AI ALIGNMENT FORUM
AF

Modularity

Dec 02, 2021 by Raymond Arnold

The second book is about modularity. Well designed or evolved structures are often not just made of parts, but made of parts with simple interfaces. These interfaces allow the parts to be reused in alternative contexts, and thus recombined in different ways.

One of the most important benefits is inspection and debugging. Modular systems are easier to understand, because their components are self-contained. However, this can also be a curse; when components have boundaries that are simple relative to their internals, they cannot share all of their information with the other components, which can lead to communication and coordination failures.

The essays in this collection have been split into four clusters, to make them easier to and and remember. This book contains essays related to modularity, both as a thing that we engineer, and as a thing that we discover in the world.

This book will explore modularized systems through topics such as biology, machine learning, companies, human emotion, and more.

47Book Review: Design Principles of Biological Circuits
johnswentworth
6y
0
38Reframing Superintelligence: Comprehensive AI Services as General Intelligence
Rohin Shah
6y
29
31Building up to an Internal Family Systems model
Kaj Sotala
6y
0
52Being the (Pareto) Best in the World
johnswentworth
6y
0
33The Schelling Choice is "Rabbit", not "Stag"
Raymond Arnold
6y
0
22Literature Review: Distributed Teams
Elizabeth
6y
0
36Gears-Level Models are Capital Investments
johnswentworth
6y
0
57Evolution of Modularity
johnswentworth
6y
6
39You Get About Five Words
Raymond Arnold
6y
0
25Coherent decisions imply consistent utilities
Eliezer Yudkowsky
6y
0
58Alignment Research Field Guide
Abram Demski
6y
8
39Forum participation as a research strategy
Wei Dai
6y
0
39The Credit Assignment Problem
Abram Demski
6y
32
60Selection vs Control
Abram Demski
6y
14