[ Techniques
to Represent Knowledge | Semantic Similarities | Frames of Knowledge ]
he field of AI has
expanded the role of the computer from data-processors to knowledge-processors. It
is believed that a key step for intelligence is to make the computer not only process the
facts(data) about the world around it, but also how each piece of fact relates to one
another to form knowledge and understanding.
So far, the modern-day computational computer has made a great impact on society by
becoming the modern-day record-keeper. Computers precisely keep track of medical
records, financial statements, library books; not to mention holding an the entire Encyclopedia
Britannica on a CD-ROM.
Data-processing computers can merely store billions and billions of
words and numbers and understanding how they all relate to each other. So, the words
describing a patient's history of asthma attacks in the computer's memory would be
isolated from the section describing what medication to use with asthma. In this
case, the user would have to manually find out what to treat for asthma attacks.
With knowledge-processing computers, the user may look up a patients'
record on their asthma problem and then have the computer suggest the medication to combat
the problem. In this case, it is the computers' job to relate the asthma with the
treatment--an intelligent action, obviously.
However, beyond making information more easily accessible, knowledge representation is
a definite step towards creating AI. In natural language processing, a
computer knowing how words relate to one another allow it to better understand a word if
it were taking in context to the other meaning of the other words in the sentence.
Combined with pattern-recognition
techniques, a computer can group a collection of blocks that is arranged in such a way
that the structure can be called an arch(see example below).
[ Top ]
Hierarchy
. One way to organize how facts relate to one another
is through a hierarchy from general to specific things. This common technique is
used in taxonomy in which life is broken down into kingdoms, phylum, classes, orders,
families, genuses, and species. An abstract computer implementation of the hierarchy
technique of representing knowledge would be:
From the relationship in the figure above, love, jealousy, and hate are all emotions;
each concept is enclosed in a white box and is called an object while the black boxes
labels how the objects are related(a.k.a. "linked" in computer terms) as
depicted by the white lines. Therefore, if a computer were to be asked, "Love
is an emotion. Give me another example of an emotion," it could reply
"jealousy" or "hate." A more robust organization called semantic
networks is created by having the objects cross-linked so that they can provide more
knowledge about the concepts. Therefore,a semantic network about some emotions would
look like this:
Like the previous network, love, jealousy, and hate are examples of emotions. A
uni-directional link from love to jealous possesses the knowledge that "love can lead
to jealousy" and furthermore, "jealousy can lead to hate." Finally,
this semantic network knows that "love is the opposite of hate" and vice versa,
as denoted by the bi-directional arrows.
Pattern-recognition techniques can incorporate semantic networks to help the computer
identify how objects to be analyzed are related to one another. The definition of
this arch:
in terms of relationships is this:
(Kurzweil 287)
Though semantic networks have been somewhat successful in representing knowledge, its
complexity was built by programmers object by object, link by link--a very time consuming
process. Researchers are now trying to develop networks that can develop structure
on their own as they gain facts and knowledge.
[ Top ]
Even if semantic networks prove to be a key ingredient to create artificial
intelligence, it is yet to be determined that the human brain organizes knowledge in such
a way. Only by observing human behavior can the structure of knowledge in the brain
be discovered.
What has been known about the brain is that it has many built-in redundancies.
Some speculate that the brain loses over 50,000 neurons
each day, yet most people do not feel as though a large part of their memory
disappeared. If the estimates are correct, it suggests that the "objects"
in the brain, speaking in semantic network terms, have many links; some repeated links may
exist as well. This redundancy theory also explains brain's ability or inability to
process new information. Every newly-perceived piece of information is stored in the
brain, but only those that are processed and linked to by other "objects" the
most are the ones that actually stay in memory. Information with very few links to
it is eventually forgotten. Information that apparently contradicts the brain's
paradigm of knowledge is stored into memory as well, but until there is enough repeated
exposure to that same information to create more links to it, much of the old-ideas that
the new piece of information contradicts remains unchanged. That is why any new
knowledge gained as an adult that contradicts beliefs learned growing up do not often make
much impact to the individual.
Another human behavior that is displayed by semantic networks is the ability to let the
mind wonder from one idea to the next. This phenomenon likens the traveling from one
object to the next through links that relate to them. With so much in-linking and
cross-referencing, it explains why it is sometimes difficult for people to master language
to effectively communicate individual ideas.(Kurzweil 288-289)
Semantic networks are not the only ways to represent knowledge. Similar to how
people tend to store related information together, frames are packaged data that allows a
computer to form relationships between things.
Most people acquire and store knowledge in broad conceptual terms like a
"boy." Then, as more information is gained about the "boy,"
facts like his name, his age, his shoe size, and so forth can be input into the slots
under the general term "boy." This organization can be best illustrated
below:
