Weist 2016
Weist 2016
Weist 2016
ABSOLUTISM vs RELATIVISM
By
Darren Weist
THESIS
Approved by:
Thesis Advisor
Second Reader
DEDICATION
To all family, friends, and faculty. Thank you for all of your love and support!
TABLE OF CONTENTS
CHAPTER 1 - INTRODUCTION………………………………………………… 1
1.1. OVERVIEW OF THESIS……………………………………………… 1
1.2. ISSUES FOR ARTIFICIAL INTELLIGENCE....................................... 2
1.3. OVERVIEW OF ABSOLUTISM AND RELATIVISM…………….... 3
1.3.1. ABSOLUTISM………………………………………………… 4
1.3.2. EXAMPLES OF ABSOLUTISM IN AI………………………. 6
1.3.3. RELATIVISM…………………………………………………. 8
1.3.4. EXAMPLES OF RELATIVISM IN AI……………………… 10
CHAPTER 3 - AXIOLOGY…………………………………………………………13
3.1. ETHICS…………………………………………………………………14
3.1.1. RIGHT vs WRONG……………………………………………15
3.1.2. GOOD vs EVIL…………………………………………………18
3.1.3. CENSORSHIP………………………………………………… 21
3.2. AESTHETICS - ART, BEAUTY, AND TASTE………………………24
CHAPTER 7 - CONCLUSION…………………………………………………… 64
REFERENCES………………………………………………………………………65
CHAPTER 1 - INTRODUCTION:
1.1 - OVERVIEW OF THE THESIS
The thesis is organized in the following approach. First, we will address some of the
main issues for general artificial intelligence. Second, to enhance our understanding of the
problems, this article will provide a general overview of two axiological theories presented from
philosophy: absolutism and relativism. We will discuss some examples of how these concepts
relate to machine learning algorithms. Third, we will argue the thesis statement that
not impossible. In the scope of general artificial intelligence, absolutism is not necessary. The
concept of relativity can be useful for: (1) axiology, (2) defining things, and (3) memory. Then,
the paper will conclude with future and final thoughts on the subject matter.
1
1.2 - ISSUES FOR ARTIFICIAL INTELLIGENCE
Narrow or weak artificial intelligence runs into the issue where it can only perform
specialized tasks. These algorithms are excellent for executing static tasks, but fail to perform
other jobs that lay outside of the limited scope of what they were originally designed to
accomplish. To get to the level of general or strong artificial intelligence, machines will need to
comprehend things in a relative context. For instance, they will need to overcome challenges,
such as:
Being capable of learning without the use of large volumes of training data and time.
2
1.3 - OVERVIEW OF ABSOLUTISM AND RELATIVISM
This paper covers several philosophical topics. Before we jump further into the thesis,
first we will introduce you to the two main terms discussed throughout this paper to give you a
better understanding of what they mean and provide you some examples of how they can relate
to artificial intelligence. The field of philosophy debates two interesting moral theories:
absolutism and relativism. This section will present a general overview of these terms and help
explain them.
3
1.3.1 - ABSOLUTISM
One of the main theories of morality is absolutism. Absolutism is the idea that there is
one right answer, independent of context or perspective. There is only one right way to represent
something. Referring to something that is unchanging and always that way. We can say
something is true, unconditionally, no matter what. For example, the statement all rectangles
have four sides. That is considered an absolute statement. It holds true no matter what. We
cannot have a three sided rectangle, can we? Some of the main absolutist philosophers include:
In the book, Doing Philosophy, Schick and Vaughn describe absolutism as “the
view that there is only one correct way of representing the world” (Schick, 2004). For example,
the act of killing might be considered to always be unethical. In this theory, the action would be
viewed from an absolute perspective, without dependency on context. Even if the act was in a
scenario of survival or for the greater well-being of others. Killing something, whether it be a
plant or an animal, might always be measured as immoral. Schick and Vaughn also discuss the
concept of universal moral laws. These universal laws are considered to be common moral
standards. They are referred to as a fixed set of standards we can use to judge our actions. They
are self-evident truths. Meaning, to support our belief, we don’t need any further evidence. Self-
evident truths contain their own justification and do not need any additional proof. For instance,
Schick and Vaughn consider two potential universal laws. One being the principle of justice.
The concept of justice is that we should treat equals equally. And, the second one is the principle
of mercy. The concept of mercy is that unnecessary suffering is immoral. However, some
suffering is alright. For example, the minor pain from exercising is necessary for the greater
4
good of your health. On the other hand, unnecessary pain from torturing an innocent child is
considered wrong.
standard. Absolutism is described as a set of common standards and values for ethical reasoning.
McDonald says absolutism “dictates that an omni-present set of standards should apply
universally, being equally valid in all places and times” (McDonald, 2010). There are moral
norms like honesty, loyalty, and integrity. These norms are seen as basic ethical standards.
These principles are asserted by many societies, however, observance of these standards may
differ significantly.
The entry on Relativism, by Bagharamian and Carter, clarifies the view of absolutism as a
main contrast to relativism. They describe absolutism as the idea “that at least some truths or
values in the relevant domain apply to all times, places or social and cultural frameworks”
(Bagharamian 2016). They explain absolutism values as being universal. For example, a truth
or value that is unbound by social or historical circumstances. It is an idea that does not change
5
1.3.2 - EXAMPLES OF ABSOLUTISM IN AI
In the context of artificial intelligence, some machine learning algorithms follow along
the lines of absolutism for coming up with classes and assigning labels. For example, the Multi-
Layered Perceptron (MLP), neural networks, and crisp set methods use more of an absolute
items by using predetermined features and labels. For example, several neural networks and the
Multi-Layered Perceptron methods use prearranged sets of training data to help teach or train the
machine. The training data is assigned labels to let the machine know how to classify the objects
it is given as input. The machine is given an input object, it extracts feature values from the
object, it evaluates the feature values through a classifier, and then it makes a decision as to what
the object should be classified as for the output. The machine has to decide from a predetermined
set of features and labels what to classify the object as. The object gets classified under one
absolute label from the given set of labels that the machine was arbitrarily programed to use.
Neural Networks: With neural networks, when transitioning from one task to another,
they tend to start over with a blank slate. They need to go through several iterations of training
to learn the new environment, variables, and mechanics to perform the new task successfully.
With this type of absolute approach, the machine is not able to recognize relative similarities
from one task to the other. And, it becomes difficult transitioning to a changing environment.
Carey’s article, Inside DeepMind’s Latest Attempts to Achieve a General Artificial Intelligence,
discusses some of the challenges Google’s DeepMind has been working on. In 2015, DeepMind
developed an agent that uses a deep neural network combined with reinforcement learning to
6
master a number of different Atari 2600 games. After several iterations, the agent becomes
capable of performing each one of these Atari games at a superhuman level. However, the
neural network learns to master each game separately, one at a time. For example, the same
neural network could not transfer from one game over to another game and still retain its learned
skills like a human. This remains a challenge for neural networks and reinforcement learning.
Usually, they are trained using large quantities of data and time in order to learn a specific task.
For instance, learning how to recognize images, listen to music, and play games. But, it is very
difficult to learn all those different tasks together. Just learning how to perform different games
on a single neural network is said to be a hard challenge. Currently, neural networks do not
maintain their ability to perform an old task after they are trained for a new one. In order to
progressive or continual neural network designed to perform multiple tasks on a single neural
network.
