Expert system and symbolic machine learning are a special form of
artificial intelligence that uses deductive methods to simulate the decision
making process of humans. They are designed to solve problems which
are too difficult to solve using traditional computer programs or methods. 
    An expert system is a computer program that contains a modifiable
knowledge base (rules and facts), an interface to users, and an
inference engine which makes logic-base decisions. 
An expert system 
applies rules to the facts to draw conclusions or make decisions. During
this process, machine learning program may deduce more facts, called
asserted values, because the machine learning system asserts them
rather than getting them from outside world. The trained program is able
to deduce the answer and explain its reasoning by forward chaining or
backward chaining. Forward chaining looks at the facts, applies the
rules, produces more facts, applies more rules, etc. until final decision is
made. Backward chaining starts with a conclusion and test the rules
which could have caused that conclusion to be reached, finds the facts
that caused those rules to be applied, asserts a new *conclusion*, find
the rules and facts which produced it, etc. until the symbolic machine
learning is able to explain (in terms of applied rules and facts) why the
final conclusion was reached. Clearly, rules are the basic form of
knowledge used in an expert system.

One of the main
problems in applying expert systems is so called knowledge acquisition
bottleneck. This problem is summarized by the
following question: where do the rules come from?
In classical knowledge acquistion, rules were acquired from
the experts using specialized acquistion techniques. These techniques
are often time-consuming, expensive, and unreliable.
Symbolic machine learning addresses the knowledge
acquisition bottleneck through the development of programs that learn
rules from examples.

Machine learning tackles a number of tasks, of which classification is
the main one. In a classification task, the learning system is given a
training set, consisting of examples. Each example in the training set is
labelled with the class to which it belongs. The learning program induces,
from this training set, general rules for assigning class membership to
examples that have not been seen during training. Rules obtained by
a machine learning system can ten be used in a classifier-type
expert system that tackles a classification task.

   An expert system shell or development system is a tool
designed to help build an expert system. Using a shell often requires
learning a kind of high level programming language. The developer of an
expert system can be almost anyone; a programmer, a knowledgable
engineer, someone with very little computer experience, or even the
expert whose knowledge will be made available and put into the expert
system. The end user may be also almost anyone: an operator, an
executive, a professional or developer himself. A good expert
expert system shell will accommodate
all kinds of people who develop and use the system.  

Machine learning systems are even simpler to use. They do not require
any programming, and no knowledge of programming or
programming language is assumed. The user has
to provide labelled examples, and the machine learning system
will induce the classification rules. In the recent years, user-friendly
systems for PC have become available.