Download Page 113 - JUfiles

Week 8, Lecture 19
Fuzzy Logic [Page 113]

Fuzzy Logic is a mathematical method of handling imprecise or subjective information.
o The basic approach is to assign values between 0 and 1 to vague or ambiguous
information. The higher the value, the close it is to 1.
o Therefore, you might assign 0.8 to the value “HOT”.
o Then you would construct rules and processes called algorithms to describe the
interdependence among variables.
o A fuzzy logic algorithm is a set of steps that relate variables representing inexact
information or personal perceptions.

Fuzzy logic is an important part of most text analytics systems. If you remember, text
analytics converts textual information into structured information. Here, fuzzy logic is
often employed to assign numerical values to comment data on a s survey form.

Fuzzy logic is also very good at disambiguation, the ability to specifically identify a
named entity recognition based on surrounding text.
Genetic Algorithms [Page 113]

Today significant research in AI is devoted to creating software capable of following a
trail-and-error process, leading to the evolution of a good result. Such a software system
is called a genetic algorithm.

A Genetic algorithm is an AI system that mimics the evolutionary, survival-of-the-fittest
process to generate increasingly better solutions to a problem.

A Genetic algorithm is an optimizing system that finds the combination of inputs that
give the best outputs. Here are a few examples of genetic algorithms in action:
1. Staples used a genetic algorithm to evaluated consumer responses to over 22,000
package designs to determine the optimal set of package design characteristics.
2. Boeing uses genetic algorithms in its design of aircraft parts such as the fan blades
on its 777 jet. Rolls Royce and Honda also use genetic algorithms in their design
processes.
3. Retailers such as Marks and Spencer, a British chain than has 320 stores, use
genetic algorithms technology to better manage inventory and to optimize display
areas.
Dr. Anas Aloudat
Page 1 of 6



Suppose you need to create a portfolio with the combination of 20 stocks out of 30
possible. Some of the conditions you would specify for the genetic algorithm is to
examine the combination of a 20 based on:
 The number of years the company has been in the business
 The performance of the stock over the past five years.
 Price to earnings ratios
 Other information
The possible combinations from 30 stocks are 30,045,015. For a 40-stock pool, the
number of combinations rises to 137,846,500,000. It would be an impossibly timeconsuming. This is the sort of repetitive number-crunching task a computer and genetic
algorithms would excel at.
Genetic algorithms are important part of the analytics professional’s tool set. And because
genetic algorithms are best suited for problems with millions and millions of possible
outcomes, they facilitate the evaluation of all those outcomes with great speed and
efficiency.
Agent-Based Technologies
o An agent-based technology, or a software agent, is small piece of software that acts on
your behalf (or on behalf of another piece of software) in performing tasks assigned to it.
o Essentially, an “agent-based technology” is an “agent” much like an agent that represents
a movie star or athlete, performing assigned tasks like negotiating contract terms and
setting up media blitzes.
o There are five main types of agent-based technologies (see Figure 4.8). They include:
1. Autonomous agent–software agent that can adapt and alter the manner in
which it attempts to achieve its assigned task.
2. Distributed agent–software agent that works on multiple distinct computer
systems.
3. Mobile agent–software agent that can relocate itself onto different computer
systems.
4. Intelligent agent–software agent that incorporates artificial intelligence
capabilities such as learning and reasoning.
5. Multi-agent-system–group of intelligent agents that have the ability to work
independently but must also work with each other in order to achieve their
assigned task.
o Our focus will be on the last two–intelligent agents and multi-agent systems, both of
which incorporate some form of artificial intelligence (AI).
Dr. Anas Aloudat
Page 2 of 6
Intelligent Agents [Page 114]

Intelligent agents incorporate some form of AI–like reasoning and learning–to assist you,
or act on your behalf, in performing repetitive computer-related tasks. There are four
main types of intelligent agents:
1. Information agents (Buyer agents or shopping bots)
2. Monitoring-and-surveillance agents
3. Data-mining agents
4. User or personal agents

We will focus on data-mining agents because of their use in the field of analytics. But, we
will define the other three as well to be familiar with them:
o Information agents– are intelligent agents that search for information of some kind
and bring you the information back. The best known information agents are buyer
agents (also known as shopping agents), agents on a Website that help you, the
customer, find products and services you need.
o Monitoring-and-surveillance-agents–intelligent agents that constantly observe and
report on some entity of interest, a network, or manufacturing equipment, etc. to warn
of possible trouble before it happens, including bottlenecks and failures.
o User agents (personal agent)– intelligent agents that take action on your behalf, such
as sorting your emails by priority, dumping unsolicited email into your spam folder,
and playing computer games as your opponent.
Data-mining agents [Page 115]

A data-mining agent operates in a data warehouse discovering information.