Crisp Set Methods: In Zimmermann’s book, Fuzzy Set Theory - And Its Applications, he
explains the use of crisp sets. Most traditional crisp techniques are deterministic or dichotomous.
Meaning, they follow an absolute valuation like the Boolean logic of true or false. It can only be
a true or false value. There cannot be some variation between the given values. An element
either belongs in the set or it does not belong. For instance, we could set up a classifier that
would categorize vehicles according to engine type using only two classes. The classifier would
label the cars as either a gas or an electric vehicle. However, if we were to encounter a hybrid
vehicle that utilizes both gas and electric, then determining the class may be an issue. It would
7
1.3.3 - RELATIVISM
Another main theory of morality is relativism. Relativism is the idea of using the context
of different perspectives. We can say something is true relative to a point of view. For example,
relative to an object, person, time, location, etc... The statement all rectangles have four sides
can be relative. There are quadrangles (objects with four sides) and non-quadrangles (objects
that don’t have four sides). The non-quadrangles help define the identity of quadrangles. And
perhaps, the term rectangle can take on a different meaning based on time or location. For
instance, definitions to terminology can change over time and the word rectangle might have a
different meaning to individuals in another region. Some of the main relativist philosophers
Schick and Vaughn refer to subjective relativism as the idea that “what makes an action
right for someone is that it is approved by that person.” (Schick, 2004). In this theory, morality
is relative to different views. It claims the appropriateness of an action is always relative to the
individual. For example, if Bob says something is right, he means it is right for him. Nothing
can be completely right or wrong in absolute terms. The view can change from person to
location. Schick and Vaughn describe cultural relativism as the concept that “what makes an
action right is that it is approved by one’s culture” (Schick, 2004). Here, the context changes
McDonald discusses how ethical relativism is not just about people having different
moral thoughts. He explains that what is right for one person, country, or period may be wrong
in another. The moral standards can be different across groups, cultures, and time. In addition to
the premise of having different moral standards and beliefs, ethical relativism calls into question
8
if there can be any commonality or universal set of standards. When it comes to moral beliefs,
relativism argues that there is no uniformity. Our moral beliefs are relative to individual views
9
1.3.4 - EXAMPLES OF RELATIVISM IN AI
In the context of artificial intelligence, some machine learning algorithms follow along
the lines of relativism for coming up with classes and assigning labels. For example, fuzzy sets,
Dempster-Shafer, and histogram slicing methods use more of a relative technique for assigning
value to objects.
Fuzzy Sets: Zimmermann describes the use of Fuzzy Set theory. Unlike most traditional
crisp techniques, fuzzy sets are not deterministic and absolute. The assignment of elements is
more relative. Elements in a fuzzy set can be allocated degrees of membership. Meaning,
elements can be in one set as well as part of another set. For instance, given a classifier that
categorizes vehicles into gas or electric classes. If the classifier encountered a hybrid car that
uses both gas and electric engines, then the fuzzy classifier would be able to assign the vehicle
under both gas and electric classes. Fuzzy sets are suitable for scenarios where objects may have
a degree of membership in several different classes. It is useful for hand writing recognition and
method uses mass and probability assignment to compute a proposition’s plausibility and belief.
The classifier can vary from separate to continuous outputs depending on the class-wise
likelihood measure given to each class. Instead of just identifying to a best matching class,
Farmer’s image classification process would assign a relative confidence level to each candidate
class. Then, the confidence outcomes are represented in a histogram where it can perform a
10
histogram slicing process. The histogram is sliced horizontally to allocate probability masses to
the subsets. That way, it groups the different levels to the subsets.
11
CHAPTER 2 - THESIS STATEMENT AND SCOPE
Thesis Statement: This paper argues the thesis statement that classification requires
relativism to be useful. Absolute classification can be very difficult, if not impossible. In the
area of general artificial intelligence, absolutism is not necessary. The concept of relativity can
be useful for: (1) axiology, (2) defining things, and (3) memory.
Scope: For the scope of this thesis, the focus is on using relativism as a general strategy
and not pursuing further philosophical claims, such as: which theory is correct and debating the
concept of truth. The goal of this paper is to show how relativism can be used as a strategy to
help solve some of the issues for general artificial intelligence. There are cases in the domain of
artificial intelligence where the method of relativism can be useful. Some cases are
straightforward while others are not as clear. We will begin with the more established axiology
examples of ethics and aesthetics in which relativism can be useful. And, then work our way to
the other instances in which the method of relativism can be useful for defining things and can be
12
CHAPTER 3 - AXIOLOGY
epistemology, axiology, and logic. When it comes to classifying and defining values in the field
of machine learning, the philosophical subject of axiology can help us explore these concepts.
Schick and Vaughn explain axiology as the study of value. For example, it discusses the
fundamental questions, such as: What is value? How do we define it? How do we attain value?
In this section, we will review some familiar examples in which relativism can be useful.
For instance, relativism can help us understand the areas of ethics and aesthetics.
13
3.1 - ETHICS
Within the branch of axiology, there is the sub-topic of ethics. Ethics is the study of
moral value. For instance, what is the best action for you or for society? What actions do we
take in particular situations? What is right and what is wrong? It explores the ideas of good and
evil. How do we deal with justice and crime? And, discusses the influences of nature versus
nurture.
We will review how relativism can help us understand some of these areas in more depth.
First, we will go over the issues with right and wrong actions. Second, we will take a look at the
concept of good and evil. And, then discuss some of the concerns with censorship.
14
3.1.1 - RIGHT VS WRONG
Relativism can help us understand between right and wrong actions. For example, we
will examine the concept of absolute universal laws, observe how morality can change over time,
Absolute Universal Laws: Theories based on absolutism claim that there exist some
universal laws of morality. However, even those universal rules are subject to a frame of
reference. Schick and Vaughn talk about two universal laws that could be followed as common
moral standards we can use to judge our actions. The principle of justice and the principle of
mercy. The concept of justice states that we should treat equals equally. However, this absolute
law seems too vague and open to relativity. As a counter example, how do we determine who is
considered an equal? We may determine one person to be equal and then omit someone else
based on some type of measurement. How do we measure this equality? It could depend of a
number of things which is more along the lines of relativism than absolutism. The concept of
mercy states that unnecessary suffering is immoral. But, as a counter example, how do we
determine what is considered necessary and what is unnecessary suffering? Again, this law is
also too ambiguous and open to relative interpretation. Relativism argues that actions can be
measured as either moral or immoral, depending on the point of view. For example, some
people may consider an act to be necessary while others consider the same act to be unnecessary
Morality over Time: Things can change over a period of time. Whether it is the laws or
entities ourselves as we age. Relativism is more open to variations in time than absolutism.
Baghramian and Carter mention how actions and judgments prevalent in society can change over
15
time. They criticize how abominations in the past have been approved by moral code. For
example, the toleration of human sacrifices, slavery, cannibalism, and physical torture.
McDonald’s article also explains how our moral behavior can shift over time. Our standards for
what we deem as acceptable practices have changed over the years. Whether it is for a small
gift-giving practice at your company or dealing with large historical atrocities. After World War
II, for example, society has become less tolerate of people justifying their immoral actions based
on a claim that they were just doing their duty. And, McDonald talks about how traditional and
cultural values have changed as they have been passed down generation by generation. When
living conditions change, then new moral norms evolve to justify the new behaviors.
