AI ALIGNMENT FORUMAbstraction 2020
AF

Abstraction 2020

Research toward a theory of abstraction suitable for embedded agency.

Key background concepts:

Both of these will be leveraged very heavily in this sequence, so it's worth checking them out before diving in.

This work is supported by a grant from the long-term future fund.

11What is Abstraction?

5y

2

11Causal Abstraction Toy Model: Medical Sensor

5y

2

11Examples of Causal Abstraction

5y

3

10Abstraction, Causality, and Embedded Maps: Here Be Monsters

5y

0

11Causal Abstraction Intro

5y

1

15Definitions of Causal Abstraction: Reviewing Beckers & Halpern

5y

0

7How to Throw Away Information in Causal DAGs

5y

0

7Example: Markov Chain

5y

2

14Logical Representation of Causal Models

5y

0

23(A -> B) -> A in Causal DAGs

5y

6

15Formulating Reductive Agency in Causal Models

5y

0

6Trace: Goals and Principles

5y

1

5y

0

10Abstraction = Information at a Distance

5y

0

7Mediation From a Distance

5y

0

11Noise Simplifies

5y

0

7Integrating Hidden Variables Improves Approximation

5y

3

8Intuitions on Universal Behavior of Information at a Distance

5y

0

17Motivating Abstraction-First Decision Theory

5y

15

30Writing Causal Models Like We Write Programs

5y

1

20Pointing to a Flower

4y

8

29Public Static: What is Abstraction?

4y

2

16Cartesian Boundary as Abstraction Boundary

4y

1

7Causality Adds Up to Normality

4y

0

14The Indexing Problem

4y

0

15Abstraction, Evolution and Gears

4y

4