Wednesday, December 23, 2009

Building a Bot-Net Brain: Easing the way to Singularity

Here is a list of important concepts and ideas that could help in accelerating the progress towards Singularity.

I want to work with such a list, so to distribute practical operative load to many people, working in many areas, and willing to dedicate time to the project, following a model similar to Open Source Development, that showed great effectiveness.

EEE Engage, Explain, Educate
Engage and excite
Explain clearly what we need and why
Educate to collaboration
Promote Social Peer-Peer Learning, and give practical assignments, with some kind of level gaining (like role based games or like wikipedia hiearchy)
Ask teachers to collaborate (see howard reingold great thoughts)
Develop needed new tools
Create a global social consciousness of a future we build together
Promote DIY, open source mentality

Experiments and Simulations
(many specialist things here)

Architectures fot the Bot-Net Brain
Base Systems and modules
Networking and protocols for the different levels
Basic Node to Basic Node (BN-BN)
Basic Node to Aggregation Node (BN-AN)
Aggregation Node to Interface Node (AN-IN)
Basic Node types: (Storage, Sensory, Processing, Actuatory)
Mutiple distributed reasoning engines
Distribution servers for code and tasks
Multi peer-peer node communication (twitter like)
Each user node execute a module (see seti or boinc)
Collective emerging features

Main components
Basic Node (1' level)
Aggregation Node (2' level)
Interface Node (3' level)
Goal Nodes (4' level)
Areas: a subset of nodes in a specific layer
Modules: sets of areas (spanning different layers)
Knowledge representation
Hierarchical Decision System (democratic, weight based, with weight is function of age, and # of connections)
Management of internal conflicts (tradeoff, reward, punishment)
Interaction Manager
Evaluation functions
Perception functions
Linguistic Functions
Innate Knowledge
Basic Goals (instincts)
High level Goals
Low Level Control Systems (vital functions, omeostasis)
High Level Control Systems (more abstract)
Feeling representation
Consciousness as overall emerging property
Self observation and self perception: multi-level Feedback
Self Evolution Engine, with nodes independently rebooting into autodeveloped new versions with similar interfaces

Knowledge access
Search Engine interfaces
Internet Access Interfaces (read)
Synthetic World Interfaces (read)
Social Network Interfaces (read)
Internet Active Presence (Interaction manager)
Robot Interface (proprioception)
Human Interfaces
Lab Interfaces

Creating a sense good and evil
Goal definition
Dealing with unanswerable questions

I would like to:
  1. improve, correct, rewrite and complete this list
  2. for each of its elements write a % to see how far we are from the target, and set of links to relevant documents and references
  3. define workgroups and managers taking care of each area of development
  4. leverage open source disrtributed development, but centrally leaded and very well targeted
  5. raise money from investors (government, privates, foundations)
  6. earn money selling modular operational subcomponents
  7. earn money selling botnet processing time

What do you think?
Can you take this list, improve it and re-share it?

Marco ( @mgua on twitter )


1 comment:

Ben Goertzel said...

I agree there is great promise in the massively distributed approach to AGI. For instance, this could help greatly with language learning. If you had virtual conversation agents that people could chat with on their Facebook pages, then these agents could learn from all the conversations and get smarter and smarter. However, I doubt it's feasible to use SOLELY this kind of massive distributed infrastructure to power the AGI brain. Maybe someday, but not now. So IMO we'd want an architecture with a centralized server farm PLUS a massively distributed component to its brain.... Also, there would have to be some reason people found it fun or profitable to chat with the agents...