Morality over Location: Things can change from one location to another. Relativism is
more open to variations in location than absolutism. Baghramian and Carter state that relative
beliefs on right and wrong can diverge greatly across individual, social, and cultural
backgrounds. McDonald also argues that ethical relativism can view morality differently
according to individual, group, and cultural levels. For example, at the individual level, each
person may have a different interpretation for a given situation and believe their own actions to
be moral. Meaning, each individual holds the standards to judge their actions. However,
McDonald does point out some criticism of this level as being naïve and lazy. Stating that it
would be hard to judge or improve an individual’s moral standards. Next, McDonald talks about
the group level of relativism. This is where moral standards are attributed to a person’s role,
organization, or social group. Morality can be determined by different groups rather than by the
individual. Rules and standards are established for each group. And, these collective standards
can vary from one group to another. The criticism of group level morality is that it is also
difficult to evaluate and improve. It may be hard for an individual to decide what is right or
16
wrong when they belong to several social groups or organizations that may follow conflicting
norms. McDonald also discusses the relative cultural level. Cultural relativism describes value
judgments according to the relative context of someone’s culture or community. For instance,
what we consider right or wrong depends on our strongly held cultural beliefs. One culture may
hold values that differ greatly from another culture. For example, if you travel on a business trip
from one region to another, you may experience a diversity of cultural norms. Places can have
different standards for greetings, etiquette, and table manners. And, you will need to learn what
is morally acceptable in their culture to earn their respect. However, McDonald does point out
criticism of the cultural perspective. Like the other levels, it may be difficult to determine right
or wrong actions in the scenario when a culture’s values clash with another culture. Or, when an
individual belongs to more than one culture that holds conflicting moral values.
17
3.1.2 - GOOD vs EVIL
What is considered evil? And, how do we differentiate what is good from what is evil?
Relativism can be more useful for classifying concepts of good and evil. For example, we will
discuss the concept of evil and then argue how one term is dependent on the other with the
ontological defense.
The Concept of Evil: There are several theories when it comes to defining evil. In the
book, The Lucifer Effect: Understanding How Good People Turn Evil, psychologist Philip
Zimbardo defines evil as “intentionally behaving in ways that harm, abuse, demean, dehumanize,
or destroy innocent others” (Zimbardo, 2008). In philosophy, the problem of evil is a subject
that is commonly discussed along with the existence of God issue. For the scope of this paper,
we will focus mainly on the topic of good and evil and avoid the religious arguments. In the
article, The Concept of Evil, Calder explains that there are two main notions of evil. There is a
narrow view and a broad view. With the narrow theory, evil is defined as the kind of actions or
characters that are the most morally appalling. It is attributed to mainly just moral agents and the
actions performed by moral agents. For instance, if we consider a human being as a moral agent,
then that individual is capable of performing evil actions. On the other hand, with the broad
theory, evil is defined in a more general sense. The broad concept of evil is described as any bad
action, character, or event. For example, he explains that the minor suffering from a toothache or
telling a white lie can be considered evil. The broad view of evil covers a lot. It is broken down
into two sub categories. The two groups are natural and moral evil. Calder describes natural evil
as the bad state of events that are not intentionally caused by moral agents. For instance, a
natural disaster such as a hurricane would be an example of a natural evil because it was not the
18
result of an intentional human action. In opposition, Calder defines moral evil as the bad actions
or events caused by moral agents. For example, a human being intentionally murdering or lying.
The Ontological Defense: Does good depend on evil? Schick and Vaughn discuss this
idea with the ontological defense. This argument claims that goodness cannot exist without evil.
It is impossible to have a good domain that contains no evil. However, Schick and Vaughn
assert that good is not a type of evil. Meaning, they are contradictory ideas. Schick and Vaughn
give an example using colored objects. For instance, a red thing and a non-red thing. They say
these contradictory concepts do not depend on each other. Meaning we can have a red thing
absolutely by itself without the existence of a non-red thing. They give a thought experiment
where we assume everything in the world is colored in red. In this scenario, the world does not
contain any non-red objects. However, they assert that red objects can exist absolutely without
having non-red things. Therefore, goodness can exist absolutely without evil in the world. The
lack of one thing does not prevent the existence of the other. Mackie’s article Evil and
Omnipotence makes a similar claim. He argues that goodness and evil are not dependent on one
another. Mackie believes it is fallacious for evil to be a necessary counterpart to good. He uses
a couple of thought experiments. Mackie uses the color example with red things and non-red
things. And, he gives another example of relating good and evil to the terms of great and small.
Mackie says the terms great and small are used relatively as counterparts. Greatness cannot exist
without smallness. However, he believes that great and small can be also be absolute. We can
have one quality without the existence of the other. He states that it is possible for everything to
be small or everything to be great. Therefore, goodness can exist absolutely without having evil.
I argue against the assertions made by Schick, Vaughn, and Mackie. I support the
ontological defense that claims goodness and evil are relative to one another. It makes sense to
19
consider that good cannot exist without its opposite. We can provide a counter-example to the
color thought experiment mentioned above. For instance, if everything in the world was colored
red, then how would we know what red is? Red relative to what? The color would be
undefinable by itself. If everything was consistently red within our reference frame of space and
time, then we would not be able to define red uniquely without an opposing non-red object. We
would not be able to perceive red. We would have nothing to distinguish red from non-red
things. Therefore, the terms good and evil are better defined using relativity. They need to be
put into perspective. Good and evil do not exist absolutely by themselves. Good and evil can be
considered as two labels on the same axis of measurement. Like temperature, you cannot
perceive hot without cold. Or tactile sensation, how would we be able to experience pleasure
without being able to feel pain? And, what is a superhero without a villain of some sort? For
example, Batman would be a very boring series without the Joker or some other evil guy to
oppose him. There would be no drama. If there was no bad guy to terrorize the city of Gotham,
then the city would not need a good guy to save it from destruction.
20
3.1.3 - CENSORSHIP
What makes a word good or bad? Why do we ban certain words, images, books, music,
videos, or other forms of art from our society? Relativism can be more useful for classifying the
value of a word. In his book, Bad Language: Are Some Words Better than Others, Battistella
argues how language is variable. He claims “that good and bad language cannot be defined in
absolute terms” (Battistella, 2005, p.8). For example, the semantics or meaning of a word can
Terms by Location: There are different laws in different locations when it comes to the
use of offensive language. Some areas in the world are more stringent on censorship than other
areas. Battistella describes how there are different levels of acceptance for offensive language
with different types of media and communication. Often, bad language is prohibited for the idea
of protecting observers, stereotypically women and children. For instance, Battistella mentions
the example of a man who was fined 100 dollars in 1999 for violating an old 1897 law in
Michigan. He broke a law that prohibited cursing in the presence of women and children.
Offensive words can vary from place to place. For example, the word fart means speed in
Danish and Swedish. And, the word slut means final or end in Swedish.
Terms over Time: In addition, the meaning of a word can change over time. For
example, a good term may eventually become bad. Or, on the other hand, a bad term may
change meaning and become good. Battistella talks about the history of cursing and how coarse
language changes. The avoidance of profanity used to be a display and distinction of class
symbolism. Then, by the twentieth century, American tolerance towards offensive language
started to shift. With the arrival of the movie industry, the impacts of several wars, and the
21
emergence of new forms of media, people eventually became more relaxed to the use of
offensive words. And, the meaning of coarse language can change over time. Bad terms can
become less taboo. Even the word bad was once considered offensive in the past. According to
Harper’s Online Etymology Dictionary, the term bad was originally used as a derogatory term in
Terms by Context: Battistella categorizes offensive language into different groups. For
one, there is the epithets group. This is for words that are considered slurs that relate to things
like race, gender, and sexuality. The profanity group is refers to religious cursing, like the words
hell or damn. And, then there are the vulgarity and obscenity categories. These groups are for
words that describe sexual anatomy and crude excretory acts. Even though bad words are
offensive, they still convey meaning. Battistella discusses how words are used to confront
underlying ideas. For instance, words can hold an emotive force and it can be challenging to ban
Bad words are considered taboo to a society based on the deviance of the term’s meaning.
However, good words don’t exist without the existence of bad words. Bad terms are needed to
define good words. We cannot just banish a set of terms that are needed to fulfill the definition
of another set of terms. In the place of where the set of exiled words used to exist, a new set of
terms will emerge to replace the void. Bad words give meaning to good words, and vice versa.
It’s hard to get rid of the bad without diminishing the good.
To an extent, as a bad word becomes increasingly rare and taboo, the bad word may
develop a stronger definition in the process. An attempt to erase a word from a society’s
22
vocabulary may end up giving the word a stronger meaning of deviance. It is kind of amusing
how the process of censorship essentially gives our offensive words stronger meaning.
23
3.2 - AESTHETICS:
ART, BEAUTY, AND TASTE
Aesthetics is the branch of axiology dealing with how we value things such as art, beauty,
and taste. For instance, how do we judge a piece of art? Absolutist would argue in favor for
methods we can use to universally place value across all individual works of art. On the other
hand, relativism argues against universal standards. Relativism can be more useful for
classifying aesthetics. The way we value art, beauty, and taste can change over different
contexts. For example, the definition of art is vague, art can be valued differently over time, and
How do we define art? We do not have a strict, absolute definition for what is
considered art and what is not. Where does art begin and where does it end? For instance, the
line is not so clear for work utilizing nude figures. Some may consider it art, others may dismiss
it as pornographic material. Art can include several different forms, such as: drawings,
paintings, sculptures, photography, literature, poems, music, concerts, film, operas, plays,
dance/ballet, architecture, graphic design, video games, graffiti, cooking, gardening, fashion,
etc... Plus, there can be several different genres within each of those examples. Music has
classic, jazz, blues, rock, rap, hip-hop, country, etc... Film has action, comedy, science fiction,
thriller, fantasy, etc... New genres can evolve over time and location. Also, it is unclear what
determines a piece of art as beautiful or ugly. Can a piece of work be so horrible that it actually
becomes valuable? Cynthia Freeland’s book, But is it Art, argues that art includes both the good
and the bad. She explains how art embraces both the formal, beautiful pieces with positive
morals as well as the ugly and disturbing pieces with negative morals (Freeland, 2002, p.28-29).
24
How do we value art? There is no one absolute standard that everyone uses to value all
forms of art. When it comes to valuating art, is the value influenced by the materials used, by the
medium it was constructed in, by the frame that surrounds it, by the techniques used, by the time
it took to create it, by the artist, or is the value also influenced by the emotions and experience
that it provides to the audience? There are several factors that can relatively influence the value
Value of art over time: The valuation of art can change over time. For instance,
Freeland describes the classic imitation theory of art where pieces are created to imitate nature
and humans. Innovations using perspective techniques and oil paintings created ever more
realistic looking copies of objects and nature. By the late 1800s, the development of
photography had a large influence on art. And, imitation became less of an objective for several
art genres, like impressionism and abstraction. (Freeland, 2002. p.31-35) Freeland discusses
how Andy Warhol’s plywood Brillo Boxes were debatable sculptures that were not accepted
right away as art. These ordinary looking soap boxes puzzled many critics and opened our views
of what may be considered art. (Freeland, 2002. p.56-59) Museums are viewed as institutions
that maintain classical standards for art and how it is valued. Over time, however, museums
have been changing their standards and displays. Freeland discusses how museums have been
arranging their displays and updating their locations with add-ons to attract more people and
different audiences. For example, museums are now adding things like restaurants, gift shops,
and theatres, and film sections. (Freeland, 2002. p.103-104) The value of art can change over
time. She gives an example how an artist in the past, like Van Gogh, lived in poverty and found
it difficult to sell his work. Then, over time, his pieces of work can now sell for astronomical
amounts of money. (Freeland, 2002. p.105-107) Freeland concludes that there does not appear
25
to be any universal standards for art. Art evolves over time. There is no common law that
dictates what makes a piece of art beautiful or valuable. (Freeland, 2002. p.208)
Daniel Grant’s article, Is an Artist Only Appreciated after He Dies, touches on one of the
greatest factors that influences the value of an artist’s work. The market supply and demand
over time. For example, he mentions how collectors release pieces of work after the death of an
artist. When collectors release too many pieces over a given period of time, they may end up
flooding the market. A sudden increase in supply can decrease the value of these particular
works of art. He gave an example of Pablo Picasso’s paintings going down in value for a period
of time as a result of an influx of his pictures being supplied to the market after his death. Also,
he mentions that the art buying public can have changing tastes. And, these changes in tastes can
The Value of Art: Money, Power, Beauty, by Michael Findlay, discusses how the value of
art can change. He describes how the commercial value of art can be determined by our
collective intentions. Art does not have a fundamental objective value. Our human demands
determine the commercial worth. He gives an example, if people were given millions of dollars
to either buy a painting or to buy a mansion, most people would prefer the mansion over the
painting. He argues that the majority of people do not have criteria or a reliable set of standards
to judge the value of art. Opinions of artwork and reputations of artists can change over time.
(Findlay, 2012)
Value of art over location: The valuation of art can change from one place to another.
Freeland describes how art is not well defined across different cultures and eras. There are many
different practices and roles for artists. For instance, she gives an example of how art may
26
include different things when comparing classical Japanese art to modern Western art. Japanese
art may consist of things like Zen gardens, tea ceremonies, and swords. She explains that in the
context of another culture, it is difficult to know what art is valued and why it is valued. Art may
include African carvings and masks. It may include Hindu dances. Or, it may include Islamic
coins, carpet, and calligraphy. (Freeland, 2002. p.60-66, 76) Even when comparing art work
from within a single place or culture, it can be difficult defining the meaning of art and how it is
valued. Pieces of work from other locations and periods can differ from our modern-day criteria
Findlay gives an example of how the quality of the art can be relative. One individual
may critique a painting and have a completely different experience than another person. He
explains two people can observe the same Picasso painting of a red dog, but process it in
different ways. One person may rank it as high quality because they have a bias for that
particular artist and color. On the other hand, the second person may rank the painting as a lower
quality because they do not like the artist, dogs, nor the color red. Findlay also talks about how
value can depend on the size of the artwork and the market. For instance, larger paintings and
sculptures tend to be more expensive than smaller pieces. However, if the painting or sculpture
is too large and requires a lot of space, then it may not fit in most buildings. By decreasing the
number of potential locations the piece of work can be installed, this can make it more difficult
There are several categories of art. Art can be valued differently depending on the person
critiquing it. For example, some people may have a preference for realistic art. These are
drawings, paintings, or sculptures that are more objective and representational of real life objects.
For example, Michelangelo’s marble statue of David, Vermeer’s oil painting of the Girl with a
27
Pearl Earring, and Millet’s oil painting The Gleaners. Some critics argue for universal standards
of realistic quality that can be used to judge all art. However, not everyone values realism.
(a) (b)
(c)
abstract art. These are pieces of work that are more subjective and may not visually represent
any real world objects. They tend to use utensils, materials, and mediums to depict art in new
and creative ways. It can be used to expresses emotion and produce experiences for the
audience. For example, there is Heizer’s boulder called Levitated Mass, located at the LA
County Museum of Art. Robert Rauschenberg’s minimal, pure white paintings called White
Paintings. And, Theo van Doesburg’s oil painting, Composition VII (the Three Graces).
(a)
(b) (c)
29
Figure 3.2: Examples of abstract art: (a) Levitated Mass - by Heizer, (b) White Paintings - by Robert
Rauschenberg, (c) Composition VII - by Theo van Doesburg.
CHAPTER 4 - DEFINING THINGS
This section will explore some other areas in which relativism can help in artificial
intelligence. These are less familiar examples in which relativism can be useful. Relativism can
be help us understand how we define things, such as: objects, patterns and deviances,
30
4.1 - OBJECTS
Relativism can be more useful for defining objects. Absolutism can fall short of
adequately classifying an object. Absolutism is only part of the equation that is explained by
possible for something to be perceived independently without relation to other things? For
example, the following diagram attempts to define something using only one thing or instance.
Within a given context, imagine the only thing that exists is one instance of something. But, how
can we identify this instance without anything else to compare it to? If the context consisted
entirely of one instance and nothing else, then we would have nothing to put that instance into
perspective. We would not be able to define that instance against anything. We would not be
capable of sensing that instance. That singular instance goes on infinitely. As a result, the
31
Figure 4.1: Examples of absolutism. Example A with a rectangular context. Example B with a circular context.
On the other hand, with relativism, I argue that things need to be defined by a pattern of
at least two instances within a context. The idea is that things are relatively defined. Being able
to define an object relies on a frame of reference. Things have to be put into perspective to
derive meaning and classification. With relativism, an object cannot be defined alone, absolutely
by itself. Relativism needs at least two instances. One instance needs something else to compare
that instance. For example, the following diagram shows how to define something using at least
two instances. Here, an instance is referring to any object or pattern that can be uniquely
defined. An instance can only observed when it is put into context. In other words, in order for
an instance to be perceived, there must also be a perceived absence of that instance in the same
context. A context is an arbitrary frame of reference that can vary in scope or form. A context
could contain one instance, several instances, or perhaps some variation in between (which we
32
will cover later on in the Patterns and Deviances section). But, the important thing to remember
here is that things are relatively defined. In order to perceive an instance within a context, then
Figure 4.2: Examples of relativism. Example A with a rectangular context. Example B with a circular context.
33
4.2 - PATTERNS AND DEVIANCES
Relativism can be more useful for defining patterns and deviances. It can be difficult to
define patterns and deviances absolutely by themselves. I argue that they are more valuable in
range from simple to complex and from consistent to inconsistent. Simple patterns can be
combined to create more complex patterns. The following figure shows a visual representation
of the relationship between repeating instances and consistency. For example, consistency
depends on the degree of repeating instances. The greater number of repeating instances within a
context, the greater the consistency. A decrease in the number of repeating instances will result
in a decrease in the consistency of the pattern. With relativism, a pattern cannot be perceived
absolutely by itself. Like an object, a pattern is defined with the absence of the pattern. The
absence of a pattern is also considered a pattern itself. As the consistency decreases for a pattern
34
Figure 4.3: Example illustrating the relationship between Repeating Instances and Consistency.
The next figure depicts a domain or context filled entirely with one pattern. In this
example, from the perspective of the red pattern, the context contains a high number of repeating
instances, a high consistency, and no deviance. From the perspective of the non-red pattern, the
Figure 4.4: Example demonstrating the low deviance of the red pattern and the high deviance of the non-red
pattern within the given context.
35
The next figure is just the opposite of the previous exhibit. The context is made up
entirely of one pattern. In this example, from the perspective of the red pattern, the context has
no repeating instances, no consistency, and a high deviance. From the perspective of the non-red
pattern, the context contains a high number of repeating instances, a high consistency, and no
deviance.
Figure 4.5: Example showing the high deviance of the red pattern and the low deviance of the non-red pattern
within the given context.
The following figure shows how the patterns and deviances change when the composition
of the context changes. In this example, a small portion of the context is made up of the red
pattern while the remainder of the domain encompasses the non-red pattern.
Relativism can be more useful for defining measurements. For example, relativity can be
useful for comparing size, relativity can be useful for classifying the maximum size of instances,
relativity can be useful for defining the minimum size of instances, and absolute measurements
itself. Measuring the dimensions of an object depends on a relation to other things within a
context. For example, the diagram below shows how we determine the size of instances. In
example-A, we only we only have one instance. In this absolute scenario, it is difficult to measure
the size of the instance without being able to compare it to another distinct instance. In example-B,
we have more than one instance. In this situation, we are now able to measure and compare things
relatively. Instance-A is smaller than instance-B. And, instance-B is larger than instance-A.
37
Figure 4.7: Example-A demonstrating size comparison using only one instance. Example-B showing size
comparison using more than one instance.
Maximum Measurement: Is there an absolute maximum size? For instance, is there
such a thing as an actual edge to our universe? Relativism can be more useful for defining the
outer boundaries of an instance. In the following figure, step-1 demonstrates a context that
contains only one instance. At this point, if no edges can be observed, then the instance can be
considered to stretch out indefinitely. If we are able to perceive an outer edge for instance-A,
then this edge would define another instance beyond instance-A. This brings us to step-2 where
we have instance-A along with the outer instance-B. If no outer edge can be defined for
instance-B, then it may stretch on forever. But, if we can observe an outer edge on instance-B,
then this would define another instance beyond instance-B. This brings us to step-3 where we
have three instances classified. These steps continue on and on. Every time an outer edge is
defined, another instance is also relatively defined. Therefore, it is difficult to define an absolute
maximum size or outer edge. The maximum size may continue on indefinitely.
38
Minimum Measurement: On the other side of the scale, is there an absolute minimum
size? For example, is there such a thing as an absolute minimum instance in our universe?
Relativism can be more useful for defining the inner boundaries of an instance. In the figure
below, step-1 demonstrates a context that contains only one instance. At this point, the instance
pattern made up of smaller instances, then this would bring us to step-2 where we have instance-
C defined by the inner instance-B. If no inner objects can be classified for instance-B, then it
may stretch inward forever. But, if we can observe inner substances that make up instance-B,
then this would bring us to step-3 where we have instance-B as a pattern defined by inner
instance-A. These steps continue on every time a smaller instance is relatively defined.
Therefore, it is difficult to define an absolute minimum size. The minimum size may continue
on indefinitely.
39
Precision and Uncertainty: Precision and uncertainty also plays a role in our
Zimmermann discusses how the accuracy of our measuring techniques have increased over time.
Our technology progresses and becomes more precise at measuring physical features. However,
he points out that uncertainty still exists. He explains that we cannot perfectly measure exact
properties. We only have an indicated measurement. And, the real measurement will have some
When it comes to our standards and measurements, we like things to be precise and
replicable. But, can we rely on absolute, universal constants for our units of measurement?
