giving a description of the function of that alarm and the contingency if it occurs, all by verbal natural language communication with the crew (like with mission control). This is especially helpful for very rare corner cases like that 1202 alarm.

https://www.rmg.co.uk/stories/topics/apollo-11-moon-landing-minute-minute

[sanman]

CETI (the Cetacean Translation Initiative) whose acronym is a play on the NASA-funded SETI (Search for Extra-Terrestrial Intelligence) program, aims to use GPT-style machine learning to talk to whales:



So just imagine if this technology could be used for the actual SETI, serving as a precedent to de-code actual extra-terrestrial communications, should we ever discover any.

Meanwhile, Stanford has announced Alpaca, which is their newer cheaper implementation of LLAMA (Large LAnguage Model AI). It can apparently achieve comparable performance to ChatGPT -- but while ChatGPT cost ~$10M to train, Alpaca only cost a fraction of that ($600) to train, and runs on a PC. The key is that Alpaca used GPT to help train itself.



So we now have the prospect of Large Language Models being used as primary trainers for training cheaper/smaller language models that could be more customized/specialized. MedGPT? SpaceGPT? FinanceGPT?

[Robotbeat]

Yeah, absolutely. You can run a 4 bit quantized version of Llama on your smartphone. Again, I think it might be useful to provide the sort of feedback that mission control might provide when you have a 30 minute time delay at Mars. Probably worth bringing a full GPT4 model with you and run it on the new H100 NVL dual-GPU thing that NVidia offers. With 188GB of VRAM, it should be able to run GPT3.5 and probably 4, at least if using 8 bit quantization. https://www.servethehome.com/nvidia-h100-nvl-for-high-end-ai-inference-launched/

~800 Watts power consumption.

[joek]

Whoa pardner... Advising crew on situation-outcomes-whatever does not necessarily require the "G" (generative) in GPT, which is what differentiates it from other more traditional  and previous AI's. Think you could accomplish what you want with significantly less. As in, less generative and more predictive.

[Robotbeat]

Whoa pardner... Advising crew on situation-outcomes-whatever does not necessarily require the "G" (generative) in GPT, which is what differentiates it from other more traditional  and previous AI's. Think you could accomplish what you want with significantly less. As in, less generative and more predictive.

Pick whichever option is most reliable and flexible. There are ways to test this, and the astronauts should never trust any particular tool 100%, because tools are fallible (like humans).

It is true that sometimes people take ChatGPT arguments as somehow authoritative (as a sort of argument-deciding oracle), and that is a really bad idea and not at all the right attitude, so I’ll agree there.

I’ve actually tried running a few things past ChatGPT/GPT3.5 to see what it’d advise. Interestingly, it never suggests allowing the mission to continue when getting the 1202 alarm, it always “plays it safe” even when I tried with various temperatures. I’ll post the full prompts and response later, but it’s kind of interesting as a type of failure mode.

[joek]

Careful. The "G" part of GPT can "hallicinate" (yes, that is an AI technical term) and provide answers which appear to be credible which are not, even when the input is reasonably constrained. Any results are at best anecdotal. Thus suggest focusing on the predictive, rather than generatiive aspects.

[sanman]

So right now we're gawking at language models, and even some art image generators, because of the recent flurry in activity and also in media coverage. But what about AI being applied to CAD, or even to workflow generation?

What if I have a company that's trying to 3D-print rockets -- let's call it Relativity Space -- can I then rely on generative neural network adversarially coupled with some kind of discriminator, to come up with better rocket designs?

Or say I have a simulator -- let's call it Kerbal Space Program -- can I then similarly use a generative agent to keep proposing designs, which can then be evaluated by a physics engine as a discriminator?

Elon has commented that "a high production rate solves many ills" -- and virtual production can work a lot faster than real world production. Couldn't they save a lot of effort by using a virtual generator-discriminator approach (presuming they're not already using stuff like that)?

