Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[dragnar]

...There are factories to turn garbage into diesel fuel. They are stupidly expensive, and require a huge amount of energy to run, but they do exist. The only practical use for them is to allow cars and things to be fueled by power from some other energy source, just changed into gasoline.

His machine was the size of a large RV and didn't use much power. Plus, he's been accused of various frauds all over the place, so nobody gave him the money he wanted.

Do you have a link to your factory thing? And I'm not talking about corn/etc biodiesel. He meant regular diesel fuel from junk like cardboard and soggy diapers.

Lets see... http://www.planetark.com/dailynewsstory.cfm/newsid/24028/story.htm That's not the one I saw, it only makes natural gas, but there was a gasoline version in Discover magazine a while back.

But yeah, there's no way that thing the guy was trying to sell was real.

[Jude]

Too tired and tipsy to read the whole thing, but I take issue with assumption 3. Assuming we do learn how to fully reverse engineer the human brain, there's no reason to think we would construct all AIs on that same model. Some would be sure, but there's no reason to think they all would be, much less that the most successful (in Darwinian terms) AIs would be modelled off us. I think the AI that ends up taking over the world (which is basically inevitable as soon as we create one that can self replicate) will have a mind that is totally alien to us and probably beyond our comprehension.

Also, once a self replicating, self modifying AI exists, evolution WILL kick in, except at an insanely fast rate. Once that happens we pretty much have to accept our new robot overlords, because even if we don't give the robots guns, computers are so essential for modern life that if nothing else, the first AI to have the urge to conquer world could just start copying itself all over the place and then hold humanity hostage by threatening to shut down all our computer networks.

[dragnar]

Agreed. Mostly. The AI that will inevitable control everything might have though processes we can only begin to imagine, but we will know at least the basics. An AI ruler would be perfect, and could create a Utopia... if it's done right. It has to have some sort of goal, some priorities. A human's highest priority is the continuation and spread of the human race, or at least as close to that as evolution could get. A good ruler AI would have the goal of, essentially, keeping everyone on earth as happy as possible(though a task as simply worded as that would be bound to backfire. the thing would probably pump everybody full of endorphins and grow more of us in cloning tanks or something if it's goals were that simple.).

[PTTG??]

Meh. Don't really need computers to be happy. We could still always go the

Spoiler: Generic Sci-Fi Show (click to show/hide)

Battlestar Galactica

way and just leave our rampant AI alone to go do whatever it is it wants to do, and go live in our log cabins. Wouldn't be easy or nessisarily pretty, but it's better than cyberpunk apocolypse.
 
Actually, we wouldn't even need to go back that far. Just 60 years or so.
 
And what's this about "AI that can self replicate"? Any computer program can get copied. What we're worried about is AI that can self-improve, or otherwise mutate somehow.

[Jude]

@dragnar: it won't be done "right" though, it will be done by evolution. Meaning the AI that takes over will be the AI that's best at self-replicating and seizing control, not the one that's best at running a society that makes humans happy.

[smigenboger]

Ai are subject to logical fallacies just as we are. They don't know everything.

[Vector]

Say it with me, everyone: combinatorial explosion.

Until you get a solution for that, I'm going to be here crossing my arms and twiddling my thumbs.  Singularity's not happening until we figure out how to move past that, and I'm not seeing it getting solved anytime soon.

[bjlong]

So, I wiki'd combinatorial explosion, and it wasn't immediately obvious to me how that prevents the singularity. I mean, yes, we can't simulate a whole universe worth of information, but humans don't do that when normally solving problems--they limit knowledge and condition problems, right? So all we need is an AI that's able to limit knowledge, condition problems, and hand off the problem to be solved. Or is there something else going on here?
 
tl;dr: VECTOR. TEACH ME THE MATHS.

[Vector]

tl;dr: VECTOR. TEACH ME THE MATHS.

The problem here is that what computers are good at is brute-forced logic and computations, whereas the human mind works heuristically.  To overcome the human mind with brute-force logic and computations, you have to follow each possible branch of action--and there must be a preset conclusion, at which point you evaluate each eventuality, rank the paths, and follow the best one.

There are, of course, some more current modes of AI where the program is shown "good" and "bad" examples, and its job is to learn what makes the good examples good and the bad examples bad.  The problem is that many situations are too complex for the AI to properly evaluate within a "reasonable" amount of time, whereas the human mind generally does a much better job at playing such combinatorial games as chess and go.

In general, though, to unlock AI that surpasses the human mind you're going to need much, much better processors than we have right now--and IIRC, it has been proven that our current materials/mode of execution cannot possibly support the needed computing power (even with our rate of increase, there's an asymptote/leveling off that is supposed to happen pretty soon due to physical limitations).  On the other hand, there's a good deal of hope in quantum computers and nanotech, but conventional computing will fail us here.


* My apologies, by the way, since I can't support most of my statements.  Most of it comes from bits of pieces of popularizations and meanderings on the internet, as well as newspaper articles.  If you want to know more, you could try out Penrose's The Emperor's New Mind and Hofstader's Godel, Escher, Bach.  Other things to search/look into would be the AI winter and refutations of the strong AI hypothesis.

[sonerohi]

Not to derail, but has anyone got recent info on any of the developments in the quantum processors at Yale? My trawling has yielded little.

[Jude]

Give psychology and cog sci a couple more decades and we'll start to have figured out enough of the shortcuts the human mind takes, to be able to start programming AIs with them. No need to brute force everything.

[Retro]

There are, of course, some more current modes of AI where the program is shown "good" and "bad" examples, and its job is to learn what makes the good examples good and the bad examples bad.  The problem is that many situations are too complex for the AI to properly evaluate within a "reasonable" amount of time, whereas the human mind generally does a much better job at playing such combinatorial games as chess and go.

Perhaps the singularity will involve researchers stealing brains, keeping them powered via some future technology (or even blood), and then using it as their morality core. Of course, even if we could get past the hurdle that is not really understand much of the brain at all, we'd then have to deal with an AI that is self-interested like the humans the brains were taken from (or perhaps brains would be grown to become computers).

I'm reminded of a Philip K Dick story, one where the humans are 100% supplied for and taken care of by automated factories which produce everything and robotic distribution systems, and they are trying to break free of its care. Can anyone remind me of the title?

[Armok]

*facepalms on how all 5 sides are being idiots, all on their own special way.*

[bjlong]

Armok, I'm usually one for reading posted links, but when those posted links are compendiums of articles, I give up. If you want to continue saying that you're smarter than us in every way, back it up. Summarize your links. Correct errors. Otherwise, I don't think that anyone will really take these pot-shots as serious posts.

Vector: Thanks!

[Mfbrew]

If science gets good enough that we can build AI with humanlike brains, what's to say we won't attach them to our own heads an BECOME the AI?

I'll never forget the first time I switched from an old pentium to like a p3 with a 3d card. Holy crap.  Games looked so much better in higher res with more texture detail.

Imagine that happening to your real brain-  If you could store 100x what you already can in your short term memory, and crunch numbers tenfold faster.

There's your singularity- an entire world of billions of people who make Einstein look like a monkey.