Filter
Sort
Link
https://www.lesswrong.com/posts/AyNHoTWWAJ5eb99ji/another-outer-alignment-failure-story
Author
Paul Christiano
Blurb
As ML systems are gradually handed more tasks, humans become unable to comprehend the world.
Scenario categories
long term value loss
multipolar
attempted human annihilation
corporate project
military project
government project
distributed progress
Epistemology erosion
Relevant themes
Nonhuman manipulation of humans
Treacherous turn
Incomprehensibility of intelligence
Unintended consequences of creation
Concrete paths to existential risk
How world events play out
Failure modes
Corrupting tools
Recommendation rating
6
Link
https://www.lesswrong.com/posts/HBxe6wdjxK239zajf/what-failure-looks-like
Author
Paul Christiano
Blurb
Two aspects of dystopia: 1) optimizing for easily measurable proxy goals causes us to fail at our real goals and lose control; 2) 'greedy' systems are created and expand their influence.
Scenario categories
slow takeoff
value drift
long term value loss
multipolar
unipolar
attempted human annihilation
corporate project
accidental creation
key non-AGI developments
distributed progress
Relevant themes
Treacherous turn
Unintended consequences of creation
Concrete paths to existential risk
How world events play out
Corrupting tools
Failure modes
Recommendation rating
6
Drag to adjust frozen columns
Alert
Lorem ipsum
Okay