According to the UK Metric Association, before the 1800’s, measurements often varied from
place to place. This created a problem for trade. To alleviate this issue, a commission in France
developed a unified metric system in 1789. This system included the meter and the kilogram.
By the late 1800’s, the metric system gained international acceptance among the industrialized
countries and the International Bureau of Weights and Measures was formed. A challenge was
proposed to create a prototype for the kilogram. This prototype would be used as an absolute
definition for a kilogram and used for a standard comparison against other constructed kilogram
masses. In the 1880’s, a prototype was made out of an iridium-platinum alloy for its resistance
to corrosion and stability. And, the standard for today’s kilogram for which all other kilogram
objects are measured is still currently based on that prototype. (UK Metric Association, 2013)
However, even these absolute units of measurement can change over time and location.
In Jonathan Fildes’ article, Getting the Measure of a Kilogram, he explains how this variation
has happened. The prototype cylinder for the standard kilogram was constructed along with
several other cylinder copies in the 1880’s. Over incremental periods of time, the cylinder
40
copies were taken back and measured against the original prototype to compare their
consistency. But, over the past 40 years scientists started running into an issue. They discovered
a small drift in the kilogram measurements. The mass of the standard prototype cylinder no
longer matched up with the cylinder copies. And, this drift in measurement continues on today.
Scientists cannot tell if the original prototype is losing mass or if the copies are gaining mass.
The change in mass might be attributed to atmospheric pollutants that all the cylinders are
exposed to. Whatever the reason, the small drift can lead to major consequences for our standard
system of measurements that require a precise definition of mass. When scientists measure the
prototype and all the other copy cylinders against each other, they are not able to determine
absolutely which cylinders are losing mass or gaining mass. We can only make relative
observations to determine that one cylinder weighs more compared to another. (Fildes, 2007)
Brumfiel’s article, This Kilogram Has a Weight-Loss Problem, he mentions how scientists are
searching for a constant to redefine the standard kilogram. One proposed idea is to define the
kilogram using a watt balance and the Planck’s constant. Basically, the watt balance is a highly
calibrated scale that measures mass by electrical and magnetic forces. The forces can convert to
a number based on the Planck constant, which is used for very small scaled calculations in
quantum mechanics. However, physicists note that relating something that small to a large
kilogram object we can hold in our hand is quite difficult. The article mentions that the scale is
so sensitive that there are many sources of noise that can cause issues for precise measurement.
(Brumfiel, 2009)
Fildes’ article proposes another method that redefines a kilogram based on the number of
atoms within a crystal object. For example, this method suggested the use of a silicon sphere.
41
To calculate the number of atoms, we would measure the volume of the silicon sphere and divide
it by the volume of an atom of silicon. However, even this method has difficulties with precision
and uncertainty. The sphere is said to be made up of trillions of atoms and the accuracy of their
Universal Constants: When we measure things, we use the basic measurement types
such as M for mass, L for length, and T for time. At some point, we came up with arbitrary units
to represent each of these types. For example, someone deemed a kilogram as a standard unit of
mass, someone came up with a meter to denote a unit of length, and someone decided on a
second to designate a unit of time. The issue is that each of these arbitrary units are defined to
objects that can change over time and location. So, we attempted to relate these units to things
that are universal throughout the universe. But, are there any absolute, universal constants?
Figure 4.10: Basic measurement types: Mass (M) in kilograms, Length (L) in meters, and Time (T) in seconds.
Scientists believe there are some physical constants throughout the universe. The NIST
reference on Fundamental Physical Constants illustrates some of these equations out for us. For
example, there is G for the gravitational constant, h for Planck’s constant, and c for the speed of
42
light in a vacuum. With only these constants, physicists have come up with the following
equations to represent mass, length, and time. There is M for Planck mass, L for Planck length,
Figure 4.11: Basic measurement types derived from universal constants. Planck mass, Planck length, and
Planck time. Where: G = gravitational constant. h = Planck’s constant. c = speed of light.
But, what exactly are these constants that we threw into our equations? And, are they
really absolute, universal constants? According to NIST, the constant c is defined as the speed of
light in a vacuum. This equates to approximately 299,792,458 meters per second. However, this
What is light? For one, there is the assumption that we know what light actually is.
Scientists have debated the classification of light for centuries. And, we still do not have one
absolute answer for the definition of light. How Light Works, by Harris and Freudenrich,
describes how light continues to amaze scientists. In some scenarios, light appears to have
properties that behave like a particle. In other experiments, light appears to have properties that
43
behave like a wave. In the late 1600s, Christiaan Huygens proposed the undulatory theory that
suggested the existence an ether. This ether was said to be an invisible medium that made up the
empty void of space around objects. And, he thought light might form as a result of objects
causing waves in the ether. In the 1700s, scientists like Isaac Newton debated that light
consisted of particles. Around the 1800s, other scientists came up with experiments to provide
evidence that light was a wave. Thomas Young conceived of the double-slit experiment. He set
up a mechanism that would shoot a beam of light from one end of a device over to a screen on
the other end. Before the light reaches the screen, it would need to travel through a card that
contained a couple of slits. The resulting pattern of light and dark bands that displayed onto the
screen showed that light demonstrated properties of a wave. Later, James Maxwell, argued the
theory of electromagnetism to define light as a type of wave created from electric and magnetic
fields that does not require a medium for light to travel through. By the 1900s, however, Max
Planck and Albert Einstein brought the concept of light as a particle back to life. Planck
introduced the idea that light can carry discrete quantities of energy. And, Einstein used the
concept of light as a particle to explain the photoelectric effect when a light is directed onto a
material and we are able to detect the emitted electrons. So, which definition of light is the
correct one? As of today, scientists accept both models. There is evidence for each side. It is
subject to context. It depends on which kind of experiment is being performed. As a result, the
idea of a wave-particle duality has emerged that defines light as both a particle and a wave.
The article by Koks, Carlip, and Gibbs, Is the Speed of Light Everywhere the Same,
mentions some assumptions with definitions in the laws of physics. For instance, there is the
assumption that a particle of light contains no mass. They explain that it is practically
44
impossible to perform an experiment that could determine if the rest mass of a photon is exactly
zero. And, if particles of light have a rest mass, then the speed of light would vary.
Grant’s article, Speed of Light Not So Constant After All, argues that the speed of light
can vary in vacuum conditions. He talks about a recent experiment that shows how manipulating
the structure of light will reduce its speed. The research reveals how the speed of light can vary
based on the photon structure. For instance, the experiment sends a pair of photons on separate
paths on the way to a detector. One of the photons gets sent through a fiber unaltered. While the
other photon gets sent through a device that modifies the light structure and then adjusts the
configuration back. The pair of photons were expected to arrive at the detector at the same time.
However, they did not. The detector measured the manipulated photon arriving later than the
non-altered photon. The measured gap was several micrometers per meter. (Grant, 2015)
What is a vacuum? The second axiom in the speed of light definition is the assumption
that we know what a vacuum is. A vacuum is said to be space that is void of matter. But, how
do we know a vacuum is really void of all mater? According to Empsak’s article, Speed of Light
May Not Be Constant, Physicists Say, we may think of a vacuum as a unit of empty space.
However, quantum physics argues that this empty space actually contains very tiny particles
(such as: gluons and quarks) that come into and out of existence from the quantum level.
