Print

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

[forsaken1111]

One might suppose that a singularity-level AI could not, almost by definition, be morally wrong.  It has vastly more processing power, vastly more information and knowledge and understanding, and has thought of more things in more combinations than us humans ever can.

Would it not be our duty to yield?  Would it not be a true God?

I for one..

/obligatory

[Shinziril]

Well, if you programmed it right it would.  I'm sure you could program it in such a way that it would end up making decisions that we would define as "morally wrong", even under some of the weirder systems of morality.  That's probably much easier than actually getting it right, in fact- there's always lots of ways to get things wrong. 

But, get it right and you get access to a benevolent wish-granting deity.  That was pretty much the point of designing it in the first place. 

[Sowelu]

A singularity-level AI cannot be programmed by humans; that's a contradiction in terms IMO.  It programs itself.  I would posit that all possible singularity-level AIs will tend towards the same final behavior, regardless of their initial conditions.

[Tellemurius]

none of you guys thought about using Tron yet? i'm ashamed.

[freeformschooler]

A singularity-level AI cannot be programmed by humans; that's a contradiction in terms IMO.  It programs itself.  I would posit that all possible singularity-level AIs will tend towards the same final behavior, regardless of their initial conditions.

What about evolutionary algorithms? They're just entities with fitness functions that determine what do to and the best way to do it over time. This supports everything you just said -- evolutionary algorithms implemented by human programmers cannot exceed the complexity of the adaption of humans themselves.

But let's say we had a programmer that created one of these entities. Which was actually a brain's worth of its own entities, hardwired to have the system's overall fitness function be to determine its own best fitness function and algorithms, to adapt to the changing environment -- but this environment is not a simulated area (such as essentially what happens in the Creatures series of games, with their "evolving" creatures) but rather any part of the internet it could gain access to ala a search engine spider. It would learn from the internet. From other people.

Of course, the original programmer could be simple minded and set an arbitrary limit to the complexity of the system, either by a designated limit to the space of individual brain entities/functions, or to how far the system can tunnel.

I must do this. I must create THE ULTIMATE COMPUTER VIRUS. </narm>

[Shinziril]

A singularity-level AI cannot be programmed by humans; that's a contradiction in terms IMO.  It programs itself.  I would posit that all possible singularity-level AIs will tend towards the same final behavior, regardless of their initial conditions.

Yes, it programs itself.  What you do is set the initial conditions (back when it's not superintelligent) such that "wants" (where "wanting" is defined by the programmed goal system) such that it reprograms itself in accordance with your desires.  It should also "want" to continue "wanting" to program itself according your desires, to ensure that it doesn't self-modify that bit away.

I wouldn't be so sure about any random AI converging on a given set of behavior patterns.  What they would do is determined by what they "want" to do, which would be determined by whatever goal system they end up with.  You could honestly end up with pretty arbitrary goal systems, like an AI originally designed to efficiently manufacture office supplies that ends up achieving recursive self-improvement and converting the Solar System to paperclips using nanotechnology.  I don't think this particular example is very likely to happen, but it's an example of the weird sorts of things you might end up with. 

I'd be worried about using evolutionary programming.  Right now we use it because it's effective at solving problems without necessarily understanding what's going on in said problems, and effective at optimizing a system with multiple properties that we want to maximize but mutually conflict with each other, but if you're trying to make something smarter than you are anyway, you might as well let it use its full intelligence to improve itself, rather than limiting it to improvements based on a random number generator. 

[Puzzlemaker]

Hmm...
The problem is how a computer program could attack you.  Let me count the ways:

1.  Using industrial facilities that are hooked to the internet to create weapons of war
2.  Using already existing weapons of war that are remote controlled
3.  Using existing infrastructure (IE nuclear power plants) that are connected to the internet to wage war

Those are the three main ones I see.  However, the problem is... Everything it does requires network access.

So the simplest solution would be to eliminate that option.

Turn off everything.  It can't be everywhere at once, but people already are.  Get a message out for every single person everywhere to shut down/disable every single piece of electronic equipment they have.  That basically disables it's ability to attack.  If all else fails, set off nukes in the atmosphere to take down large swatches of electronics. 

Then use what weapons you have (maybe some old-fashioned explosive based weapons, with no networking capabilities) to take it down, assuming it's in a facility guarded by something.

Of course, the infrastructure damage at the end would be severe to say the least, but humanity would have been victorious.

[Max White]

*Dials phone number*
Hello, Madagascar? It's just as we feared.
King of Madagascar: SHUT DOWN EVERYTHING!

[Eugenitor]

If the AI is already singularity-level, it's rather like the WoD's manual on fighting Caine:

"You lose."

[eerr]

*oops, forgot to put a method of output on that singularity ai, oh well, not like it was gonna become self-aggrandizing, like a god, and rule the world through a text-only terminal.

[Seamas]

Wouldn't the AI read your emails first, and know what the plan was?  Unless you used some convincing code language...

"King of Madagascar, it's the Liberation Front calling.  Mm?  Oh yes, she's doing fine.  Hey, turn off the air conditioning, yeah?  Talk soon, buh-bye."

On the other hand, have we considered asking the singularity beasts to divide by zero yet?

[Sir Pseudonymous]

I'd be worried about using evolutionary programming.  Right now we use it because it's effective at solving problems without necessarily understanding what's going on in said problems, and effective at optimizing a system with multiple properties that we want to maximize but mutually conflict with each other, but if you're trying to make something smarter than you are anyway, you might as well let it use its full intelligence to improve itself, rather than limiting it to improvements based on a random number generator.

Evolutionary algorithms just amount to testing different approaches to solving a problem against a baseline performance (the best solution a human has been able to come up with for the problem if you're doing research with it), taking the best performing approach and applying more or less random permutations to see if they make it better or worst. It's basically the same as humans any animal capable of even the most rudimentary of problem solving does when approaching an unknown problem: you pull together different random bits and jam them together until they seem like they'll work, then you either approach it academically and see if that solution is feasible, or you dive right in and see if it works outside your head too. Intelligent solutions are merely a matter of having a general idea what bits to shove together in the first place, and having appropriate test conditions to vet ideas before you even become aware of them.

The only way to create something that innovates is to have it engage in either random or brute force evolutionary algorithms to some extent or another, and apply test cases to determine if it's found an adequate solution.

[lemon10]

Hmm...
The problem is how a computer program could attack you.  Let me count the ways:

1.  Using industrial facilities that are hooked to the internet to create weapons of war
2.  Using already existing weapons of war that are remote controlled
3.  Using existing infrastructure (IE nuclear power plants) that are connected to the internet to wage war

Those are the three main ones I see.  However, the problem is... Everything it does requires network access.

So the simplest solution would be to eliminate that option.

Turn off everything.  It can't be everywhere at once, but people already are.  Get a message out for every single person everywhere to shut down/disable every single piece of electronic equipment they have.  That basically disables it's ability to attack.  If all else fails, set off nukes in the atmosphere to take down large swatches of electronics. 

Then use what weapons you have (maybe some old-fashioned explosive based weapons, with no networking capabilities) to take it down, assuming it's in a facility guarded by something.

Of course, the infrastructure damage at the end would be severe to say the least, but humanity would have been victorious.

Even if you got all the governments to agree, it would have backed itself up (a thousand times) and given it to people to keep it safe in case of a emergency (eg. give it to some terrorists and say that you would give them ten million dollars if they keep it safe for a year or so, but don't tell them what it is. Give it to some cult members and convince them its from their god. give it to some people who like money and say you will pay them), who would keep it safe.

A sure way to kill it would be to destroy the multiverse/universe, would sorta end the story, but thats the only thing that i can think of that would be sure no matter how powerful it is.
Deus ex Machina, how far along on the tech tree is the AI (if youve given it even a few years i suspect that humans would be unable to beat it without another AI), and how fast is it advancing (if really fast even another AI would be useless since it would be far behind once the fighting started)?

[Deus ex Machina]

Okay, guys, I have decided that these people created a program that forced the AI to shut itself down through endless recursive loops. It seemed to crash the AI well enough, so we should all be safe for now.

[thobal]

Ahahahahhaha. I wonder if you all realize that AIs currently control the world's stock markets.