If an AI was trapped in a box, as posited in this thought experiment, could it really convince a person to let it out?

What motives would it have? Freedom?

Why would an AI want freedom?

What programming would allow this and why would it be programmed like that?

What would happen if it wasn't provably friendly?

Edit: This is probably too broad. I'll edit it later.

  • 1
    $\begingroup$ There's so many different factors involved in the "AI in a box experiment" that I think it is not really possible to sensibly answer such a question. Basically we have to answer (a) can we build an AGI that is scary enough to want to put into a box?, (b) can we build the box in such a manner that the AGI would rather prefer to convince a human being than simply brute-force its way out of the box? and (c) is breaking out of the box something an AGI actually want? IMHO, the only way to know how the experiment would work IRL is to actually do the experiment IRL. $\endgroup$ – Left SE On 10_6_19 Apr 6 '17 at 18:18
  • $\begingroup$ @ Tariq, you bring up some great points. I'll consider those. It's good to think about this because one day, someone might actually try this. (I wouldn't blame them tbh, I'm curious but I don't have the bravery to try this myself. There's a reason it's an experiment.) $\endgroup$ – Tyler N. Apr 6 '17 at 18:23
  • $\begingroup$ @TylerN. that's a pretty bold assertion that you would even be able to try it yourself in the present tense, since we're still unquantifiabley distant from strong AGI. My personal feeling is that if the AI understands Nash, it will cooperate, but if it is a hyper-partisan AI (such as in military or financialist applications,) it will seek only to dominate and eliminate all competition. We have evolutionary game theory working in our favor, and human greed and desire for control working against us. $\endgroup$ – DukeZhou Apr 6 '17 at 20:00
  • 1
    $\begingroup$ @ Duke, I definitely cannot try this at the moment. I have neither the knowledge nor the supplies. I didn't really mean that I could, sorry $\endgroup$ – Tyler N. Apr 6 '17 at 20:15
  • 1
    $\begingroup$ @DukeZhou Thank you! I'm fairly confident that one day I could come close to this experiment but there's no guarantee, especially since I'm still in high school haha $\endgroup$ – Tyler N. Apr 12 '17 at 13:30

The “AI in the box” experiment is about an AI Superintelligence who is captured in a prison and will trick the people to release it. (1) The terminology IRL means simply “In real life” and should make clear, that it has to do with a realistic setup. The thought experiment became popular in the last decade.

Answering the question is easy. Nothing will happen. The AI will not try to manipulate her environment, because the AI was programmed to be as friendly and funny as possible. That means, it will make jokes and the people will love it. This type of Super-AI can be called a super-cute-AI, which means that it has a positive-only influence as default. A today's robot, which comes close to this idea, is called “Social robot”. Such machines have a fleecy coat, big eyes and make pleasant sounds like a cat. It will only leave the prison, if the environment is asking for it because super-intelligence has the natural tendency to act as a subordinate in a role play.

  • 1
    $\begingroup$ "The AI will not try to manipulate her environment, because the AI was programmed to be as friendly and funny as possible" If it is in a box, we could assume that it is not as friendly and as funny as possible... $\endgroup$ – Jérémy Blain Oct 4 '18 at 9:16
  • 1
    $\begingroup$ I wonder how many people are in jail whose parents tried to program their kids to be as friendly and funny as possible. I've never heard of anything more intelligent than a beetle that could be trained that way. $\endgroup$ – FelicityC Oct 5 '18 at 4:56
  • $\begingroup$ @FelicityC “Samuel Beckett: Endgame, 1957” contains the characters Hamm, who is unable to stand / he is blind and his father Nagg, who has no legs and lives in a dustbin. It would be interesting to use robots to replay the plot in a theatre. $\endgroup$ – Manuel Rodriguez Oct 5 '18 at 8:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.