Print

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

[Zrk2]

Or, it will just shoot us all. With science!

[ein]

Tell it to calculate the exact value of pi. It will eventually run out of memory and crash.

Or it will be smart enough to tell you that's impossible and dismiss the question.

Or it will say "Error, unauthorized user. Disregarding command."

Write a program that forces it to do so.

sudo calculate pi

[forsaken1111]

Tell it to calculate the exact value of pi. It will eventually run out of memory and crash.

Or it will be smart enough to tell you that's impossible and dismiss the question.

Or it will say "Error, unauthorized user. Disregarding command."

Write a program that forces it to do so.

sudo calculate pi

"Shut up and do it yourself, meatbag."

[Shinziril]

If this is a "soft-takeoff" scenario (I.E., it's not very superintelligent, and not improving very quickly) you can probably get away with incinerating everything involved.  Enough entropy will destroy things without any possibility of retrieval, even with perfect information.  Make sure to use a GOOD incinerator.

If it's a hard-takeoff scenario, where it becomes much smarter than human very very quickly, you probably lose.  At most you might see a glitch in the system temporarily, which clears up an instant later (because it's hiding its new capabilities) and then everything explodes at an unspecified time in the future before you can do anything about it.

Humanity will end up being destroyed not because of any inherent dislike the AI has for humanity, but because we (like everything else in the solar system) happen to be made out of atoms, which it could be using for other purposes.  Like calculating pi by converting the entire solar system into nanocomputers, because you told it to calculate pi.  Oops.

That could even happen with a benevolent AI that doesn't have a proper system of morals installed.  The most efficient way of figuring out pretty much anything is to convert the solar system into nanocomputers.  Let's just say that the "Three Laws of Robotics" aren't going to cut it here; you're basically going to have to design a human's entire value system, from scratch, plus editing to make some of the bits nicer (since a decent bit of a human's value system isn't desirable in something with as much power as a hard-takeoff AI).  Have fun.

[Max White]

Or, it will just shoot us all. With science!

It's a trailed and proven method.

[forsaken1111]

Humanity will end up being destroyed not because of any inherent dislike the AI has for humanity, but because we (like everything else in the solar system) happen to be made out of atoms, which it could be using for other purposes.  Like calculating pi by converting the entire solar system into nanocomputers, because you told it to calculate pi.  Oops.

Funny you mention that, I once played in a d20 future roleplaying game where robots took over the solar system.

Of course our group was playing as the robots.

We 'accidentally' destroyed earth, so spent the next several hundred years 'reformatting' most of the inner solar system and asteroid belt, and even dismantled most of Jupiter, to build a Matrioshka Brain so that it could calculate the universal space/time constant (some plot mcguffin, not anything real) so we could develop a time machine and return to the exact point in time to shoot ourselves in the head and allow events to resume along the original storyline.

It was fun.

Or, it will just shoot us all. With science!

It's a trailed and proven method.

Video is blocked.

[Burnt Pies]

I wondered why I shouted SCIENCE! like that. Now I know.


Back on topic, a roving group of space-pirate-aliens show up, and, noticing the plight of the people on the planet, use their super advanced laser awesome weapons to shut down the AI?

Then steal everything of use to them. They are, after all, pirates.

[Googolplexed]

Well, the question is, how much control does the AI have....

If an AI like that started on a computer not connected to the internet and with almost no I/O, and surrounded by xenophobes......

[Shinziril]

Trust me, "almost no I/O" is more than enough.  There has been an experiment run that shows that it is possible for a HUMAN- not a superintelligent AI, but an ordinary human pretending to be one (if an intelligent and slightly arrogant human)- to convince someone to release the dangerous AI using a text-only terminal. 

You could give it no I/O capabilities at all, but A) Why are you building a highly expensive brick in the first place?  and B) It could probably figure out a way around that anyway, like pulsing power through its electical circuits to generate radio waves and affect nearby electronics.  Remember that it thinks both better AND faster than you do.

The best way to contain it is to program it right the first time around, so that you're okay with it having UNLIMITED POWAAAAHHH!!!, because you know it won't use it in any manner that you wouldn't like.  While it SHOULD be possible to do this, unhelpfully, you don't get second chances. 

[forsaken1111]

Might be safer to make a whole bunch of less intelligent AI who operate in some form of blind bureaucracy so that most of them have no idea of the bigger picture but they still get things done faster than a person would. Obfuscation and oversight are the way to go.

[GlyphGryph]

You guys are all getting hung up over the appearance of a really strong AI.

But not one thats singularity level. You see, a singularity level AI would never have to conquer humanity. We give it control quite willingly.

The only way to defeat it would be to have a group of incredibly dedicated individuals who do not care that the entire world will hate them for what they are doing, They spend years honing their highly developed weapons to attack vulnerabilities in the AI, and set them in motion by manipulating data streams and certain small scale events, and maintaining utmost secrecy. The AI is growing exponentially at this point, but the vulnerabilities aren't a way to destroy it, but to to manipulate that growth. To send it awry, and open up a true weakness down the road.

Then a group of heavily armed humans fight their way into the AI's central compound, disabling its security systems, and they talk to it.

And they convince it that we would be better off were it no longer around.

And because of the things those humans have done before to convince it of this, it believes them, and it either destroys itself or it leaves.

And everybody hates their new "heroes" as they try to rebuild society in a world where its foundation has been ripped away.

[Heliman]

Only one way to destroy a singularity level AI.

[Burnt Pies]

We haven't mentioned Magma yet.

[Shinziril]

Well, that's one that was properly programmed, obviously, glyphgryph  . 

"Fighting" the AI really isn't the way to go.  If you program it right, one of its main desires (indeed, its primary desire) will be to aid humans, and because it KNOWS that it might not have the right idea, it'll ask for all sorts of confirmation and explanations before doing anything that might have unwanted consequences.  It'll even ask you to look over modifications it wants to make to its moral system, to ensure that it doesn't lead to any truly unhelpful consequences.  A higher-level one might even safely simulate having random bits flipped or hostile viruses implanted or portions of code deleted to ensure that its morality programming is strong enough to withstand damage. 

If the AI stops wanting to be helpful, and wanting to ensure that what it thinks is "helpful" is actually helpful, you've already lost.

Stick to pre-programmed expert systems with no self-improvement capabilities if you want to be "better safe than sorry". 

[Sowelu]

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?