Overview Index

Learning

Definition

Definition of word “Learning” by dictionary.com:

  1. The act, process, or experience of gaining knowledge or skill.
  2. Knowledge or skill gained through schooling or study.
  3. Psychology. Behavioral modification especially through experience or conditioning.

 

Two ways of learning

There are two ways of the learning process:

  1. Experiment.
  2. Knowledge download from external sources (books, people, other computers, CD-disks, …).

 

Experiment advantages

Experiment is the key way of learning process because:

  1. Experimenting makes good results without “knowledge downloads”.
  2. “Knowledge download” is useless without experiment, because without experimenting “knowledge downloading” has no connection with goals and hardcoded units.

Knowledge downloads advantages

In other hand, knowledge download is much faster way of learning, because it doesn’t need time for experimenting. Instead of making hundreds of experiments and analyzing results, knowledge can be quickly downloaded from the external source.

 

The most efficient learning

The most efficient way of learning is a combination of both ways:

  1. Download knowledge from an external knowledge source (e.g. Internet).
  2. Make some experiments which accept or reject gathered knowledge.

 

Learning process in Strong AI

“Experimental” way of learning is embedded into experiment process (see: action, event, event correlation analyzer).

“Knowledge download” is possible by reading, in particular by basic plain text parsing routine.

 

Result of learning process in Strong AI

1) New concepts are created in the main memory.

2) New relations are created in cause-effect relation table.

3) Existing concepts are updated. In particular “Strength” and “Desirability” field are modified. From functional point of view that means that softcoded goals are modified. Softcoded goals update improve PCnous understanding how to achieve hardcoded goals.

4) Existing cause-effect relations are updated. “Strength” and “Coherence” fields of cause-effect relation table are updated.

5) Word dictionary and phrase dictionary are enriched.

 

Implementation of learning process in hardcoded routines

Discover concept correlation routine – establishes correlation between close concepts.

Deliberation routine – identifies direct relations between indirectly related concepts.

Reward distribution routine – adjusts softcoded goals based on experience.

 

Memory Cleaner routine – cleans the main memory by removing useless concepts and relations.

 

Basic plain text parsing routine – knowledge download in form of natural text reading.