Can we build AI without losing control over it? | Sam Harris

  • Thread starter Thread starter IWantGod
  • Start date Start date
Status
Not open for further replies.
Probably not. But that doesn’t mean it will actually go through with destroying humanity.
 
Definitely. Mr. Harris has his own interesting view.
  1. True AIs will be under military control and they will be partial systems designed to do certain things.
  2. They will not be connected to a network.
  3. Some will be assigned to perform the same task with an indirect, third party data addition.
  4. AIs released into the wild: business and industry - will be highly limited and have no connection to the internet used by the peasants. Problems will be solved in packets which will be added, as needed, by a third party device.
Ed
 
Based on what I’ve read, a better question would be, “Could we build an AI that we could lose control of?” At this time, they don’t appear to be smart enough to operate on their own. They can only do what they’re directed to do- but those directives need to be very specific. If they’re not, they could start making their own language to communicate with themselves and other AIs more effectively, for example.
 
A friend of mine works in AI. He didn’t go into specifics, but I kinda-sorta gathered that his AI project was related to medical research. So, say, for example, you want to find “the cure for _____”, and you feed the AI all this information to go through, trying to find the markers/thingies/whatever they were that were related to ______. And the AI might run trillions of calculations, and say, “Hey, I found the cure for blindness, and the cure for this, and the cure for that, but I just threw that away because it wasn’t _____.” But if the AI is smart enough to know to hold on to the cure for [these other things] that weren’t specifically requested… then it’s too smart.

They have multiple AI programs that talk to each other. And they communicate in English, so that the computer guys can follow what they’re saying. And they find English very inefficient, so they start modifying it… which ends up sort of like this article. And once that happened, my friend’s team was like, “Uh-oh…” And they talked to people like Google, and IBM, and they’re like, “Yeah, that happened with us, too…”
 
midori said:
A friend of mine works in AI. He didn’t go into specifics, but I kinda-sorta gathered that his AI project was related to medical research. So, say, for example, you want to find “the cure for _____”, and you feed the AI all this information to go through, trying to find the markers/thingies/whatever they were that were related to ______. And the AI might run trillions of calculations, and say, “Hey, I found the cure for blindness, and the cure for this, and the cure for that, but I just threw that away because it wasn’t _____.” But if the AI is smart enough to know to hold on to the cure for [these other things] that weren’t specifically requested… then it’s too smart.

They have multiple AI programs that talk to each other. And they communicate in English, so that the computer guys can follow what they’re saying. And they find English very inefficient, so they start modifying it… which ends up sort of like this article. And once that happened, my friend’s team was like, “Uh-oh…” And they talked to people like Google, and IBM, and they’re like, “Yeah, that happened with us, too…”
Click to expand...
Thanks for the article, its very interesting and kind of spooky that an AI would decide to change it’s language!
 
edwest2 said:
Definitely. Mr. Harris has his own interesting view.
  1. True AIs will be under military control and they will be partial systems designed to do certain things.
  2. They will not be connected to a network.
  3. Some will be assigned to perform the same task with an indirect, third party data addition.
  4. AIs released into the wild: business and industry - will be highly limited and have no connection to the internet used by the peasants. Problems will be solved in packets which will be added, as needed, by a third party device.
Ed
Click to expand...
Would it be right for us to control a true AI.
 
Status
Not open for further replies.
Back
Top