SSG 311:

Mathematical
Modelling for
Artificial
Intelligence
Systems

MODULE 3
Reasoning Under Uncertainty
 Intended Learning Outcomes
At the end of this class, you should be able to do the following:

Understand what reasoning is

Understand different types of reasoning
Understand what uncertainty is
Understand and implement the concept of certainty factor
Revision on Probability concept

2
Contents
What is Reasoning?
Types of Reasoning
What is Uncertainty
Certainty Factor
Definition of Reasoning

Reasoning is the process of using existing knowledge to draw

conclusions, make predictions, or construct explanations.
Reasoning is the process by which we use the available existing
knowledge to draw conclusions or infer something new about
a domain of interest.
It also refers to how inferences are made about known and
assumed facts.
Types of Reasoning

1. Reasoning by Induction: one draws a general conclusion

based on some specific facts. It deals with generalise
attributes from cases seen or experience to cases unseen. It
is useful in Machine Learning, automated systems e.t.c.
In Inductive reasoning, conclusion is
merely likely
Inductive reasoning begins with observations that are specific
and limited in scope, and proceeds to a generalized conclusion
that is likely, but not certain, in light of accumulated evidence.
You could say that inductive reasoning moves from the specific
to the general. Much scientific research is carried out by the
inductive method: gathering evidence, seeking patterns, and
forming a hypothesis or theory to explain what is seen.
 Examples
Either Tope or Wande was going to win the award. Tope didn’t
win, so Wande must have won.
I think Kalu committed the murder. He had the motive, he had the
opportunity, his bloody glove was found on the scene, and he has
no alibi.
I’m sure Alice knows how to swim. After all, most people know
how to swim.
Every time you eat peanuts, your throat swells up and you can't
breath. So, you are allergic to peanuts.
All cats that you have observed purr. Therefore, every cat must
purr.
2. Reasoning by Deduction: this is the most logical reasoning . It
follows the rules of standard logic and is based on deriving
logical expression from a given premise. It is the process of
reasoning from general information about a class or object
or event to specific information about a given member of a
class.

In Deductive reasoning, conclusion is

guaranteed
Deductive reasoning starts with the assertion of a general rule and
proceeds from there to a guaranteed specific conclusion. Deductive
reasoning moves from the general rule to the specific application:
In deductive reasoning, if the original assertions are true, then the
conclusion must also be true.

For example, math is deductive:

If x = 4
And if y = 1
Then 2x + y = 9
 More Examples
i) Elephants are bigger than dogs.
Dogs are bigger than mice.
Therefore, Elephants are bigger than mice.
ii) Some dogs are great hounds.
Some grey hounds run fast
Therefore, some dogs run fast.
iii) In mathematics, If A = B and B = C, then A = C.
iv) Since all humans are mortal, and I am a human, then I am mortal.
v) All dolphins are mammals, all mammals have kidneys; therefore all
dolphins have kidneys.
3. Reasoning by Abduction: it is the process of reasoning from the
state of event to what caused it. It is a useful way of reasoning
and often involve using best guesses.

In Abductive reasoning, best shot is taken.

 Examples
A medical diagnosis is an application of abductive reasoning: given
this set of symptoms, what is the diagnosis that would best explain
most of them?

Likewise, when jurors hear evidence in a criminal case, they must

consider whether the prosecution or the defense has the best
explanation to cover all the points of evidence. While there may be
no certainty about their verdict, since there may exist additional
evidence that was not admitted in the case, they make their best
guess based on what they know.
4. Monotonic Reasoning: it involves unidirectional parameter
instantiation or estimation. i.e. when you start an approach or concept
you always follow that concept without changing it regardless of
whatever new information may become available.

For example, the statement, once a thief, always a thief.

In nonmonotonic reasoning, parameters can be reinstantiated if new
information warrants the assignment of a new value to the parameter.
5. Forward Chaining/Forward Reasoning or Data-Driven Search,
bottom-up or antecedent driven. It begins with known facts and
attempt to move towards the goal.
It is usually in the format:
If (data condition)
Then (conclusion)
If we can assume that people normally like those that they trust,
then a forward chaining rule to convey that assumption is:
If person x trust person y
Then person x likes person y
The conclusion part of a rule may become the condition part of
another rule. Thus, we can extend the above example to the one
below:
If person x trust person y
Then person x enjoys-the-company-of person y.
6. Backward Chaining/Backward Reasoning or Goal Driven, Top-
Down, Hypothesis process driven or Consequent-Driven. The
reasoning process starts from a goal state and backtracks to the
paths that might have led to the goal.

It is generally of the form:

Goal state
If (data condition)
A backward chaining rule based on the previous example is:
person x likes person y
if person x trusts person y

This example asserts that liking someone requires a condition of

trust. If for example, only 80% of the population fell in the category
of the rule assertion, then we could assign some level of certainty
or confidence to the rule. Then the rule might be stated as:
 person x likes person y
If person x trusts person y (certainty factor = 0.8)
Backward chaining is often implemented in expert systems in the
coding format of a forward chaining rule. In that case, the goal is
specified in the antecedent of the rule and the conclusion leading
to the goal is specified in the conclusion part of the rule. For the
example of inter person relationship, the backward chaining rule
can be written as:
If person x is-to-like person y
Then person x must-trust person y.
 Uncertainty
Uncertainty is a situation in which something is not known or
certain.
In terms of Decision making, it is a Situation where the current state
of knowledge is such that (1) the order or nature of things is
unknown, (2) the consequences, extent, or magnitude of
circumstances, conditions, or events is unpredictable, and (3)
credible probabilities to possible outcomes cannot be assigned.
Although too much uncertainty is undesirable, manageable
uncertainty provides the freedom to make creative decisions.
 Uncertainty
Uncertainty can be looked into with two different approaches:
1. Concept of Certainty Factor (CF)
2. Concept of Fuzzy Logic.

The most common representation of heuristic weights is the use

of CF.
 Certainty Factor or Confidence Factors
In this approach, numbers greater than zero are used for positive
factors while numbers less than zero are used for negative factors.
E.g. -1 to +1, -100 to +100.
If for instance I say:
Tomorrow (Sunny) [0.6], Tomorrow (warm) [0.8]
If I now combine the two statement
Tomorrow(Sunny) ᴧ Tomorrow (warm) [?]
How can I conclude the level of certainty in the statement above?
In this case Certainty Factor is proposed to solve this kind of
situation.
The CF indicates the degree of certainty with which each fact
or rule is believed to hold and is number between -1 and +1
 Mycin Example
The popular MYCIN expert system uses CF in handling uncertainty.
It was developed to diagnose and recommend therapies for
bacterial infections in blood. It associates a CF to each of the
production rules and a weight or confidence factor is attached to
each of them.
In MYCIN consultation, a frequency fatal cause of a disease would
be assigned a higher certainty than the one that is more likely but
rarely fatal. To evaluate MYCIN’s production rule, the following
steps are followed:
1. The CF of a conjunction of several facts is taken to be the
minimum of the CFs of the individual facts. The weakest link
of a chain determines the strength of the chain.
2. The CF of a disjunction of several facts is taken to be the
maximum of the CFs of the individual facts.
3. The CF for the conclusion produced by a rule is the CF of its
premise multiplied by the CF of the rule. E.g. If temperature
is high and patient is feverish, then malaria.
4. The CF for a fact produced as a conclusion of one or
more rules is the max of the CF produced by the rule
yielding that conclusion.
 EXample
Suppose we want to establish fact D.
If facts A,B,C,H,I,J are known with respective CFs of
0.7,0.3,0.5,0.8,0.7,0.9 and we have the following rules:
Rule 1: If A and B and C then conclude D with CF of 0.8
Rule 2: If H or I or J then conclude D with CF of 0.7
What is the CF of the combination?
 Solution
For Rule 1, CF is Min of (A,B,C)
i.e. Min (0.7,0.3,0.5) = 0.3
CF of D based on Rule 1 is 0.3 * 0.8 = 0.24
For Rule 2, CF is Max of (H,I,J)
i.e. Max(0.8,0.7,0.9) = 0.9
CF of D based on Rule 2 is 0.9 * 0.7 = 0.63
Rule Combination
CF(D) = max(CF(D)1,CF(D)2)
= max(0.24,0.63) =0.63