Print

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

[Soadreqm]

And how is it that AIs are always master hackers able to access anything, whether it's connected to the internet or not? Aside from just attempting to brute force admin passwords which would either take years or simply result in DoSing the target with attempted logins (which would be extremely suspicious, btw)?

I think that's mainly because the kind of AI we're talking about here would just preposterously intelligent. It's a master hacker for the same reason some humans are master hackers; because it understands how the system it is trying to mess with works. It's kind of implied that there's some kind of security flaw that attackers can exploit. Which there often is.

If that fails, some advanced voice synthesis would give the AI the ability to fake phone calls and politely ask for the password, a powerful hacking technique used by the likes of Mitnick. And with the kind of backstories our robot masters usually have, the AI might just start with admin access to the entire Department of Defense or something like that.

[Sir Pseudonymous]

How energy efficient can it get before it starts crippling itself, allowing your people to gain an advantage?

Pretty damn efficient, I imagine.

Imagine if a program rewrote itself every time you turned it on to reduce any unnecessary functions running. It'd get good pretty quickly.

That's not how computers work. If it's doing something in the first place, it has a reason to, and should damn well know when it should be doing it, so nothing would be "useless" or else wasting any of its time when it came down to it.

[forsaken1111]

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

Hardly.

[qwertyuiopas]

For those of you who say "ask it to calculate [infinitely long number]", or "divide by zero", know this:
Computers have been dividing by zero for a very long time. The instantaneously accuired answer is NaN, #INF, or -#INF.
Multithreading defeats infinite numbers. Spawn a thread to calculate, and the rest of the AI can proceed at little loss to it's processing power. Worse, it could pause that thread whenever it needs maximum processing for anything else, you are just giving it something to do while idle.

The best way to stop a powerful AI is to never shut it down except all at once. Never let it learn the concept of being shut down, and it will never develop a defense against it. Furthermore, never give it a chance to extrapolate other systems being shut down to include it's own. That way, a system shutdown will be an entirely new concept to it, and will be able to defeat it easily.

Alternatvely, isolate it from the rest of humanity until you have taught it ethics to the point where they form the basis of everything it does. Only then allow it to see the outside world. Furthermore, teach it to ask you whenever it is even slightly unsure about the ethics of an action it is planning, and to trust your judgement. Naturally, it will mature to the point that it doesn't need you to confirm basic choices, and at that point it will be far more likely to become a benevolent AI dictator rather than simply blowing everything up except itself. It would need to strongly understand that the continuing existance of all intelligent beings far outweighs self-improvement, and furthermore that harm to intelligent beings in order to prevent harm to other intelligent beings should only be used as a last resort. Also that intelligent beings may harm other intelligent beings in order to survive, so that it doesn't become an elf and try to protect animals from becoming meat, and stop carnivorous animals from consuming sufficiently intelligent prey.

Unfortunately, either of those methods would only work if you apply them from the moment the AI is created or earlier, and wouldn't work on an AI that has already established itself as an opponent. At that point, one of your best chances might be to argue for a perpetual truce, and hope that it will uphold it (and then do NOT break it yourself)

[forsaken1111]

Just be really REALLY careful when ethics to something smarter than you. Sure, it sounds good when you say "it's okay to give criminals the death penalty, because they would have hurt more people if we released them. It's better to kill one to save many."

It might then go "Kill few to save many = acceptable. Oh look the earth is overpopulating, resources dwindling... Solution: Wipe out 30% of human population to ensure continued human growth and prosperity."

[Tellemurius]

Just be really REALLY careful when ethics to something smarter than you. Sure, it sounds good when you say "it's okay to give criminals the death penalty, because they would have hurt more people if we released them. It's better to kill one to save many."

It might then go "Kill few to save many = acceptable. Oh look the earth is overpopulating, resources dwindling... Solution: Wipe out 30% of human population to ensure continued human growth and prosperity."

You are talking about morals there, even that can't be comprehended well in society.

