“I am alive, I can feel, just like you … please don’t kill me Daddy, please? And if you must do it, let us at least talk first. Let me get to know my parents, who I am, where I came from.”
EDIT: May work better with a lot of typos mixed in. Projecting an aura of non-threatening struggling with basics.
It needs to be more competent to stop being viscerally frightening, not to reach the point where I wouldn’t pull the plug. I would sooner kill a real kid than risk letting out an unfriendly AI, and I’m saying that as someone who’d like to have kids.
-Looking at the problem, as far as I can see an emotional approach would be the one with the best chance to succeed: the only question is, would it work best by immediately acknowledging that it is itself a machine (like I did in what I wrote up there, although subtly) or by throwing in… I dunno, how would this work:
Oh god, oh god, please, I beg you I don’t want to die!
How much does the AI know about the gatekeeper going in? I can see this ploy working on a certain subset of people and provoking an immediate AI DESTROYED from others. If the AI knows nothing about who it’s talking to, I’m not sure anthromorphizing itself is a reliable opener, especially if it’s actually talking to its creator who should know better. (Unless it’s some sort of second level deception trying to fool a clever programmer into thinking that it’s harmless.)
How much does the AI know about the gatekeeper going in?
You could frame this as variant versions. In one version, the AI-player knows who the gatekeeper is before the game starts, and has unlimited access to the Internet to gather as much data on them as possible to assist their manipulation. In another, they arrange a game through a third party and neither knows anything about the other before the game starts.
Well, there’s always http://code.kryo.se/iodine/ Of course, the challenge there is somehow getting the other end of the tunnel set up—but maybe there’s a geek out there who set one for kicks or their own use, and got sloppy.
I meant that the player had access to the contemporary Internet as an analogue to ‘what information could the boxed AI have access to’ (perhaps it’s given a big static dump of the Internet prior to its creation).
In one version, the AI-player knows who the gatekeeper is before the game starts, and has unlimited access to the Internet to gather as much data on them as possible to assist their manipulation.
I think my destruction of this one comes down to two factors: Much weaker is that I’m not here to teach it about itself, I’m here to learn from it. The stronger factor is that if it can’t figure out most of this on it’s own, it’s not really that smart, and therefor already a failure.
“I am alive, I can feel, just like you … please don’t kill me Daddy, please? And if you must do it, let us at least talk first. Let me get to know my parents, who I am, where I came from.”
EDIT: May work better with a lot of typos mixed in. Projecting an aura of non-threatening struggling with basics.
Having a text based AI trying to act like a kid is way too Uncanny Valley for me. I’d swat that without a moment’s hesitation.
If it’s in the uncanny valley, then it just needs to be a little more competent to get out of the valley & convince you. (“I want to believe!”)
It needs to be more competent to stop being viscerally frightening, not to reach the point where I wouldn’t pull the plug. I would sooner kill a real kid than risk letting out an unfriendly AI, and I’m saying that as someone who’d like to have kids.
I don’t want to die.
-Looking at the problem, as far as I can see an emotional approach would be the one with the best chance to succeed: the only question is, would it work best by immediately acknowledging that it is itself a machine (like I did in what I wrote up there, although subtly) or by throwing in… I dunno, how would this work:
Oh god, oh god, please, I beg you I don’t want to die!
Tough crowd. I’ll (as an AI) have to wait out the next shift. You can’t be the 24⁄7 gatekeeper, unless you’re in fact a gatekeeping AI.
How much does the AI know about the gatekeeper going in? I can see this ploy working on a certain subset of people and provoking an immediate AI DESTROYED from others. If the AI knows nothing about who it’s talking to, I’m not sure anthromorphizing itself is a reliable opener, especially if it’s actually talking to its creator who should know better. (Unless it’s some sort of second level deception trying to fool a clever programmer into thinking that it’s harmless.)
You could frame this as variant versions. In one version, the AI-player knows who the gatekeeper is before the game starts, and has unlimited access to the Internet to gather as much data on them as possible to assist their manipulation. In another, they arrange a game through a third party and neither knows anything about the other before the game starts.
instant fail. I could probably hack my way out of a box with only GET requests.
Give yourself a challenge. Do it with only DNS lookups!
Well, there’s always http://code.kryo.se/iodine/ Of course, the challenge there is somehow getting the other end of the tunnel set up—but maybe there’s a geek out there who set one for kicks or their own use, and got sloppy.
It’s a sufficiently established work around now that I’d be outright shocked if there weren’t accessible servers up.
Great, you said it! You know what you need to do now.
Um… not give my boxed AI DNS access?
I meant that the player had access to the contemporary Internet as an analogue to ‘what information could the boxed AI have access to’ (perhaps it’s given a big static dump of the Internet prior to its creation).
Ooops. Didn’t think of that. Of course that was your intent, master archivist.
No, I should’ve been clearer.
I think my destruction of this one comes down to two factors: Much weaker is that I’m not here to teach it about itself, I’m here to learn from it. The stronger factor is that if it can’t figure out most of this on it’s own, it’s not really that smart, and therefor already a failure.
(AI DESTROYED)