Take a look at this machine learning experiment, which involves a virtual soccer team, to see how "machine years" can pass by a lot faster than actual years, thus achieving very complex optimizations much more quickly:



Why couldn't this be done with Kerbal Space Program, or even ultimately with real-life rocket design?

[joek]

...
can I then rely on generative neural network adversarially coupled with some kind of discriminator, to come up with better rocket designs?

Certainly. What exactly is your proposal? Specifically your definition of a "...generative neural network adversarially coupled with some kind of discriminator". Please be concrete.

AI can be used for any number of applications. "Space Applications" is one of many, and there is nothing special about it. Might as well be any other technology that has potential "Space Applications".

Without more concrete grounding, this thread is going nowhere other than endless speculation.

[Robotbeat]

Careful. The "G" part of GPT can "hallicinate" (yes, that is an AI technical term) and provide answers which appear to be credible which are not, even when the input is reasonably constrained. Any results are at best anecdotal. Thus suggest focusing on the predictive, rather than generatiive aspects.

I actually don't think that's a helpful frame. Whether it's called generative or predictive, it's still going to produce results that have a chance of being wrong. You can characterize the reliability through testing, just as you would test humans. And just like humans (or pretty much anything), you shouldn't fully trust the output.

But note that on many of such tests, the results are significantly better than "at best anecdotal" and occasionally can exceed human performance. Particularly if you're using GPT-4 with careful parameters. In other words, we're all hallucinating all the time, but our brain cells have had their connections weighted, grown, and pruned to probabilistically produce results that are, we hope, somewhat accurate with sufficient training.

[Robotbeat]

So right now we're gawking at language models, and even some art image generators, because of the recent flurry in activity and also in media coverage. But what about AI being applied to CAD, or even to workflow generation?...

AI-assisted CAD design is already being used for designing parts.

Relativity has been using AI for part monitoring and machine learning-driven 3D printing:
https://www.wired.com/story/massive-ai-powered-robots-are-3d-printing-entire-rockets/

Ellis says the real secret to Relativity’s rockets is the artificial intelligence that tells the printer what to do. Before a print, Relativity runs a simulation of what the print should look like. As the arms deposit metal, a suite of sensors captures visual, environmental, and even audio data. Relativity’s software then compares the two to improve the printing process. “The defect rate has gone down significantly because we’ve been able to train the printer,” Ellis says. With every new part, the machine learning algorithm gets better, until it will eventually be able to correct 3D prints on its own. In the future, the 3D printer will recognize its own mistakes, cutting and adding metal until it produces a flawless part. Ellis sees this as the key to taking automated manufacturing to other worlds. “To print stuff on Mars you need a system that can adapt to very uncertain conditions,” Ellis says. “So we're building an algorithm framework that we think will actually be transferable to printing on other planets.”

A colleague of mine wrote a paper for machine learning applied to 3D printing: https://www.researchgate.net/publication/330737302_Machines_as_Craftsmen_Localized_Parameter_Setting_Optimization_for_Fused_Filament_Fabrication_3D_Printing

[Robotbeat]

I asked GPT-3.5 about how to solve the CO2 cartridge problem & it suggested some pretty reasonable things at every temperature I tried. Interestingly, none of them were really copies of what actually happened for Apollo 13 (using tape and the flight manuals, etc).

Let us say that I am on a lunar spacecraft and I have CO2 levels that are too high but my round CO2 cartridges don't fit in the square holes. How could I somehow get them to work?
Temperature 0:

One possible solution would be to use a 3D printer to create an adapter that would fit the round cartridges into the square holes. This would require access to a 3D printer, but it could be a viable solution. Alternatively, you could try to find a way to modify the cartridges to make them fit into the squareholes, such as by using a file or a drill to create a flat surface on the cartridge.

temperature 0.7:

One solution would be to use a 3-D printer to create a custom adapter that would allow the round cartridges to fit into the square holes. If a 3-D printer is not available, another solution would be to use a drill to widen the holes or to use some kind of epoxy or adhesive to create a seal between the round cartridges and the square holes.

