REASONING UNDER UNCERTAINTY

GROUP 3
CONTENT
⮚ The meaning of uncertainty and theories devised to deal with it.
⮚ Types of errors attribute to uncertainty.
⮚ Errors associated with induction.
⮚ Features of classical probability, experimental and subjective probabilities.
⮚ Compound and conditional probabilities.
⮚ Hypothetical reasoning and backward induction.
⮚ Temporal reasoning.
⮚ Markov chains, odds of belief, sufficiency and necessity.
⮚ Role of uncertainty in inference chains, implications of combining evidence.
⮚ Role of inference nets in expert system.
⮚ How probabilities are propagated.
• The meaning of uncertainty and theories devised to deal with it.

Uncertainty refers to the lack of definite knowledge or information, leading to a state of doubt or unpredictability. There are
various theories and models that attempt to address and deal with uncertainty, such as:

o Probability theory – Probability theory is a mathematical framework that deals with the likelihood of events occurring. It
is used to quantify uncertainty and make predictions based on incomplete information. Probability theory provides a
rigorous framework for reasoning under uncertainty, and it has many practical applications in fields such as finance,
engineering, and statistics.

An example of probability theory is flipping a coin. When we flip a coin, we know that there are only two possible
outcomes: heads or tails. The probability of getting heads or tails is equal, which means each outcome has a probability
of 0.5 or 50%.
Suppose we want to know the probability of getting heads twice in a row. We can use probability theory to calculate
this. The probability of getting heads on the first flip is 0.5, and the probability of getting heads on the second flip is
also 0.5. To find the probability of both events occurring, we multiply the probabilities together: 0.5 x 0.5 = 0.25
So, the probability of getting heads twice in a row is 0.25 or 25%.
This example illustrates how probability theory can be used to calculate the likelihood of an event occurring, even
when we don't have complete information. By using probability theory, we can make more informed decisions and
make predictions based on incomplete or uncertain information.
• The meaning of uncertainty and theories devised to deal with it.

o Fuzzy Logic - Fuzzy logic is a mathematical theory that deals with uncertainty and imprecision. It allows for the
representation of partial truths, and it can handle situations where there are degrees of truth, rather than just true or
false. Fuzzy logic is used in applications such as control systems, decision-making, and artificial intelligence.

Fuzzy logic can be used to control the temperature of a room based on imprecise or uncertain information. For
example, the word "warm" can be represented as a fuzzy set, and a fuzzy logic controller can adjust the temperature
based on the degree of membership of the "warm" fuzzy set. This allows us to make decisions based on uncertain
information and handle situations where there are degrees of truth or ambiguity.

o Bayesian inference – it is a method for updating beliefs based on new data. It uses Bayes' theorem to update prior
beliefs to posterior beliefs, incorporating new information and uncertainty. It is used in fields such as machine learning
and AI to make predictions and decisions based on uncertain data.

Bayesian networks can be used to model uncertain relationships between variables. For example, in healthcare, a
Bayesian network can be used to predict the likelihood of a patient having a particular disease based on symptoms and
medical history. The network represents the conditional probabilities of each symptom given the presence or absence
of the disease, and the probabilities of the disease given the presence or absence of each symptom. This allows us to
make more informed decisions and predictions based on incomplete or uncertain information.
• Types of errors attribute to uncertainty.

Uncertainty in decision making can be attributed to several types of errors, including:

o Measurement error: Measurement error is a type of error that is attributed to uncertainty in

measurements. It occurs when the measured value of a variable deviates from its true value due to various
factors such as equipment limitations, environmental conditions, and human error.
❖ An example of measurement error is a weighing scale that is not calibrated properly. Suppose a product
is supposed to weigh 1 kilogram, but due to measurement error, the scale reads 1.05 kilograms. This
means that the measured weight deviates from the true weight of the product, and this deviation is
attributed to measurement error.
o Model error: Model error is a type of error that is attributed to uncertainty in the model used to make
predictions or estimations. It occurs when the model does not accurately represent the underlying system
or process being modeled, leading to inaccurate predictions or estimations.
❖ An example of model error is using a linear model to predict the growth of a population that exhibits
exponential growth. In this case, the linear model would not accurately represent the underlying
population growth process, leading to inaccurate predictions.
• Types of errors attribute to uncertainty.

o Sampling error: Sampling error is a type of error that is attributed to uncertainty in the process of sampling
data from a population. It occurs when the sample used to make inferences or predictions about the
population is not representative of the entire population, leading to a discrepancy between the sample
statistic and the population parameter.
❖ An example of sampling error is taking a sample of students from a school and estimating the average
height of students in the entire school based on the sample. If the sample is not representative of the
entire school, the estimated average height may deviate significantly from the true average height of all
students in the school.
o Bias error: Bias error is a type of error that is attributed to uncertainty in the data or information used. It
occurs when there is a systematic deviation between the true value and the measured value, leading to a
consistent overestimation or underestimation of the true value.
❖ An example of bias error is conducting a survey on the popularity of a political candidate and only
surveying individuals who live in a specific geographic region. This could lead to a biased sample, as the
opinions of individuals in other regions may differ, and the survey results may not accurately reflect the
true popularity of the candidate across the entire population.
• Types of errors attribute to uncertainty.