Data mining is the process of looking through the data warehouse to find information that
you can use to take some action, such as ways to increase sales.
o It called “data mining” because you need to sift through a lot of information for
unknown-before key knowledge piece of information or indicators.
Dr. Anas Aloudat
Page 3 of 6

Data mining could help you to answer questions that you cannot answer using regular
databases queries, such as “what else do people buy on Thursday afternoon when they
come to buy baby diapers?”

One of the most common types of data mining is classification, which find patterns in
information and categorize items into those classes (exactly as the work of neural
networks). That is why neural networks are usually part of many data-mining systems.
Data-mining agents are another integral part, since these intelligent agents search for
information in a data warehouse.

A data-mining agent may detect a major shift in a trend or key indicator. It can also detect
the presence of new information and alert you. For example, Volkswagen uses an
intelligent agent system that acts as an early-warning system about market conditions.
Multi-Agent Systems [Page 115]

What do a cargo transport system, book distribution centers, the video game market, a flu
epidemic, and an ant colony have in common?
 They are all complex adaptive systems and thus share some common
characteristics.

By observing parts of an ecosystem (‫)النظام البيئي‬, like ant or bee colonies, artificial
intelligence (AI) scientists can use hardware and software models that incorporate insect
characteristics and behavior to:
1. Learn how people-based systems behave;
2. Predict how they will behave under a given set of circumstances; and
3. Improve human systems to make them more efficient and effective.

This concept of learning from ecosystems and adapting their characteristics to human and
organizational situations is called biomimicry.

In the last few years, AI research has made much progress in modeling complex
organizations as a whole with the help of multi-agent systems.

A multi-agent system, is a group of intelligent agents that have the ability to work
independently but must also work with each other in order to achieve their assigned task.
Again, think about ants. Each has a specific task and works throughout the day on that
task without any management oversight. But, no ant alone can build a colony of canals,
an ant hill structure, and so on. That takes all the ants working together.

Multi-agent systems are being used, for example, to:
1. model stock market fluctuations (‫ تقلب‬،‫)تأرجح‬.
Dr. Anas Aloudat
Page 4 of 6
2. predict the escape routes that people seek in a burning building.
3. estimate the effects of interest rates on consumers with different types of debt.
4. anticipate how changes in conditions will affect the supply chain.
Swarm (Collective) Intelligence [Page 116]

The ant ecosystem is one of the most widely used types of simulations in business
problems. Individual ants are autonomous, acting and reacting independently. But, ants
are also social insects.

The “social” term implies that all the members of a colony work together to establish and
maintain a global system that is efficient and stable. So, even though the ants are
autonomous, each ant contributes to the system as a whole.

Ants’ extraordinary evolutionary success (been on Earth for 40 million years) is the result
of ants’ collective behavior, known as swarm (‫ مجموعة‬،‫ سرب‬،‫ )حشد‬intelligence.

Swarm (collective) intelligence is the collective behavior (‫ )سلوك الجماعي‬of groups of
simple agents that are capable of devising (come up with, create) solutions to problems as
they arise, eventually leading to coherent global patterns.

So, how are the workings of ant colonies related to IT in modern business?
Swarm intelligence gives us a way to examine collective systems where groups of
individuals have certain goals, solve problems, and make decisions without
centralized control or a common plan.
o In an ant colony, the forager ants have the sole responsibility of providing food to
the colony. They don’t form committees and discuss strategies or look to a central
authority for direction; they just find food and bring it back to the colony, and in
doing so they follow a simple procedure.
o Say two ants (A and B) leave the same point to search for food, the shortest path is
the one that will be selected. A trail of pheromones (a biological smell from the
ant) is left by the ant to indicate the path to the food so as other ants can use the
same short path to find food.
o So the ants’ approach is straightforward but effective, and can be expressed in the
following three simple rules:
1. Rule 1: Follow the trail if one exists, otherwise create one.
2. Rule 2: Find food.
3. Rule 3: Return to the nest, making a pheromone trail.
Dr. Anas Aloudat
Page 5 of 6

The problem that the ants just solved is one of the oldest problems in business world
and it is very similar to the one solved by the ants. It is known as “the shortest path
problem” or “travelling salesman problem”. Any company that schedules drop-off
and pick-up routes for delivery trucks, or schedules jobs on the factory floor, or even
colors maps, making sure that no two adjacent components have the same color, has
had to find a solution to the same type of problem.

AI researchers built sets of small robots and incorporated software that allowed the
robots to follow rules and interact with each other in the same basic ways as the ants.
They also dispensed with the physical forms altogether, creating virtual ants in the
form of small autonomous blocks of code that we call intelligent agents. And each
code block could follow certain rules, interact, and adapt. These virtual ants were then
arranged into multi-agent systems that were further refined into agent-based models.
Dr. Anas Aloudat
Page 6 of 6