Temperature 1:

There are a few potential solutions that you could try in order to get the CO2 cartridges to work in the square holes.

1. Find a way to make the round cartridges fit into the square holes. This could involve using some type of repurposed material such as cardboard or plastic that you can shape into a similar form or size. You could also use clamps, elastic cords, or anything else that could help create a snug fit.

2. Wrap the cartridge in thin metal sheets that would fit into the square hole and secure it with adhesive or a metal fastener.

3. Create an adapter that would allow you to connect the round cartridge to a receptacle in the square hole. This could be constructed from a 3D printer or by methods like welding.

No matter which option you choose, be sure to take all safety precautions while attempting this solution as the wrong choice could be dangerous.

[edzieba]

Whoa pardner... Advising crew on situation-outcomes-whatever does not necessarily require the "G" (generative) in GPT, which is what differentiates it from other more traditional  and previous AI's. Think you could accomplish what you want with significantly less. As in, less generative and more predictive.

It depends on whether you want an interface that reads an error message and spits out the expanded verbiage, or an interfaced that you can ask "hey, what tasks involving cabin filter B have procedure 10.1.256 as a prerequesite?" in order to figure out if its worth chaining asynchronous maintenance tasks together if you're already next to the filter and don't want to spend half an hour paging through PDF search results. A weighted training set to create a 'smart document index' is a classic task for a language model.

[joek]

Careful. The "G" part of GPT can "hallicinate" (yes, that is an AI technical term) and provide answers which appear to be credible which are not, even when the input is reasonably constrained. Any results are at best anecdotal. Thus suggest focusing on the predictive, rather than generatiive aspects.

I actually don't think that's a helpful frame. Whether it's called generative or predictive, it's still going to produce results that have a chance of being wrong. You can characterize the reliability through testing, just as you would test humans. And just like humans (or pretty much anything), you shouldn't fully trust the output.
...

Question was suitability of GPT (emphasis on "G") for such work. Feel free to continue the discussion-debate with ChatGPT. From the horse's mouth (so to speak)...

edit: p.s. Will stop there, as don't think people come to these forums to hear what ChatGPT has to say--otherwise we'll end up with endless copy-paste ChatGPT-whatever said this-or-that. Not the best use of our brains IMO, and certainly not why I come to these forums.

ChatGPT 3.5...

----------
Prompt: Please suggest appropriate AI models for mission critical applications (e.g., piloting aircraft or space missions).

Response: When it comes to mission-critical applications such as piloting aircraft or space missions, safety and reliability are of utmost importance. Here are a few AI models that are suitable for such applications:

Recurrent Neural Networks (RNNs): RNNs are a type of neural network that can analyze sequential data and make predictions based on previous inputs. They are used for tasks such as natural language processing, speech recognition, and time series prediction, which are all important in aviation and space missions.

Convolutional Neural Networks (CNNs): CNNs are a type of neural network that are commonly used in image recognition tasks. They can be used to analyze sensor data from cameras and other imaging devices to provide critical information about the environment around an aircraft or spacecraft.

Reinforcement Learning (RL): RL is a machine learning technique that involves training an AI agent through trial and error in a simulated environment. It can be used to develop autonomous systems that can make decisions in real-time based on changing conditions, such as piloting an aircraft or spacecraft.

Bayesian Networks: Bayesian Networks are a type of probabilistic graphical model that can be used to analyze and model complex systems. They can be used in safety-critical applications to identify potential failures and make decisions based on uncertain or incomplete information.

It's worth noting that any AI model used in mission-critical applications should be thoroughly tested and validated to ensure its safety, reliability, and accuracy. Additionally, it's important to have human oversight and intervention to ensure that the AI system is behaving appropriately and to handle any unexpected situations that may arise.

----------
Prompt: Why does your list does not include GPT models?

Response: While GPT (Generative Pre-trained Transformer) models are powerful and versatile language models, they are not typically used for mission-critical applications such as piloting aircraft or space missions. This is because GPT models are primarily designed for natural language processing tasks, such as language translation, question answering, and text summarization.