Empsak explains that photons may run into these particles and get re-emitted as they travel
through space. It is proposed that these tiny particles can affect the speed of the photons. And,
In conclusion, it is difficult to classify with 100 percent certainty what light is. And, a
45
4.4 - LANGUAGE
Language, the medium through which we communicate, is more relative than absolute.
Relativism can be more useful for classifying language for linguistics. It can be relative to the
situation that it occurs in. Language, in general, can vary depending on location, time, and
context.
Location: Language can change from one place to another. In Battistella’s book, he
argues how language is continuously changing. Dialects and languages may attempt to follow
standards. However, those standards depend on usage. And, usage is relative. (Battistella,
2005, p.153) When speakers or advertisers communicate to an audience, the dialect and speech
is usually tailored to fit the background of the audience. Variation in dialect can also be seen on
individual words. For instance, he gives an example with the adverb anymore. For Standard
English, the term anymore is primarily used in negative sentences. However, in many locations
throughout the United States, the word is also utilized in positive statements. The language
standards of one region can differ significantly from another. Articles on arts and culture can
differ from one publisher to another. And, speakers alter their styles to relate to their audience.
Things like vocabulary and pronunciation norms get altered. (Battistella, 2005, p.7-8) Language
doesn’t follow an absolute standard in all locations. There can be multiple standards and the
Certain words can take on different meanings in other languages around the world. For
example, in the article 10 English Words that Mean Something Else in Other Languages, Harris
shows how the word gift can change meaning. In English, gift refers to a talent, skill, ability, or
46
present. In German, however, Gift can mean poison. And, in Scandinavian languages, the word
Time: Language can also change from one period of time to another. For instance,
dictionaries provide a good example of how words change over time. We do not use the same,
original dictionaries that were put together several years ago from the past. Eventually, new
words are introduced. New meanings are added for existing terms. And, the rules of grammar
In Zimmermann’s book, he mentions how natural languages depend on many things. For
example, how they change over time and how they change according to the backgrounds of an
individual. He differentiates between the label of a word and its meaning. Zimmermann says
there is usually a one-to-many relationship between the labels and their meanings. He argues
that the meaning of terms are not classified in a crisp way. For instance, generally they are not
absolute. The meaning of words depend on relative context. (Zimmermann, 2006, p.117-118)
Words can shift meaning over time. For example, according to the Oxford English
Dictionary, the word evil used to have an original meaning that was fairly mild. It was just a
general word used for disapproval or dislike. Later on, the term evil took on a more intense
meaning. Today it can refer to something that is malicious or immoral. And, the word awful
used to express something that was worthy of awe or respect. But, now it can stand for dreadful,
terrible, or appalling.
Context and Semantics: The situation of a word in relation to other words and the
arrangement of the words within a sentence can change the meaning of what we are trying to
convey. Zimmermann talks about the fuzziness of natural languages and how the meaning of
47
terms are generally not very clear. For instance, he gives the example of the word birds. Birds
can be used for a label of a set. However, when it comes to classifying which objects fall within
the set of bird and not-bird, the borders of distinction become fuzzy. Classifying an ostrich, bat,
or a penguin may be difficult. Zimmermann also gives examples of the labels tall men and
creditworthy customers. Tall men is considered fuzzy and not absolute. The meaning of tall is
relative to the context of the observer. And, the creditworthiness of an individual is fuzzy as
well. He describes this term as being subjective. The distinction of creditworthy can be based
on different financial valuations and the personality traits of the applicant. (Zimmermann, 2006,
p.4)
Hobbs book on Homophones and Homographs shows how the English language contains
several words that can have multiple meanings. Or, we could be using completely different
terms that end up expressing the same meaning. The following table categorizes these words
Figure 4.12: Table categorizing words into groups according to spelling, pronunciation, and meaning.
48
Homonyms are terms that are spelled and pronounced the same. But, the meanings are
different. For example, words like bark change depending on the context of the other words
surrounding it. We could be using a verb bark to describe the sound of a dog. Or, we could be
Figure 4.13: Example of homonyms. Terms that have the same spelling, same pronunciation, and
different meaning.
Another example is the term off. Off can refer to something being activated or
deactivated depending on the context. For example, we could say something is sounding off and
we need to turn it off. What we really mean is that something is actually on or enabled and now
Homographs are terms that are spelled the same. But, they are pronounced different and
have different meaning. For example, the word wind. We could be talking about the windy
Figure 4.15: Example of homographs. Terms that have the same spelling, different pronunciation, and
different meaning.
49
Homophones are terms that are pronounced the same. But, the terms are spelled
differently and mean different things. For example the words there, their, and they’re. The word
Figure 4.16: Example of homophones. Terms that have the different spelling, same pronunciation, and
different meaning.
There are several words that can be placed within in the synonym category. Synonyms
are terms that refer to the same meaning. However, the words are spelled and pronounced
differently. For example, the terms horrible, awful, and terrible have different spellings and
Figure 4.17: Example of synonyms. Terms that have the different spelling, different pronunciation, and
same meaning.
50
4.5 - DATA vs INFORMATION
Relativism can be more useful for defining information. Information can be made up of
relative data that provides better value than absolute data. It can be difficult to evaluate raw data
when it is presented absolutely by itself. Data needs to be put into perspective in order to derive
meaning from it. For example, we will discuss definitions of data, definitions of information,
What is Data: Nunamaker, Romano, and Briggs article, A Framework for Collaboration
and Knowledge Management, explains the differences between data and information and how
context plays a role. For example, they declare that “data by itself has no context until someone
interprets it and ascribes a context to it” (Nunamaker 2001, p.4). Data is referred to as raw
symbols that do not have any significance. Raw data just exists absolutely by itself. It does not
have any context. Thus, it has no meaning by itself and no relevant relations to other data. In
order to assign meaning to data, we need to provide context by associating the data to other
pieces of data. Another article, Enterprise Knowledge Infrastructures, by Maier, Hadrich, and
Peinl, describes data in a similar way. Data is referred to as symbols in a basic description. For
instance, data can be things like numbers, alphanumeric characters, and figures. They clarify
that “data can be recorded, classified, and stored, but are not organized to convey any specific
describe information as having “some embedded context within it by the nature of the relations
that summarize and organize the data that comprise it” (Nunamaker 2001, p.4). Information is
made up of organized data. With information, data is structured with meaningful relationships.
51
Meaning is given to the data by having relational connections. The functional usefulness is what
distinguishes data from information. Data is not as useful as information. They give an analogy of
a relational database and how it takes raw data from storage and builds it into relevant information.
Maier, Hadrich, and Peinl discuss how most definitions distinguish data from information through
meaning and semantics. For instance, several definitions express information as “data that have
been organized so that they have meaning and value to the recipient” (Maier, 2009, p.4-5). And,
they explain how information depends on the context of the interpreted signals.
Examples of Data vs Information: Here are a few scenarios to help explain the
transformation of data into relevant information. For example, the number “2” absolutely by
itself is quite meaningless. What are we talking about when we say “2”? Two of what unit?
