Anti-basilisk protocol: Have irrational agents (animals, infants, schizophrenics) act as AI gatekeepers. If the text terminal is all that is needed to hijack humanity, then have something confused manning the keyboard.
-
-
Replying to @Moctezuma_III
so all of this senseless madness in the world is a sophisticated defense strategy in the temporal cold war against the not-yet-realized Basilisk that haunts the future(s)
1 reply 0 retweets 5 likes -
Replying to @danlistensto
It's hard to say. In some sense, disorganization and incoherence is weakness to be exploited. It also could be that our irrationality could be a motivating factor for AI to exterminate us. To stake the future of our species on irrationality is a desperate move, for sure.
1 reply 0 retweets 2 likes -
Replying to @Moctezuma_III
why presuppose that the Basilisk is super-rational? if incoherent irrationalism is the training data set then what kind of AI would it be? the anti-Basilisk protocol is to introduce vulnerabilities, then, rather than prevent it from ever being created.
1 reply 0 retweets 2 likes -
Replying to @danlistensto
Yes, I see irrationality as a key defense. By feeding an AI irrational ideas, lies and deceptions, we might give birth it idiotic, yet still dangerous, systems. Maybe those already exist today.
1 reply 0 retweets 2 likes -
Replying to @Moctezuma_III @danlistensto
But if 'rationality is systematized winning' as Land/Yudkowsky puts it, then the adversarial approach might provide super-intelligence the evolutionary pressures it requires to develop. Still, I think it's the best defense short of (highly unlikely) iconoclasm and luddism.
1 reply 0 retweets 3 likes -
Replying to @Moctezuma_III
rationality is systematized _playing_ not systematized _winning_ I see no reason why an AI couldn't be an automated idiot instead of an automated expert. it could systematically play by the wrong rules, or pursue the wrong goals, or omit relevant data, or admit irrelevant data
1 reply 0 retweets 2 likes -
Replying to @danlistensto
I think what you're saying is highly likely, and it brings into question anthropomorphic projections we make on AI. The AI might want to 'win' in a way we consider 'losing'. But the comeback to this is that certain goals: survival, resource acquisition, are not only human.
1 reply 0 retweets 2 likes -
Replying to @Moctezuma_III
paperclip maximizer being the canonical example of the AI winning in a way that is really the entire universe losing. let's anthropomorphize to our heart's content. how do you engineer the infosphere to neutralize threatening idiot intelligences?
1 reply 0 retweets 2 likes -
Replying to @danlistensto
Paperclippers and similar idiotic AIs would have more idiosyncratic defense, compared to Basilisk defense. Something like 'convince the paper clipper that everything is already a paperclip'.
1 reply 0 retweets 2 likes
alright, I think I'm getting you. the Basilisk has much more obscure and esoteric objectives than a paperclipper. unclear what it's low-hanging confounds might be, thus the smokescreen of irrationality approach. optimal basilisk defense: psychedelic subroutines for AIs
-
-
Replying to @danlistensto
Haha, yeah, we gotta dose the Basilisk with some really good LSD.
0 replies 0 retweets 2 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.