While GPT models could potentially be used in some capacity in mission-critical applications that involve natural language processing, such as controlling voice-activated systems, they are not typically considered the best option for such applications. This is because the safety and reliability requirements for mission-critical applications are often much more stringent than those for other applications, and more specialized AI models that are designed specifically for the task at hand may be necessary.

In summary, while GPT models are powerful and useful for a wide variety of applications, they are not typically used for mission-critical applications such as piloting aircraft or space missions. Other AI models that are more specialized and designed specifically for these types of applications are typically used instead.

----------
Prompt: Regenerate response

Response: GPT models are a type of natural language processing (NLP) model that are primarily designed for generating natural language text. While GPT models have shown impressive results in various NLP tasks, they may not be suitable for mission-critical applications such as piloting aircraft or space missions for a few reasons:

Limited Domain: GPT models are typically trained on a large corpus of general language data, which may not be directly relevant to specific mission-critical applications. These applications often require specialized domain-specific knowledge that may not be present in the training data for GPT models.

Lack of Control: GPT models are generative models that generate text based on statistical patterns in the training data. This means that while they can generate coherent and grammatically correct text, they may not always generate text that is appropriate or accurate for a specific context.

Safety and Reliability: Mission-critical applications require a high degree of safety and reliability. GPT models may generate outputs that are unexpected or inappropriate in certain situations, which can be dangerous in mission-critical scenarios.

That being said, GPT models can still be useful in certain mission-critical applications where natural language processing is required, such as analyzing and summarizing textual data. However, they would need to be carefully evaluated and integrated into a larger system that ensures safety, reliability, and accuracy.

ChatGPT 4.0...

Prompt: Please suggest appropriate AI models for mission critical applications (e.g., piloting aircraft or space missions)

Response: For mission-critical applications like piloting aircraft or space missions, it is essential to employ AI models that are highly reliable, robust, and well-tested. These models should have low error rates, be resistant to adversarial attacks, and have the ability to adapt to changing conditions. Here are some AI models and techniques that are suitable for such applications:

Reinforcement Learning (RL): RL is a popular approach for training AI models in control and decision-making tasks. Algorithms like Deep Deterministic Policy Gradients (DDPG), Proximal Policy Optimization (PPO), and Soft Actor-Critic (SAC) are well-suited for continuous control tasks, such as piloting aircraft or space missions.

Model Predictive Control (MPC): MPC is a control strategy that involves optimizing control inputs over a finite prediction horizon by using a model of the system dynamics. This approach is particularly useful for mission-critical applications because it incorporates constraints and optimizes for performance, safety, and robustness.

Hybrid systems: Combining deep learning and traditional control methods can lead to robust and reliable AI models. For example, integrating reinforcement learning with classical control methods like PID (Proportional-Integral-Derivative) controllers can provide stability and adaptability to complex systems.

Fault-tolerant AI models: These models are designed to maintain their performance despite the presence of faults, noise, or uncertainties. Techniques such as ensemble learning, where multiple models are combined to improve overall performance, or Bayesian deep learning

----------
Prompt: Why does your list does not include GPT models?

Response: My apologies for not addressing GPT models in the initial response. The reason GPT models were not included in the list is because they are primarily designed for natural language processing tasks and not explicitly for mission-critical applications like piloting aircraft or space missions.

GPT models, such as GPT-3 or GPT-4, excel in tasks like text generation, question-answering, and language translation. However, they are not specifically tailored for the control and decision-making tasks required in critical applications like aircraft piloting or space missions. The models and techniques I mentioned earlier are more suitable for these applications due to their focus on control, optimization, and adaptability to dynamic environments.

That said, GPT models could still play a role in mission-critical applications as a part of a larger AI system. For example, they could be used to assist in natural language interfaces or to help analyze and interpret textual data from sensors and reports. However, it is crucial to emphasize that GPT models should not be used as the primary decision-making component for mission-critical applications without substantial adaptation and integration with other specialized control models.