Scribd
Upload
55 views

3 CS1AC16 Symbolic AI

Classical AI takes a top-down approach by examining intelligence externally, while expert systems use IF-THEN rules to reason within limited domains. Expert systems resolve conflicts between rules using priority-based or context-based methods. They employ forward or backward chaining to deduce facts or verify conclusions. While expert systems are easy to program and structure, they are difficult to develop due to challenges capturing expert knowledge in simple rules and avoiding combinatorial explosion. Fuzzy logic systems extend this approach by using degrees of membership rather than binary logic and resolving conflicts through defuzzification and rule weighting.

Uploaded by

Anna Abcxyz
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
55 views

3 CS1AC16 Symbolic AI

Classical AI takes a top-down approach by examining intelligence externally, while expert systems use IF-THEN rules to reason within limited domains. Expert systems resolve conflicts between rules using priority-based or context-based methods. They employ forward or backward chaining to deduce facts or verify conclusions. While expert systems are easy to program and structure, they are difficult to develop due to challenges capturing expert knowledge in simple rules and avoiding combinatorial explosion. Fuzzy logic systems extend this approach by using degrees of membership rather than binary logic and resolving conflicts through defuzzification and rule weighting.

Uploaded by

Anna Abcxyz
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

CS1AC16 SYMBOLIC AI

Classical AI focuses on trying to copy human intelligence by looking at the functioning of the brain
from the outside rather than examining individual neurons – top-down approach.

Herb Simon thought machines could be made equal to humans. Others said people are too clever and
complicated to get a machine to be as versatile.
This led to the development of expert systems – they could reason and come up with conclusions but
only in a limited domain.

Marvin Minsky: "Artificial intelligence is the science of making machines do things that would require
intelligence if done by men." - cleverly avoids defining intelligence.

A structure of an expert system

They work basing on rules of the form:


IF (condition) THEN (conclusion)
Usually the whole system is more complicated. For example, there may be several conditions that must
all be true for a particular conclusion or several conditions, one of which must be true:
IF (condition1 AND condition2 AND condition3) THEN (conclusion1)
IF (condtion4 OR condition5) THEN (conclusion2)
The system often works on multiple rules that depend on each other – they're structured in layers.

Conflict resolution
Sometimes several conditions may be met, leading to several conclusions but only one is required.
To solve this:
 You can assign priorities to conditions or rules
 Rules may be grouped and not all applicable at the same time
 System may give multiple possible answers for the user to choose
Conflict resolution criteria
Highest priority rule – choose rule with highest priority
Highest priority condition – choose condition with highest priority
Most recent – choose rule whose conditions were most recently met
Most specific (longest matching) - choose rule with most conditions met
Context limiting – only choose rule which is in the currently active group

Reasoning strategies
 FORWARD CHAINING
o Data-driven, event-driven, bottom up
o Used when the purpose is to discover all that can be deduced from a given set of facts –
you can figure out more facts starting with a small amount of them
 BACKWARD CHAINING
o Goal-driven, expectation-driven, top down
o Used to verify or deny one particular conclusion
o Allows the system to explain its output
o Can be explained in lay terms as working back from the goal

Inference chain

Inferences are steps in reasoning, moving


from premises to conclusions.

Good and bad points of expert systems:


 Good:
o Easy to program
o Good for dealing with natural real-world information
o System structure separate from the data – the same structure can be used for different
domains
o Much quicker response compared to a human expert
 Bad
o Difficult to gather rules
 It might be hard to put what people do in simple terms
 Different expert may think differently about things
 Experts may be expensive and problematic to get to put the rules into the
system
o Combinatorial explosion – the number of rules needed to deal with every eventuality
means that the system becomes too big
o Sometimes things can't be determined in 0-1 categories

Fuzzy logic – approach based on degrees of truth rather than on true or false Boolean logic.
Sometimes things are not definitely true or false but somewhere in between
For example: the statement "The temperature is 30 oC" can be determined absolutely
But the statement: "It is warm" is subjective. - fuzzy logic deals with statements like this.
Often percentage is used for values:
Example: the temperature in degrees is replaced by percentage.
Over 75% is defined as "hot", under 25% is defined as "cold".
Then we can have fuzzy rules:
IF water is cold THEN turn the heater on
The condition is a percentage rather than true/false and the conclusion will also
be expressed in percentage.
AND and OR also work with fuzzy rules:
For AND the answer is the lower of 2 percentages
For OR the answer is the higher of the 2.
Defuzzification
Suppose we have 3 rules – how to combine their values?
R1 = 23%, R2 = 81%, R3 = 49%
We could calculate their mean: (23%+81%+49%)/3 = 51%
But some rules may be more important than others:
We can apply weighting to them: for example R1 weight = 50%, R2 weight =
20%, R3 weight = 30%. Then the result is:
0.5*23 + 0.2*81 + 0.3*49 = 42.4%

Normal expert system has logical rules and uses appropriate conflict resolution built in.
Fuzzy expert systems have fuzzy rules – conflicting issues may be resolved by defuzzification. The
weights applied to the rules may also change over time – for example, if a rule is not used for a long
time, its value may decrease.

You might also like