From: Ben Goertzel (ben@goertzel.org)
Date: Thu Jun 13 2002 - 16:11:12 MDT
> On Thu, 13 Jun 2002, Ben Goertzel wrote:
>
> > I think that in a few decades, it will be possible for students in an AI
> > class to build an AGI during the course of a semester, making use of
> > programming languages and toolkits that are nowhere near
> available as yet.
>
> Yes, I believe you get the point of my thought exercise which is that
> it might be the case that the kernel capabilities to achieve Seed AI, are
> modest compared to the overall capabilities delivered by AGI. So I wonder
> (as have yourself and others on this list have wondered) just what those
> essential capabilities might be?
>
> -Steve
Steve, since you asked, the essential capabilities are as follows ;)
First, a knowledge representation capable of compactly representing both
procedural and declarative, and both localized and distributed knowledge.
Secondly, the following dynamical processes, acting on this knowledge
representation:
1 Activation spreading
2 Importance updating (determining how much importance, i.e.
resource-worthiness, each entity in the system has)
3 Forgetting/Defrosting (moving knowledge from RAM to disk and back based on
relevant criteria)
4 Association formation
5 Logical relationship formation (based on probability estimation)
6 First-order inference (inference on links)
7 Higher-order inference (inference on compound relation nodes)
8 Concept creation (out of bits of existing concepts with some randomness)
9 Compound relation formation (i.e. formation of logical combinations of
basic relationships)
10 procedure execution
11 Reinforcement learning on procedures
12 Schema encapsulation/optimization (taking procedures that are used often
and making them more efficient, i.e. routinizing them)
13 Feeling evaluation (evaluating aspects of the system's overall state)
14 Homeostatic parameter adaptation
15 Time-based Data Mining (specialized concept formation based on temporal
patterns)
16 Clustering
17 Information gathering (special schema orienting the system toward
gathering relevant knowledge from outside itself)
-- Ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT