Difference between revisions of "Pivotal act"
| (3 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
A '''pivotal act''' is (putting down my guess here) something that allows a single entity (?) to take over the world? form a [[singleton]]? prevent others from taking over the world or forming a singleton or achieving AGI? | A '''pivotal act''' is (putting down my guess here) something that allows a single entity (?) to take over the world? form a [[singleton]]? prevent others from taking over the world or forming a singleton or achieving AGI? | ||
| + | |||
| + | things to clarify: is pivotal act only for "good" actors, or does it also include destroying the world? also pivotal capability vs pivotal act | ||
according to some views of [[AI takeoff]], a pivotal act is an important part of the plan for preventing [[existential doom from AI]]. | according to some views of [[AI takeoff]], a pivotal act is an important part of the plan for preventing [[existential doom from AI]]. | ||
| Line 6: | Line 8: | ||
* "Build self-replicating open-air nanosystems and use them (only) to melt all GPUs."<ref>https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns</ref> | * "Build self-replicating open-air nanosystems and use them (only) to melt all GPUs."<ref>https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns</ref> | ||
| + | * Make progress on the full (i.e. not restricted to a limited AI system like present-day systems or [[minimal AGI]]) alignment problem faster than humans can<ref>https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns</ref> | ||
==Non-examples== | ==Non-examples== | ||
| + | |||
| + | * make dramatic progress in proving mathematical theorems (without explaining the results to humans) -- not a pivotal capability because it's unclear how to perform a pivotal act<ref>https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns</ref> | ||
==External links== | ==External links== | ||
Latest revision as of 15:15, 26 February 2022
A pivotal act is (putting down my guess here) something that allows a single entity (?) to take over the world? form a singleton? prevent others from taking over the world or forming a singleton or achieving AGI?
things to clarify: is pivotal act only for "good" actors, or does it also include destroying the world? also pivotal capability vs pivotal act
according to some views of AI takeoff, a pivotal act is an important part of the plan for preventing existential doom from AI.
Examples
- "Build self-replicating open-air nanosystems and use them (only) to melt all GPUs."[1]
- Make progress on the full (i.e. not restricted to a limited AI system like present-day systems or minimal AGI) alignment problem faster than humans can[2]
Non-examples
- make dramatic progress in proving mathematical theorems (without explaining the results to humans) -- not a pivotal capability because it's unclear how to perform a pivotal act[3]
External links
References
- ↑ https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns
- ↑ https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns
- ↑ https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignment-difficulty#1_1__Deep_vs__shallow_problem_solving_patterns