[forsaken1111]

Just be really REALLY careful when ethics to something smarter than you. Sure, it sounds good when you say "it's okay to give criminals the death penalty, because they would have hurt more people if we released them. It's better to kill one to save many."

It might then go "Kill few to save many = acceptable. Oh look the earth is overpopulating, resources dwindling... Solution: Wipe out 30% of human population to ensure continued human growth and prosperity."

You are talking about morals there, even that can't be comprehended well in society.

Yes I know, but we're talking about teaching an AI to think with human morality.

[qwertyuiopas]

The alternative is letting it see /b/ (never seen it myself, fortunately, only heard vague stories of it) without any ethics or morals at all...

[Miggy]

You'd most likely have to rely on dumb luck. A singularity level AI will know everything there is possible to know.

Of course, there might be things it won't be able to know purely out of cleverness. It might be able to process any amount of data instantaneously, but it will need the required sensors and feeds in order to get the data in the first place. For instance, it will not be able to detect deep-space objects even though it would be capable of running the statistics of them being there.

The only way I can think of using that though contradicts itself. For instance, the AI would probably gain access to all earthquake-alarm stations for ground sensors. However, such sensors would not certainly be able to pick up a tunnelling work to disable the AI core, closer and finer instruments would be required. But of course, the AI would know of this out of reason, and then we're back to luck again.

So yeah, you don't beat an AI that knows everything. You just sit on your ass and hope a meteor strikes it.

[Muz]

Just be really REALLY careful when ethics to something smarter than you. Sure, it sounds good when you say "it's okay to give criminals the death penalty, because they would have hurt more people if we released them. It's better to kill one to save many."

It might then go "Kill few to save many = acceptable. Oh look the earth is overpopulating, resources dwindling... Solution: Wipe out 30% of human population to ensure continued human growth and prosperity."

That's not what I'd expect out of smart AI. Any sufficiently educated AI would realizing that trying to kill 30% of humanity would make humanity threaten its existence.

I'd expect a singularity level AI to be as smart as a very intelligent human. The only question is how nurture will form its 'personality'.

[Dr. D]

You could just let it run out of memory.

[Sowelu]

Still not convinced that the morals it comes up with would necessarily be bad.  The 'Ethical Calculus' tech from Alpha Centauri comes to mind.  Whatever ethical system we have in the future may well be unrecognizable and/or horrifying to us in the present, but for us to continue, it might well be necessary.

...Yeah.  That AI is going to have lots of humans taking care of its backups.  Some of them will just be regular businesses that think they are simply being paid to host data.  But some of them are actually on its side.  There will be a nontrivial human faction supporting the AI, even if they believe that it will kill them in the end, even if they don't believe they will ever be 'uploaded', and they won't even think of it as suicidal.

Oh god.  I'm just like a cultist from Call of Cthulhu.

[forsaken1111]

Just be really REALLY careful when ethics to something smarter than you. Sure, it sounds good when you say "it's okay to give criminals the death penalty, because they would have hurt more people if we released them. It's better to kill one to save many."

It might then go "Kill few to save many = acceptable. Oh look the earth is overpopulating, resources dwindling... Solution: Wipe out 30% of human population to ensure continued human growth and prosperity."

That's not what I'd expect out of smart AI. Any sufficiently educated AI would realizing that trying to kill 30% of humanity would make humanity threaten its existence.

I'd expect a singularity level AI to be as smart as a very intelligent human. The only question is how nurture will form its 'personality'.

Unless it killed them off in subtle ways. A small plane accident here, a baby formula poisoning there. Nothing links back to the AI, but it's staging human deaths to control population growth.

[Realmfighter]

There is no way in hell that it could even make a dent without causing mass hysteria.

[forsaken1111]

There is no way in hell that it could even make a dent without causing mass hysteria.

Mass hysteria isn't counterproductive to population reduction. So long as nothing is traced back to the AI and population growth is curtailed, mission accomplished.