Rogue AI: Bioweapons, Cyberattacks, Military Force, Bargaining | Carl Shulman
Ғылым және технология
Full Episode: • Carl Shulman (Pt 2) - ... (June 2023)
Transcript: www.dwarkeshpatel.com/carl-sh...
Apple Podcasts: bit.ly/3r1HBJk
Spotify: bit.ly/437t50c
Follow me on Twitter: / dwarkesh_sp
Carl's blog: reflectivedisequilibrium.blogs...
Пікірлер: 22
would love to see Carl back again after enough time has passed that its worthwhile
@saiparepally
Ай бұрын
Yes! One of the best guests ever
I wonder if someone said we are making a new genetically engineered species that thinks and communicates 1000x faster than us but is made to be "subservient", would people feel so relaxed?
@DwarkeshPatel Thanks for doing (and re-posting) this interview with Carl. I can just listen to the whole podcast and slowly shake my head for 3 hours. His points are so clean, so clear. How could anyone listen to this (or Roman Yampolskiy, Max Tegmark, Yoshua Bengio, Geofry Hinton, Jaan Tallinn, etc, etc and think... "Nah, impossible. We got this!"?
might wanna disable transcript on this one bro
@DwarkeshPatel
2 ай бұрын
😅
@juliandunn8412
Ай бұрын
Lol
The AI will also have a lot of carrot to offer, not just the stick, no? Assuming that it's not capable of superhuman persuasion, which seems like it mightbe, if it's able to plan to this extent without losing coherence.
@mrbeastly3444
Ай бұрын
For sure! Anything that has super-human general intelligence should be able to (legally) make a lot of money? E.g. by trading stocks, building new digital products and services, replacing millions of Human workers with millions of copies of it's more intelligent self, etc, etc. Then, yes it could use this vast amount of money to provide "carrots" to millions of Humans to do it's bidding in the physical world. E.g. Fight wars, destroy competitor AIs, build robot armies, etc, etc. Anything with super-human intelligence would quickly be able to develop a lot of carrots and sticks. And, be very unpredictable to lowly Humans (e.g. as humans are very unpredictable to chimps, or insects...)
Wow, that guy from Scrubs hasn't aged a day
9:28 "acquire a lot of illicit wealth..." True, but it could probably do (nearly) all of this without breaking any human laws or rules as well? You would think that any Super-Human intelligence could acquire a lot of money legally. E.g. trading stocks, making new digital products and services, replacing millions of human workers with millions of copies of it self, etc, etc. Then, it could use that vast wealth to pay humans to take any actions that it wants in the physical world. E.g. fight wars, destroy rival AIs, build armies of robots, etc. And, once it has access to millions or billions of robots (which Carl mentions could be built in only a few years with existing car manufacturing plants), then why risk keeping humans around at all? As, they _could_ just try to create more rival ASIs...
This was in June 2023.
why keep posting old videos? Unsubscrıbed
Doomer
@juliandunn8412
Ай бұрын
Stop it. Totally plausible with a logarithmically intelligent AI... tell me how you would control something a billion times smarter than you? Quit being naive. It's like you don't understand the potentialities at all...
@Iitex
Ай бұрын
@@juliandunn8412 I am a proud acceronalist 😀 for the future!!
@princep1398
Ай бұрын
@@juliandunn8412 could a dead man kill switch work to shut down a rogue ai?
@danaut3936
Ай бұрын
@@princep1398 The 'dead man' being humanity?
@dg-ov4cf
Ай бұрын
@@princep1398bill gates already has one rigged from his heartbeat to the internet's power switch i saw an infowars special about it