o Human error: Human error is a type of error that is attributed to the actions or decisions made by humans,
leading to unintended consequences or outcomes. It can occur due to factors such as fatigue, distraction,
miscommunication, or lack of trainings.
❖ An example of human error is a pilot inputting incorrect data into the navigation system of an airplane,
leading to the plane flying off course. This error could be due to fatigue, distraction, or lack of training,
and could have serious consequences.
o Structural error: Structural error is a type of error that is attributed to uncertainty in the structure or design
of a model or system, leading to inaccurate predictions or estimations. It occurs when the assumptions or
simplifications made in the model do not accurately represent the underlying system or process being
modeled.
❖ An example of structural error is using a linear model to predict the stock market returns of a highly
volatile market. In this case, the linear model would not accurately capture the non-linear and complex
relationships between the various factors that influence the stock market returns, leading to inaccurate
predictions.

Dealing with these types of errors is an important part of reducing uncertainty in decision making, as they can
affect the accuracy and reliability of the results and decisions made based on the data.
• Errors associate with induction.

Induction is the process of making generalizations or predictions based on limited data, and is subject to several
types of errors, including:

o Overgeneralization: Making overly broad generalizations from limited data, which may not accurately
reflect the underlying patterns or relationships.
o Confirmation bias: The tendency to selectively seek and interpret information that confirms existing beliefs,
while ignoring or discounting information that contradicts them.
o Hasty generalization: Drawing conclusions based on insufficient data, without adequately considering
alternative explanations or evidence.
o Illogical reasoning: Using flawed reasoning or incorrect logic to arrive at conclusions.
o False cause: Assuming a causal relationship between two events when there is none or failing to consider
alternative explanations.
o Unrepresentative sample: Using a sample of data that is not representative of the larger population, leading
to inaccurate or biased results.

Reducing these errors and avoiding these pitfalls is important in induction, as they can affect the accuracy and
reliability of the results and generalizations made.
• Features of classical probability, experimental and subjective probabilities.

Classical probability, experimental probability, and subjective probability are different approaches to modeling
and quantifying uncertainty. Some of the key features of each approach are:

o Classical probability: It is based on the concept of equally likely outcomes, and is used when the outcomes
of a random event are known and the number of outcomes is finite. In classical probability, the probability
of an event is calculated as the number of favorable outcomes divided by the total number of outcomes.
o Experimental probability: This is based on the results of repeated experiments, and is used when the
outcomes of a random event are not known but can be observed. In experimental probability, the
probability of an event is estimated based on the proportion of times that event occurred in a large number
of trials.
o Subjective probability: It is based on personal beliefs and opinions, and is used when there is uncertainty
and no objective data is available. In subjective probability, individuals express their beliefs about the
likelihood of an event by assigning a probability value.

Each approach has its strengths and limitations, and the choice of approach depends on the specific situation
and the available data. The classical and experimental approaches are more objective, while the subjective
approach is more subjective and personal.
• Compound and conditional probabilities.
Compound probability and conditional probability are important concepts in probability
theory that are used to model and analyze complex uncertainty.

o Compound probability: It refers to the probability of two or more events happening

together, or the probability of an event that is the combination of multiple events.
Compound probability is calculated as the product of the probabilities of each event.
o Conditional probability: It refers to the probability of an event happening given that
another event has already happened. Conditional probability is calculated as the
probability of the first event given the second event, divided by the probability of the
second event.

Conditional and compound probabilities are essential in many applications, including risk
analysis, decision making, and statistical inference, as they provide a way to quantify complex
relationships between events and to analyze the impact of uncertain events on each other.
• Hypothetical reasoning and backward induction.
Hypothetical reasoning and backward induction are two related concepts used in decision making and problem
solving.

o Hypothetical reasoning: It involves considering possible outcomes and consequences of a decision or

action, based on hypothetical scenarios. Hypothetical reasoning is a useful tool for considering the potential
impact of decisions, exploring different options, and making informed choices.
o Backward induction: It is a decision-making method used in game theory and decision analysis, where the
decision-maker starts with the end state of a problem and works backward to determine the optimal
decision in each step along the way. Backward induction is often used to analyze sequential decision
problems, such as strategic games, where the outcome depends on the decisions made by multiple players.

