From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Mon Sep 07 1998 - 21:17:50 MDT
http://pobox.com/~sentience/AI_design.temp.html
Coding a Transhuman AI
also known as
Notes on the Design of Self-Enhancing Intelligences
The Art of Seed AIs
How to Build a Power in 3 Easy Steps and 35 Hard Steps
Applied Theology 101: Fast Burn Transcendence
A fairly massive tome, about halfway complete at 210K.
This is a temporary version for use in the Singularity Colloquium, in case the
completed page takes another two weeks. The completed version will be reannounced.
Table Of Contents: (Implemented sections only.)
Paradigms
1. Pragmatism and allowable methods.
2. Seed AIs.
3. Grounding (vs. symbols and stochastics).
Principles
1. Self-enhancement and domino enhancement.
2. Domain modules: Domdules.
3. World-model: Integration of domdules.
4. RNUI: Representing, Noticing, Understanding, and Inventing.
5. Symbols and memory.
6. Adaptive code and self-organization.
7. Reflexive reasoning: Self-awareness and will.
8. Causality and goals.
View from multiple levels
1. Meet the AI.
2. Sample thought. (Internal TOC.)
3. The AI Advantage.
Details
1. Self-swallowing compilers.
13. Interim Goal Systems.
Precautions
3. The Prime Directive: No Asimov Laws.
Credit goes to Max More for stimulating this by challenging the assertion that
human AI implies transhuman AI. I tried to specify all the principles that
would affect the AI's trajectory, and at about 3 AM realized it should
probably be a separate page...
--
sentience@pobox.com Eliezer S. Yudkowsky
http://pobox.com/~sentience/AI_design.temp.html
http://pobox.com/~sentience/sing_analysis.html
Disclaimer: Unless otherwise specified, I'm not telling you
everything I think I know.
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:49:33 MST