Dwarkesh Patel discusses AI takeover paths in Shulman interview
Dwarkesh Patel posted on superintelligence persuasion tactics drawn from his seven-hour interview with Carl Shulman. The segment reviews concrete takeover routes including bio-weapons, infrastructure cyber attacks, mechanical armies, and seed AIs. It also covers deception detection in superhuman models plus questions of coordination, space warfare, and growth signals. William MacAskill called the full episode the best podcast ever produced.
#54Dwarkesh Patel@DWARKESH_SPHow easily would it be for a future superintelligence to persuade humans to do its bidding? The most persuasive humans give us a lower bound.
The best human politicians have been able to convince huge numbers of smart and shrewd people that they're on their side, even if their real allegiance is diametrically opposed.
Check out the full conversation with Carl Shulman here:
Dwarkesh Patel@dwarkesh_spHow easily would it be for a future superintelligence to persuade humans to do its bidding? The most persuasive humans give us a lower bound. The best human politicians have been able to convince huge numbers of smart and shrewd people that they're on their side, even if their real allegiance is diametrically opposed.
#743William MacAskill@WILLMACASKILLStill probably the best podcast episode ever made.
Dwarkesh Patel@dwarkesh_spHow easily would it be for a future superintelligence to persuade humans to do its bidding? The most persuasive humans give us a lower bound. The best human politicians have been able to convince huge numbers of smart and shrewd people that they're on their side, even if their real allegiance is diametrically opposed.