| Home | Table of Contents |
General Intelligence :
Version 2.0
( For detailed requirements for each module, please refer to the module's page. )
The following requirements are summarized from AI:MA (Artificial Intelligence: A Modern Approach, 2nd ed). They aim to encompass all the functions of a general knowledge maintenance system. AI:MA has dealt with these functions separately, the challenge is to integrate them into a coherent whole.
The list may seem daunting, but we can focus on R1-3 first.
The entire sensory experience must be compressed and stored in the memory systems.
In cognitive science memory is sometimes organized into declarative (episodic, generic, and semantic) memory, and procedural memory, but we don't know how exactly they are implemented in the brain.
This is the job of the Recognizer. For example, the input may be the video sequence of 2 men fighting with each other. The Recognizer first performs low-level vision, so we have an internal representation describing 2 human bodies and the movements of their arms and legs etc. Then at the higher level the Recognizer recognizes this as "2 men fighting", where "fight" is the higher-level concept.
Humans interact with the GI mainly through this function.
Traditionally inference means given a current knowledgebase KB and a statement p, determining whether KB entails p.
The statement p may be input as a natural language sentence, which is then translated to the internal representation.
Queries may take these forms:
The GI should be able to discover new "useful" concepts. "Useful" means that the concept can be applied to diverse objects and that it may help achieve information compression. This function may require a lot of computing power.
New knowledge can be generalized facts or explanations. They are stored as declarative memory. An explanation is simply a set of statements that entails the thing to be explained. For example, "I cannot see the doll" may be explained by "the doll is behind the mirror".
Another example is generalizing the rule that "all things tend to gravitate towards the earth". This rule may later be superceded by knowledge of gravity in physics. This means that explanations are also subject to learning.
Coming up with a sequence of actions to achieve a goal.
Sometimes answering difficult queries requires planning.
Planning does not always involve physical actions; it can also be involved in solving cognitive problems.
You may provide some instant feedback with this form:
| Home | Table of Contents |
4/May/2006 (C) GIRG [ Notice about intellectual property on this page ]