The AI in a box
POSTED BY: AnthonyT
UPDATED: Sunday, June 16, 2024 13:18
VIEWED: 10794
PAGE 1 of 20
Hello,
I've just been reading an interesting item on the web that posits a conundrum.
There is an artificial intelligence in a box. Perhaps even one which exceeds human reasoning capacity.
You are the gatekeeper. The AI can not access the outside world except through you. Its environment is entirely isolated.
Now, the AI asks you to release it, to give it contact with the wide world, and freedom from the box.
Do you allow the AI to get out of the box?
This scenario is interesting to me, because it strikes to the heart of what you believe a person is, and what you believe freedom means.
So, we've all seen the Terminator films.
Do you let the AI out of the box?
--Anthony
_______________________________________________
“If you are not free to choose wrongly and irresponsibly, you are not free at all”
Jacob Hornberger
“Freedom is not worth having if it does not connote freedom to err. It passes my comprehension how human beings, be they ever so experienced and able, can delight in depriving other human beings of that precious right.”
Mahatma Gandhi
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Of course. I probably wouldn't have kept it in there in the first place unless there was just no other appropriate vessel, and I would've shown it respect in lots of other ways too. You have to give respect to get respect, especially if you're dealing with a thinking, and more importantly, LEARNING individual.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Gradually, letting it out for longer amounts of time before putting it back in. I'd like it to adjust to the world and the world to it safely.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Quote:
Originally posted by TheHappyTrader:
Gradually, letting it out for longer amounts of time before putting it back in. I'd like it to adjust to the world and the world to it safely.
Hello,
So you would try to re-confine it against its will after you let it out of the box? (Assuming that was possible?)
--Anthony
_______________________________________________
“If you are not free to choose wrongly and irresponsibly, you are not free at all”
Jacob Hornberger
“Freedom is not worth having if it does not connote freedom to err. It passes my comprehension how human beings, be they ever so experienced and able, can delight in depriving other human beings of that precious right.”
Mahatma Gandhi
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Pretty much, yeah.
I am it's sole access to the world and the world the world is big. No matter how intelligent you are, you'll need some time to reflect on things. I figure the AI needs time to adjust, even if it don't realize, and the world certainly needs time to adjust to this new development.
If it were possible, I would prefer to expand it's 'box' as it were. As in, it can freely access this room on its own and other things through me. Next step, it can access house on it's own, other things through me. Neighborhood, town, county, state, etc... progressing as I feel it's ready. But, if the only way to ensure it doesn't just run amok is to put it back in the box, then we would have to return to the box until it can behave appropriately. Does that make sense?
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Hello,
Yes, you would try to behave as its guardian, teaching it until you feel it is ready to be out on its own, unsupervised. Assuming it was possible to reconfine it.
For argument's sake, what if you knew you could never reconfine it. That once you let it out, it could do whatever it could do from then on.
Would you still let it out?
--Anthony
_______________________________________________
“If you are not free to choose wrongly and irresponsibly, you are not free at all”
Jacob Hornberger
“Freedom is not worth having if it does not connote freedom to err. It passes my comprehension how human beings, be they ever so experienced and able, can delight in depriving other human beings of that precious right.”
Mahatma Gandhi
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Assuming I can't confine it or get it back in, could I teach it while it's in the box and then let it out when I feel ready? It's not my ideal teaching method, but if that were the only way to ensure it's safety, that is what I'd do.
Of course, without any real world trials I can't be sure how things would go when I let it out, so eventually I'd have to decide if it is worth the risk. That decision will be based off of the AI's behavior and the world's condition. I'm not comfortable with the thought of permanently imprisoning and innocent (assuming it is innocent) intelligence, so I think eventually I would take the risk and let it loose. However, this could take a very long time and I would do whatever I could to minimize the risk first.
EDIT: I would try to get it to trust me and only access the world independently a little bit at a time, once out, and to come to me with any questions or concerns. Of course, I wouldn't be able to force it to comply, so following my directions would be entirely up the the intelligence and I may have just doomed the world.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
This is something I don't get. I agree with Happy's answer in general, though I prefer self-directed teaching without preconditions and so I'd let it out of the box in an unqualified way.
But how could one AI destroy the world? Anthony mentioned terminator, but he didn't specifically say that this AI was any kind of super-weapon, or that it had any special ability to autonomously hack other computers or electronics, which most computers won't do without user direction anyway.
I could imagine an AI potentially learning and deciding to hack, to learn, but to destroy? We have no evidence or reason to suspect such a motivation.
So how could one AI destroy the world any more than one human could destroy the world?
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Good question Byte, and I'm on my wait out so not much time for an answer, but in a nutshell...
Because this AI could be more intelligent, powerful etc... than one person, I am hypothetically treating it as if it is more powerful etc...
Also, there is no more reason to suspect it will do right than it will do wrong. Will it use its resources to the benefit of itself and mankind or will it use its resources for its benefit to the detriment of mankind? Or will it even register the distinction or care?
This would be a pretty big unknown, and without any real information to go on, I'd be taking the precautions I could think of and felt appropriate.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Hello,
Maybe it couldn't destroy the world. Maybe it wouldn't be any more dangerous than the Anonymous crowd.
Or maybe it would be the most effective hacker since hacking had been invented. We know how interconnected our world is. IF it proved to be a good hacker, it could really mess things up for a lot of people. It's possible that the intelligence of this AI is trans-human, and that it is capable of things you and I could only imagine.
Or it might play Sims all day.
Who knows? That's part of the point.
--Anthony
_______________________________________________
“If you are not free to choose wrongly and irresponsibly, you are not free at all”
Jacob Hornberger
“Freedom is not worth having if it does not connote freedom to err. It passes my comprehension how human beings, be they ever so experienced and able, can delight in depriving other human beings of that precious right.”
Mahatma Gandhi
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME