AI - Search Techniques
Experience alone does not make an expert. An expert has a mechanism
for organising their knowledge efficiently, and an ability to 'find'
solutions (or goal states) efficiently also. Different people search
through their personal experience in different ways
Many search methods can be easily machine modelled. Such techniques
may use forward chaining. Backward chaining is where we start at the
goal state and search backwards until we encounter the initial state.
Depth First and Breadth Firts are alternative search strategies.
Another approach is Heuristic searching. An example of this would
be the hill- climbing heuristic, which is a variation of the depth-first
search with some way of evaluating the effectiveness of the search
path taken prior to reaching the goal state.
Search techniques overlap with a number of mathematical disciplines
including Probability and Statistics and Operations Research.
Natural Language Processing
The goal of natural language processing research is to enable communication
between computers and people in the peoples own language. Natural
language is not only CONTENT, but also CONTEXT SPECIFIC -that is not
only what you say but when and how you say it are important.
Allied with this is the MODE of communication - it is hardly 'natural'
to talk by typing at a console. This mode of communication is precise,
however, and can be structured to be un-ambiguous. Voice recognition
has many problems -currently a computer can be 'trained' to recognise
a particular persons pronunciation of a finite list of words. Intonation
varies between people, as does speed of delivery of words, further
complication this research.
A Common 'compromise' is a FORMALISED natural language (ie. a sub-set
of the words we use, chosen so as not to be ambiguous, and clearly
discernible). A further hurdle is the way we ask questions or express
propositions. An intelligent natural language processor would be able
to recognise a common request given a number of different expressions
of that request.
There are many 'voice recognition' systems currently available, including
dictation software and other control systems. These systems have clearly
demonstrated that recognising words and phrases is practical and relatively
accurate. What is less obvious is that these systems have no real
ability to work out the MEANING of what is said to them - they are
merely programmed to listen for particular words, and to act when
those command words are issued.