Both hypothetical reasoning and backward induction are useful tools for dealing with uncertainty and
complexity, and are widely used in fields such as economics, psychology, and computer science. By considering
the potential outcomes and consequences of decisions, and by working backwards from the end state of a
problem, they provide a systematic approach to decision making and problem solving.
• Temporal Reasoning.
Temporal reasoning under uncertainty involves making predictions or decisions based on
information that evolves over time. This involves considering the changing probabilities of
events and states as time progresses, and incorporating that information into the predictions
or decisions.
Temporal reasoning under uncertainty often uses models such as dynamic Bayesian
networks, hidden Markov models, or Kalman filters to represent the evolution of the system
over time. These models allow for the propagation of probabilities over time, taking into
account the changing relationships between variables and the uncertainty associated with
them.
Examples of temporal reasoning under uncertainty include predicting the stock market,
predicting the weather, or predicting the behavior of a system that evolves over time. The
goal of temporal reasoning under uncertainty is to make the best possible predictions or
decisions based on the available information, taking into account the changing state of the
system and the uncertainty associated with it.
• Markov chains, odds of belief, sufficiency and necessity.
o Markov chains: A Markov chain is a mathematical model used to describe the evolution of a system over
time, where the future state of the system depends only on its current state, not on its past history. Markov
chains are widely used in fields such as physics, economics, and computer science to model and analyze the
behavior of complex systems.
o Odds of belief: In probability theory, the odds of belief are a way to represent the degree of belief in an
event or proposition. The odds of belief are a ratio that represents the likelihood of an event happening,
with odds in favor of the event being greater than 1, and odds against the event being less than 1.
o Sufficiency and Necessity: In logical reasoning, sufficiency and necessity are two concepts used to describe
the relationship between two events. An event is said to be sufficient for another event if it guarantees that
the second event will occur. An event is said to be necessary for another event if the second event cannot
occur without the first event.

These concepts are important in decision making and problem solving, as they provide a way to reason about
the relationships between events and to determine the necessary and sufficient conditions for a particular
outcome. By understanding the relationships between events, decision-makers can make informed choices and
predictions about the future.
• Role of uncertainty in inference chains, implications of combining evidence.

Inference chains are sequences of reasoning that link evidence to conclusions. Uncertainty in inference chains
arises from limitations and imperfections in the available evidence, such as measurement errors, imprecise
data, and incomplete information. The role of uncertainty in inference chains is to account for these limitations
and to provide a means of representing and quantifying the degree of belief in the conclusions.

When multiple sources of evidence are combined in an inference chain, it can result in more accurate
predictions and stronger conclusions, but it can also increase uncertainty if the sources of evidence are
inconsistent or conflicting. It is important to carefully consider the role of uncertainty in inference chains and to
use appropriate methods for combining evidence to account for the limitations and imperfections of the
evidence and to make informed decisions and predictions.

This includes techniques such as Bayesian inference, which uses probabilities to update beliefs based on new
evidence, and decision theory, which uses mathematical models to determine optimal decisions in uncertain
situations.
• Role of inference nets in expert system.
Inference nets are used in expert systems to represent the relationships between different pieces of knowledge,
allowing the system to reason and draw conclusions based on the available information. They provide a
structured approach to decision-making based on the knowledge and rules of the domain and enable expert
systems to make accurate recommendations or decisions consistent with the available information

An example of inference nets can be seen in a medical diagnosis expert system. In this system, the inference net
would encode the relationships between various medical symptoms and possible diseases. The system would
gather information about the patient's symptoms and use the inference net to perform forward chaining or
backward chaining to arrive at a diagnosis.

For example, if a patient presents with a fever, cough, and chest pain, the system would use the inference net
to determine which diseases are associated with these symptoms. The system may perform forward chaining
by starting with the symptom of fever and branching out to the possible diseases associated with fever, such as
pneumonia, bronchitis, or influenza. Alternatively, the system may perform backward chaining by starting with
the possible diseases associated with chest pain and narrowing down the list based on the patient's symptoms.

By using the inference net to reason about the relationships between symptoms and diseases, the medical
diagnosis expert system can make accurate recommendations or diagnoses based on the available information.
• How probabilities are propagated.
Probabilities can be propagated through a system using various methods depending on the specific system and
the nature of the relationships between the variables. Here are three common methods:

• Bayesian Networks: In Bayesian Networks, the relationships between variables are represented as a directed
acyclic graph, where each node represents a variable, and the edges represent the conditional
dependencies between variables. The probabilities of each variable are estimated based on the conditional
probabilities and prior probabilities associated with the variables in the network. Probabilities can be
propagated through the network using Bayes' rule to update the probabilities of each variable given new
information.
• Fuzzy Logic: In Fuzzy Logic, probabilities are propagated through a system by using linguistic variables to
represent the uncertainty associated with each variable. The degree of membership of a variable in a
particular linguistic set is represented by a probability value. Fuzzy Logic uses rules to propagate the
probabilities of different linguistic variables through the system and arrive at a conclusion or decision.

Overall, the method used for propagating probabilities in reasoning under uncertainty depends on the specific
system and the relationships between the variables. Different methods may be appropriate for different types
of systems and applications.
 Thank you
Aljun Yosores Jutba Charlotte E Ploma Esnaira Gandawali Jarriza May M. Cuasito Meky Quirante
Nikki Babor Beljano Rhea Jane Alangcas Stefany Dizon Wilfredo Catalan