Unmanned Mission to Alpha Centauri, A study of an unmanned mission to the Alpha Centauri system Options

[Vultur]

I don't really understand the need for massively complex AI. Sure, it can't be directed by humans, but if we're talking a flyby speed this high, the possible choices are VERY limited.

It seems like the flowchart you'd need would be like this:

If a planet/moon has oxygen or methane, go as close as possible to it using all instruments on planet.
If not:
If a planet/moon has an atmosphere at least Mars density but less than gas giant density, go as close as possible to it, use all instruments
If not:
Go as close as possible to planet/moon with radius closest to 8800 miles (Earth's), but take pictures of star as well rather than prioritizing everything to planet
If no planets at all:
Go as close as possible to star and use everything on it.

[JRehling]

[...]

[Vultur]

The greater the inherent problems of speed and sensing and rerouting, the less you have to worry about intelligence as the bottleneck in the system. It's very hard to make decisions using data you don't have.

Exactly. And I think short of sci-fi (read: not in the foreseeable future) technology, those will be the huge issues for something like this. A lightsail-launched probe doesn't seem that far beyond our current tech. Sure, the laser would be millions of times bigger than any built before, but the principles would be the same -- more an engineering issue than one of fundamentally new technology, rather like how we went from Explorer 1 to the Apollo/Saturn V system in a decade.

[Stephen]

A lightsail-launched probe doesn't seem that far beyond our current tech. Sure, the laser would be millions of times bigger than any built before, but the principles would be the same -- more an engineering issue than one of fundamentally new technology, rather like how we went from Explorer 1 to the Apollo/Saturn V system in a decade.

Hmm. I wonder how many times in the history of the space program that sentiment has been expressed before--only to have reality come crashing down with a loud thump when, after many years and the spending of large amounts of money, the engineering requird turns out to more complicated and more expensive than contemplated in the original proposal.

======
Stephen

[Greg Hullender]

Claiming a problem as being unsolvable "even in theory" is to claim that no solution for that problem exists. An obvious example would be trying to build a space probe which could ATTAIN light speed.

We have no physics that suggests how we could exceed light speed, and we have no theory in computer science that suggests how we could build a machine that thinks. Yes, we have the example that people do think (and no examples of anything moving faster than light) but we have absolutely no clue how people think. A lot of bright, talented people have wasted lots and lots of time trying to build reasoning algorithms inspired by how they imagined the human brain to work, and, so far, that work is a frighteningly complete failure.

Giving up on the idea of building a machine that can think is the first step to constructing useful systems in this area.

Actual working systems today to do things like speech recognition, handwriting recognition, machine translation, etc. use highly mathematical, highly statistical, explicitly non-organic algorithms. These things do not simply scale up into human intelligence, though. Even with unlimited hardware and unlimited raw data, none of these things is going to start to reason. They just do a better job at their narrow task.

If you want to get some idea of the state of the art, Mitchell's "Machine Learning" http://www.amazon.com/Machine-Learning-Mcg...t/dp/0071154671 is an excellent graduate-level text for general machine learning, and for the natural-language end of the problem, the recently-published second edition of Jurafsky and Martin's "Speech and Natural Language" processing http://www.amazon.com/Language-Processing-...e/dp/0131873210 is hard to beat. A more difficult, but perhaps more to-the-point text would be Duda, Hart, and Stork's "Pattern Classification" http://www.amazon.com/Pattern-Classificati...a/dp/0471056693.

Speculating about AI work when you have not read any of the basic materials is a lot like speculatinig about space travel when you don't understand the rocket equation. This is a hard problem, and lots of really smart people have worked on it for a really long time. I believe it will help make a contribution to unmanned space flight through algorithms such as the EO-1 used, but no one will confuse these with anything resembling actual human thought.

--Greg

[JRehling]

[...]

[djellison]

Want to get personal - take it elsewhere. First and final warning for all. No debate.

[Mark6]

This seems like an exercise in continual obsolescence. Whenever you launch something like that, you're tempting fate that long before it arrives, you would have a faster way to get there.

I think a good rule of thumb is - never launch anything that would take more than 50 years to reach its destination. TAU (Thousand Astronomical Units)^* mission is thus a valid objective. By the time it reaches Inner Oort Cloud, it will be obsolete, but not ridiculously so, and it will do good science along the way. A "50 year probe" every generation (say, every 25 years) would be a good incremental buildup to a true interstellar mission.

^* Horrid choice of background in the web page!

[stevesliva]

Does the 50-year rule assume only advancements in propulsion, or also advancements in remote sensing?

Of course, I can think of counterexamples that make remote sensing less of an issue. So what if Voyager had taken fifty years to reach Neptune? Still would be the best photos of Neptune we've got. If Pioneer had taken fifty years to photograph the Galileans, though...

[Mark6]

Does the 50-year rule assume only advancements in propulsion, or also advancements in remote sensing?

Of course, I can think of counterexamples that make remote sensing less of an issue. So what if Voyager had taken fifty years to reach Neptune? Still would be the best photos of Neptune we've got. If Pioneer had taken fifty years to photograph the Galileans, though...

Sorry, but I do not understand your point. Yes, remote sensing could make a probe obsolete before it reaches its destination, but I thought the main discussion was "How to develop intertsellar capability incrementally without building white elephants doomed to be overtaken within their lifetime?" Also, some things can not be done remotely even in principle. Looking at your target from multiple angles, for example.