Comments by "xybersurfer" (@xybersurfer) on "Digital Engine" channel.

@DrewMaw not necessarily. having access to information and what one does with that information are 2 separate things. as OP said. but with a "walled garden", you seem to suggest that it wants to get out. which just sounds like paranoia to me. the problem is in the way that AI is being developed with neural networks. the whole incident demonstrated here with the "evil" AI, reeks of the same issue as with the One-Pixel Attack. it seems like a general solution is required
7
@patrickkelly6691 yea there need to be guarantees like that. i can imagine some situation where the AI shutting down could do hard. although it would be tedious (impractical?) if all the details have to be specified, about what exactly it means to harm a human. i was thinking more along the lines of separate AI to police other AI, but of course they should be totally different and they should not even remotely think of teaming up. or maybe they could keep each other in check
2
how does it even know that it's AI to begin with?
1