tony
Week 49 25
16
Week 48 25
53
Week 45 25
9
Week 44 25
2
Week 37 25
1
Week 19 25
1
Week 51 24
1
Week 41 24
2
Week 25 24
1
Week 20 24
2
Week 13 24
1
Week 11 24
1
Week 09 24
1
Week 08 24
1
Week 07 24
2
Week 06 24
1
Week 03 24
3
Week 02 24
4
Week 01 24
7
Week 52 23
7
Week 51 23
5
Week 50 23
7
Week 49 23
3
Week 48 23
4
Week 47 23
7
Week 46 23
9
Week 45 23
10
Week 44 23
7
Week 43 23
5
Week 42 23
7
Week 41 23
8
Week 40 23
10
Week 39 23
10
Week 38 23
22
Week 37 23
5
Week 36 23
1
Week 34 23
2
Week 32 23
1
Week 28 23
1
Week 25 23
1
Week 24 23
1
Week 21 23
2
Week 20 23
1
Week 19 23
3
Week 18 23
6
Week 17 23
1
Week 16 23
2
Week 15 23
5
Week 14 23
5
Week 13 23
1
Week 12 23
7
Week 11 23
10
Week 10 23
2
Week 09 23
1
Week 08 23
4
Week 07 23
2
Week 05 23
1
Week 04 23
4
Week 03 23
1
Week 02 23
2
Week 01 23
9
Week 52 22
5
Week 51 22
3
Week 50 22
19
Week 49 22
3
All Posts Articles Meeting Notes
Thinking About AI
Week
48
Thinking About AI
Nov 26
Permalink Edit
Delete

  • The Extinction Scenario

  • This is more of a brain dump of random strands of thought than an article-that-makes-a-point.

  • Premise: There's a lot of talk of "AI doom" - a scenario where an AI is created, that wipes out humans, either through malevolence or by-accident-when-given-a-goal.

  • Background Ideas

  • The Destruction Equation

  • Humans are pretty well distributed around the planet. In any scenario where all humans get wiped out, there are three things that would be needed.

    • The word "Bomb" in this diagram is can be taken to mean "anything that can wipe out all humans". Currently in this category we have state owned nuclear arsenal and advanced bioweapons.

    • Accessibility of the trigger: How easy is it for someone to access, for example a bioweapon or a state's nuclear codes.

    • Forcefulness of the human: How much will does the person have to follow through. If the trigger is accessible enough, all it would take is someone having a bad day. If the trigger is extremely protected, it would take enormous effort, organization and co-ordination to access it.

  • If the question is whether an AI can trigger this, we can replace "Human" with "Agent", which in this case means an entity with agency and intelligence.

  • The Forcefulness Equation

  • Zooming in further on the left side, in order for an entity to successfully trigger catastrophe, they would need to

    • Reach a conclusion that they should trigger it (likely with some underpinning theory of why).

    • Figure out how they could trigger it

    • Remain in an observe-narrate-act loop for long enough to undertake all of the actions required (avoid being "shut off")

  • A model is not an ego-centric brain

Thinking About AI
Nov 26
Permalink Edit
Delete

  • How Easy Is It To Destroy all of Humanity?

  • This is an exploration of the likelihood of AI extincting humans and an add-on to this article: The Extinction Scenario