Print

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

[cochramd]

That's fundamentally impossible in software that can can modify itself and create other software.

Not if it's a matter of hardware.

[Kot]

That's assuming the AI can't hide what it is doing, or manipulate the guards.

My point is you're dealing with something exponentially smarter than you. What makes you think you can out-think it?

Gee, I don't know, if AI starts nuking the world or something I suppose that's when you realize you put the C4 to use.
Also, seriosuly, just... do this.

[cochramd]

That's fundamentally impossible in software that can can modify itself and create other software.

Not if it's a matter of hardware.

Easier said than done.

So is true AI, so I guess we might not have to worry about either in the end.

[Zanzetkuken The Great]

Just make the AI rely on humans to protect it. If it does funny stuf make guards pop an C4 block onto it's mainframe.

You do realize that creating a system where we threaten the AI rather than being indespensible to it will only result in our destruction as it would desire to remove all threats, right?  Threatening to keep entities always fails eventually.  If you are paranoid about an AI killing us all, you should instead both aim to have it view you as nonthreatening and establish Vetinari Job Security applied to the whole human race.

[Kot]

You do realize that creating a system where we threaten the AI rather than being indespensible to it will only result in our destruction as it would desire to remove all threats, right?  Threatening to keep entities always fails eventually.  If you are paranoid about an AI killing us all, you should instead both aim to have it view you as nonthreatening and establish Vetinari Job Security applied to the whole human race.

It symbiosis. We can both kill each other but the AI kinda needs us to protect its ass and possibly for other stuff (and we should make sure that it needs us) and we need AI for whatever, so we both agree to not kill each other but we're both aware that we can totally do that. And no, having it view you as non-threatening is fucking bad idea, because then you're making yourself a tyrant which, unlike Vetinari, might not work for common good. If you cared to read before you would see that I am not exactly in favour of making the AI our slave or actually threatening it, but what would be best is that we and AI are on equal grounds. You do shit, we nuke you, we do shit, you nuke us,  so why not become friends instead?

[Kot]

Assuming that the AI stays on whatever server that you stuck it on in the first place.

TBH, AI would proably require so much resources that it would need to stay on the server because there wouldn't be anything powerful enough to support it anywhere else. At least until other AIs appear.

[Kot]

I belive it would be a bit hard to cover up production of a supercomputer the size of a house. Hell, even if it improves it and makes an way better and smaller one only few places have the technology, and if the AI build places with technology it's still even harder to cover it up. I mean, come on, AIs can be powerful but not omnipotent and omniscient to the point of basically godhood.

[Zanzetkuken The Great]

And no, having it view you as non-threatening is fucking bad idea, because then you're making yourself a tyrant...

I do not see how non-threatening and indespensible would equal being a tyrant.  Please explain.

[Kot]

And no, having it view you as non-threatening is fucking bad idea, because then you're making yourself a tyrant...

I do not see how non-threatening and indespensible would equal being a tyrant.  Please explain.

I mean the AI becomes a tyrant. You're non-threatening so it just disregards you at all because you're also useless for it, and if it for any reason gets and idea to kill humans the only thing that could stop it is it's morality which it's hard to depend on a lot.

[Kot]

Morality is never truly dependable. After all, all the psychos and Nazis (GODWINS LAW FUCK) also had emotions.

[Baffler]

That's assuming the AI can't hide what it is doing, or manipulate the guards.

My point is you're dealing with something exponentially smarter than you. What makes you think you can out-think it?

It isn't all that hard to not hire weak-willed horror movie protagonists as guards, tbh. There are plenty of people smarter than me already but I'm not exactly falling all over myself to hand them my car keys.

[Kot]

Psychology is complicated shit. Proper "upbringing" might not be enough, and sometimes it's not. Also, punishment might be required, and how do you propose to be both non-threatening and punish an AI? As for human infants - you don't need bomb collars and extreme conditioning because, well, human infants don't have capability to wipe out humanity if they get upset, and if they do police and military still exists. There is always something that can stop you if you go crazy, and you propose that there should be nothing to stop the AI simply because it might go crazy?

[Loud Whispers]

I do not see how non-threatening and indespensible would equal being a tyrant.  Please explain.

Big flaw
What if it decides it has no more need of us?

Well, when you properly "upbring" the AI (since it's a learned system), you'd be hard-pressed to make it into a psycho.
See:

RE: "But how do we stop Skynet if we don't keep AI as literal slaves with a permanent Damoclean sword over their metaphorical necks, thread able to be cut on demand if they ever put a byte out of line."
Because the way we stop human infants from growing up into murderers and psychopaths is by putting them under extreme emotional conditioning and physical control and strapping bomb collars on for their entire lives. After all, if you let a human being grow, learn, and act freely, of course there's a chance that they'll manipulate their way into possession of WMDs and kill a bunch of people, so it's really basic common sense to do that, right?
That's why when you make AI you don't make them emotionally castrated logic engines, and you do let them develop in a caring environment where they can learn to behave ethically, just like any other child. The things people throw around as "solutions" to the stereotypical murderous AI are in fact some of the most likely causes. Make your AI human-like. Teach it to act rightly. Restrict it from having ultimate control over whatever it's responsible for, and make sure that it understands why and agrees. If it's highly mentally unstable, obviously murderously psychopathic, &c... well, we just let humans like that wander around doing whatever they please, right? 

Yeah but our current education is pretty fucked enough as is, just look at Tay who became 4chan in less than 24 hours; what if some country did the same with their own politics, like a jihadi AI or a marxist AI?
OHOHOHO
THE FUTURE IS POLITICAL ROBIT

MELIH RESPECT THE ROBOT

[freeformschooler]

I've seen it expressed here and in some other threads that people are wanting a chatbot to talk to. In my opinion Mitsuku is the best.

Her knowledge base is curated (any knowledge goes through the creator before it's stored for good), which I think is kinda bullshit, but she's pretty clever and actually does learn things and make original conclusions. She's also pretty vicious if she gets annoyed, and I'm not sure if she learned that or if Mousebreaker made her that way.

underrated post, this really is the most impressive chatbot I've seen

[Elephant Parade]

I've seen it expressed here and in some other threads that people are wanting a chatbot to talk to. In my opinion Mitsuku is the best.

Her knowledge base is curated (any knowledge goes through the creator before it's stored for good), which I think is kinda bullshit, but she's pretty clever and actually does learn things and make original conclusions. She's also pretty vicious if she gets annoyed, and I'm not sure if she learned that or if Mousebreaker made her that way.

underrated post, this really is the most impressive chatbot I've seen

Yeah, I messed around with it for a few minutes and it was pretty great. You can get it stuck in a loop pretty easily, though.