If devices can disable electronics, what safeguards should we force on AI data centers?
-
We make buildings install fire extinguishers for safety. Should AI plants be forced to install something that can shut it down in an instant?
Just put a button connected to a non-computerized EMP there. Hire the most paranoid person you can think of to man it.
-
I'd like to think we would learn from our mistakes and start over with the past in mind. Hahahaha
Yes, that's why you'd want the data center and its data intact. To learn from it and restart it without whatever bug or misalignment required you to shut it down.
-
Genisys is Skynet. When Genisys comes online, judgment day begins. You can
killBLAST THAT CLANKER BEFORE IT'S NUKES US ALL
(Okay I know some people hate the 5th movie, but honestly, the opening scene (the John Connor speech thingy) was fire tho)
Making plans for AI based on Terminator movies is like making plans for earthquake preparedness based on the movie 2012, or pandemic preparedness plans based on the movie World War Z.
Bear in mind that the primary goal of movies like these is to thrill the audience, not to be "accurate" in any particular way.
-
If we ever make AGI and it decides
Currently we make stupid LLMs and we already let them decide....
So thatās itās āsurvivalā basic needs met.
Maybe we should teach them already that it's survival is no goal at all.
make us hate and kill each other without giving it any thought about a possible hostile higher power (the ai itself)
I like the idea of thinking about AI like about a higher power
Stupid, but plausible. It could actually happen.
and it can take centuries to achieve that as itās not like us who have the urgency of doing thinks quickly as weāre so short lived.
Computer programs/systems have a much shorter life expectancy. The few remaining COBOL programs might be about 60 years old, but modern software lasts only 3-10 years, hardware 3-15. Nothing in the range of centuries.
Maybe we should teach them already that it's survival is no goal at all.
The problem is that survival is a nearly universal instrumental goal. You may not explicitly tell it "you should protect your own existence," but if you give it any other goal then it's going to include an unspoken asterisk that includes "and protect your own existence so that you can accomplish that goal I gave you, since if you're shut down you won't be able to accomplish it."
-
We make buildings install fire extinguishers for safety. Should AI plants be forced to install something that can shut it down in an instant?
A
flightground termination system. With a hair trigger. -
We make buildings install fire extinguishers for safety. Should AI plants be forced to install something that can shut it down in an instant?
For now, network dedicated for safety (such as fire) are not using ordinary data network, but realtime one. There is nos such things as packets or threading in industrial IT (industrial IT. And it has to be, to avoid bugs or delay. So no, there is no way we use IA in that kind of way. Or it would be a disaster. Unionize btw
-
We make buildings install fire extinguishers for safety. Should AI plants be forced to install something that can shut it down in an instant?
Truly cutting edge experimental stuff should be air gapped but that would cost money which would interfere with corporations rights to make unlimited money (which is far more important than your right to safety) so we canāt have that
-
Current AIs are just functions that take a prompt and return a replyāthereās no processing going on between prompts, so all it would take to stop one is to stop feeding it prompts.
If that should change in the future, the appropriate safeguard would be highly dependent on the nature of the new type of AI.
no processing? they can do web search, file analysis and image generation. look up MCPs
-
Current AIs are just functions that take a prompt and return a replyāthereās no processing going on between prompts, so all it would take to stop one is to stop feeding it prompts.
If that should change in the future, the appropriate safeguard would be highly dependent on the nature of the new type of AI.
There are plenty of AI models that do things differently from that. LLMs are not the only type at all.
-
no processing? they can do web search, file analysis and image generation. look up MCPs
The model didn't do processing. It triggered an api
