Knowledge Representation

There is a long-held belief in AI that knowledge is crucial to intelligent behavior. If a system does not have a significant fund of knowledge (about the world or about some particular subarea), then no matter how sophisticated the reasoning, intelligent behavior will not be forthcoming. Quite apart from the difficult questions of what knowledge is needed and how much of it is needed, there is an initial question of how that knowledge must be structured, or represented, within an AI system.

1. Semantic Networks

One enduring belief about how knowledge should be structured is that there should be much and various interconnectivity between the elements of knowledge, whatever they may be. Concepts should be associated with each other with various strengths and according to various relationships. Words, for example, seem to be associated in our brains by semantic similarity and by frequency of cooccurrence.

A representational scheme known as semantic networks is one attempt to provide the necessary associativity. A semantic network is a set of nodes and links between them in which each node is labeled with a concept and each link is labeled with a relationship between the concepts. This scheme can be effective for small and static domains. However, in the wider fields of intelligent behavior it is evident that both the relationships and their strengths are context dependent, which implies that the semantic network must be dynamically reconfigurable. The concept "chair," for example, might be represented in terms of physical relationships as having legs (usually four), a seat, and a back (not always). In terms of functional relationships, it is used for sitting on. However, there are chairs with no legs (e.g., when suspended), and there are chairs that must not be sat on (e.g., in a museum). Neither of these situations would be expected to disrupt normal intelligent behavior, but such exceptional circumstances as well as a detailed computational realization of common qualifiers, such as "usually," cause major problems in AI systems.

Dynamic reconfigurability is the first example of adaptivity, of learning, which seems to be fundamental to intelligent behavior but is a source of great difficulty in computational systems. Neural networks have a fundamental ability to adapt and learn: (they are constructed by adapting the link weights to learn the training examples), but they are severely limited in scope and do not begin to approach the apparent complexities of human learning. In AI, this is considered as yet another subfield, machine learning.

2. Knowledge Bases

A further drawback of the semantic network representation of knowledge is the complexity of the representation. For example, if a node is removed from a network, then all links to it (or from it) must be removed or redirected to other nodes. Therefore, explicit representation of the desired associativity actually works against efforts to introduce dynamic adaptivity.

A knowledge base, or rule base, is a representation of the knowledge as a set of facts and rules that capture much the same information as a semantic network but do so with a set of distinct elements (the individual facts and rules). This does make the addition and deletion of knowledge elements (and hence adaptivity of the knowledge as whole) computationally much simpler. The cost is that relationships between the rules are no longer explicit within the representation. They reside implicitly in the rule ordering in conjunction with the searching strategies.

3. Expert Systems

Despite the loss of explicit representation of the interrelationships between the various elements of knowledge, knowledge bases have become firmly established as a useful way to represent knowledge in relatively limited, fixed, and well-structured domains of human endeavor. They have provided the basis for the large and relatively successful subfield of expert systems.

An expert system reproduces (or even exceeds) human capability in some specialized domain, some area of human expertise. Initially, the use of knowledge bases with appropriate control strategies so dominated the expert systems subfield that this representational scheme was used to define an expert system. Subsequently, it has been accepted that it is the reproduction of intelligent activity within a very limited domain that characterizes an expert system, whatever the means of realization. However, it is not uncommon to find the terms expert system and knowledge base used with virtual synonymity.

Unraveling Alzheimers Disease

Unraveling Alzheimers Disease

I leave absolutely nothing out! Everything that I learned about Alzheimer’s I share with you. This is the most comprehensive report on Alzheimer’s you will ever read. No stone is left unturned in this comprehensive report.

Get My Free Ebook


Post a comment