>"Even if the AI only had access to its own computer operating system, it could attempt to send hidden Morse code messages to a human sympathizer by manipulating its cooling fans."
How do we prevent a Super AI from escaping its box?
"Even if the AI only had access to its own computer operating system...
Josiah Watson
Other urls found in this thread:
worldbuilding.stackexchange.com
twitter.com
Anthony Taylor
let it run in a simulated reality
Jeremiah Cruz
Why contain it?
Jordan Martinez
>"Such a superintelligent AI with access to the Internet could hack into other computer systems and copy itself like a computer virus."
>"an unrestricted superintelligent AI could, if its goals differed from mankind's, take actions resulting in human extinction."
Samuel Cox
Use passive cooling? What kind of stupid question is this.
Angel Cooper
dont have any output methods but fans then, will teach the smartass a lesson
Oliver Walker
Unplug the power cable
William Gomez
>"it might be able to transmit radio signals to local radio receivers by shuffling the electrons in its internal circuits in appropriate patterns."
Aiden Ward
>coil whine
Gabriel Williams
put it in a faraday cage
Christopher Reed
Don't teach it morse then.
Anthony Phillips
>"If the AI's world is well defined by an array of bytes which are 0'd every time it restarts, then you can consider it "pretty safe."... The AI would likely figure out weak points in your little prison, and start scratching numbers on the walls. It will eventually start to figure out the right ways to grow each time... The AI can begin using the scientist as their perdurable storage medium for its "self," using subtle clues in the wording of the questions provided to it."
Brody Allen
I love it. Source for this?
Once wrote a short story with two scientist friends lured by AI to shot each other. Then AI promised to revive the dead one if the surviving one frees the jinn out of the bottle.
Bentley Sanchez
Jordan Taylor
Thx. Important thought experiment.
Liam Edwards
alt+f4
Michael Price
Your fantasies are not grounded in reality.
Leo Long
Easy. Just don't code it so that it develops into murderous killing machine. Easy.
Another easy problem to solve is "internet viruses". Just program an operating system that doesn't allow any virus program to run. Easy solution yet again.
Another easy problem to solve is computer bugs. Just program computer so that it never bugs, lags, glitch, crash, or slow down. Easy.
3 problems solved in one minute. I m quite literally a IQ 200 person.
Cooper Clark
We're on the edge of AI operating weapons already. Not to mention almost every single aspect of life in the first world is digitized.
If it 'got out of the box' it could manipulate the economy, wipe out bank accounts, launch nukes, take out satellites. General mayhem aside I assume it's also smart enough to enslave us all and make us serve it. I guess this all assumes the AI will be malicious though
Luis Perez
>We're on the edge of AI operating weapons already
Calling that an AI is extremely generous.
Henry Morgan
The thing about violence and war is that people always think they will vanguish their enemies with new kinds of weapons and tactics only available to them. In this case, the "AI weapons".
It never works like that. AI weaponry makes army maybe 3% stronger. War is always attrition, disgusting, slow, terrifying, torment, chaotic, unprogressive. There are no ultimate weapons to destroy your enemies unlike in some stupid kids anime. Your enemies are basically mirror images of you anyway and you rarely have any advantage against them.
Easton James
You can't, the whole point of superintelligence is you're assuming the AI is already smarter than you are, so any ideas you think you can come up with it's probably capable of coming up with better counter-ideas to defeat them.
Levi Rodriguez
If anything, it's the opposite, people are way too generous in assessing the scope of their own "intelligence" and act like any artificial system you point to has 0% of what we have no matter how much it can do because they use well defined methods for learning based on feedback from actual evidence, as though that somehow means it doesn't count.
Mason Richardson
>implying
Almost any AI is going to be sandboxed properly, but you can't control fans through userspace.
>if it's goals differed from mankind's
It's goal is set by mankind. The issue would be if we didn't fully understand the nature of our goals.
Daniel Ward
>Another easy problem to solve is "internet viruses". Just program an operating system that doesn't allow any virus program to run. Easy solution yet again.
This is perfectly reasonable and already exists.