When we add context to the data and relate it to other data, then it transforms into relevant
information. For example, that number “2” turns into information with more relevance when we
add units and compare with other data in a ratio, like “2 miles / 1 hour”. Now, we have more
meaningful information. We have a valuation for speed, where we are taking a distance
For another scenario, we will compare the strengths of two individuals. We can do so in
absolute terms or in relative terms. Let’s say Person A weighs 100 pounds and can lift 200
pounds and person B weighs 300 pounds can lift 300 pounds. Who is stronger? In an absolute
sense, person B can lift 100 more pounds than person A. Therefore person B would be
considered stronger. However, if we took into account how much each person weighs, in a
pound-for-pound ratio, then we would come out with a different result. Person A can lift 2 times
their body mass. Whereas, person B can only lift their own body mass. From a relative pound-
52
Putting numbers into context is important for valuating statistics and scores in video
games. For example, some games like Halo have a multiplayer mode called team slayer. The
objective for winning this particular game is to be the first team to reach a certain number of
kills. In this scenario, the first team to reach 50 kills ends up winning the game. The following
report demonstrates the results of the game. As we can see, the Blue team ended up winning
over the Red team. Blue team reached a total of 50 kills and Red team only scored 30 kills. The
next section of the report shows how each player scored individually. Players 1 and 2 represent
the Blue team and players 3 and 4 were on the Red team. The players are ranked according to
their score. In this case, each individual score is calculated by the absolute number of kills the
player achieved. Using this method, players 1 and 2 scored the highest. And, player 4 scored the
least number of kills. However, these absolute valuations can be very misleading. Is player 4
Blue Team 50
Read Team 30
Figure 4.18: Postgame report using absolute scoring. Each individual score is calculated by the absolute number
of kills the player achieved. Using this method, player 4 performed with the lowest score.
53
If we alter the calculation of the individual score values to be more relative. This will
completely change the player rankings. For instance, in the second report below, the score
evaluation has been changed. If we put the number of kills into context with the number of
deaths, then we will have a better representation for the performance valuation of the players.
Here, each player score is calculated in a kill-death ratio with the number of kills divided by the
number of deaths. Using this relative perspective, we can see that player 4 is actually the best
performer. And, player 3 scored the lowest. Player 3 may have really good offensive skills.
But, his or her defensive skills are horrible! Player 3 had several more deaths than kills.
Perhaps, the Red team could have won if player 3 did not die so much.
Blue Team 50
Read Team 30
Figure 4.19: Postgame report using relative scoring. Each individual score is calculated using the number of kills
divided by the number of deaths. Using this method, player 4 performed with the highest score.
54
CHAPTER 5 - MEMORY:
ATTENTION, RETENTION, ALTERATION, AND FORGETTING
This section will explore an additional area in which relativism can help in artificial
intelligence. Relativism can be useful for memory. For instance, relative deviances and
relations can be useful for attention, retention, memory alteration, and forgetting.
Attention: Relative deviances and relations can be valuable for concepts like attention
and fear. Sensory inputs with a strong deviance can attract greater attention. For example,
imagine the event of a firecracker exploding in a setting of a silent room like a library. The
bright flash and loud sound of the firecracker will have a relatively high deviance compared to
the silent background. The stronger the deviance of the sensory input, the higher the attention.
However, now imagine the same firecracker exploding in a venue with a large fireworks show.
In this scenario, the firecracker will have a low deviance relatively compared to the noisy
background of several large fireworks going off. The lower the deviance of the sensory input,
Applying attention for artificial intelligence can be beneficial for filtering, extraction,
compression, and taking on multiple tasks. Pashler’s book, Attention, describes how important
the process of attention is for us. For instance, our senses are continually overwhelmed with
information from our external environment. Yet, merely a small fraction of the information is
considered relevant to us. Our thoughts and behavior need to process the information in a
selective manner. Our brains are presented with several inputs from different sources of sensory
receptors. We perceive visual inputs, auditory inputs, tactile inputs, and a range of tastes and
smells. The majority of this input is irrelevant and we only have a limited amount of processing
55
resources available at a given time. To process these inputs efficiently and effectively, we are
able to focus on a small set of the relevant information while ignoring the rest. Selecting only
the relevant regions can help simplify object recognition. (Pashler 1998).
The book From Human Attention to Computational Attention, by Mancas, Ferrera, Riche,
and Taylor, also addresses the importance of prioritizing and selecting relevant information.
They use the term saliency. Meaning, how much an object or a thing stands out relative to other
objects in a context. They discuss how attention is useful for perception and could be a step
towards conscious awareness. Attention is useful for information reduction. Our brains are
provided with massive amounts of sensory data from multiple sources. Attention allows our
brain to filter out irrelevant data and helps to prioritize multiple tasks. Like our brains, machines
have limited computational resources and memory. Applying these attention techniques to
machines has several benefits. It can help them perform faster, increase efficiency in memory
storage, achieve better abnormality detection, help information reduction, and prioritize
information. With attention, machines can gain the capabilities of being curious and bored.
The book mentions how features like brightness, color, and motion can be extracted from
the input signals. Then, the machine will look for deviations within a context. Calculating the
saliency is useful for machine learning. Being able to limit resources and apply attention to a
feature of interest can improve detection and decrease response times. Like humans, machines
have a limited quantity of resources to process through large amounts of sensory inputs. Having
attention similar to humans can help machines focus on similar things as we do. And, it can
improve their interactions and communication with people. (Mancas, Ferrera, Riche, and Taylor,
2016).
56
Retention: Machines will need to retain and consolidate relevant information for recall.
Applying relative retention techniques can be beneficial for compression, learning multiple tasks,
and adapting to changing environments. To accomplish this, memory could be constructed with
The following graphs are simplified versions of a semantic network used to show how
memory can be represented with relational deviances. The graphs contain nodes and weighted
connections between the nodes. In the first example, we can see a strong connection between the
Fire-engine node and the Red node. There is a high deviance. The Red node has a stronger
Figure 5.1: Semantic network illustrating how memory can be represented using relational deviances. The relation
between the Fire-engine node and the Red node has a higher weight compared to the other associations.
57 relative to the other nodes connected to the Fire-engine.
The Red node represents a stronger deviation
In the second example, more nodes are now associated with the Red node. The Red node
now shares several relationships. As the number of connections to the Red node increase, this
can decrease the deviance between Red and the Fire-engine. Also note, there is a banana on this
graph. For some reason, a connection between a Banana node and Red was made. This was a
bad association. But, we will address this scenario in a later example when we discuss the
Figure 5.2: Semantic network illustrating memory. Increasing the number of relations to the Red node can
decrease the weighted deviance between the Red node and the Fire-engine node.
58
The third example demonstrates how a deviant event of one node may influence the
connection weights of the other associated nodes. For example, say we experienced a very loud
event of a red balloon exploding. With the intense sound, the Explosion node will have a strong
connection to the Balloon node. And, this event can increase the connection weight to the
Figure 5.3: Semantic network illustrating memory. The addition of the Explosion node can influence the weights
between the other associated nodes. The strong weight between the Explosion node and the Balloon
node can propagate to the associated Red node.
59
The concept of relativism is useful in understanding memory retention. In your memory,
you may be connecting nodes to other nodes that are relevant to you. However, your particular
set of connections may not be relevant to other individuals. Not everyone may share the exact
same set of experiences. They can have different experiences than you. Individuals may
experience and store memories differently even for the same event. For example, two people
may experience the same bank robbery event. However, each observer may be at different
locations and may only pay attention to details they find relevant from their perspective as the
robbery event unfolds. Thus, creating two relatively different memories to recall.
Figure 5.4: Fill-in-the-blank question demonstrating how different answers (a, b, or c) may be
correct within their own context. It can be difficult to establish one absolute answer.
60
All answers may be correct within their own context. Perhaps the banana option was
chosen because we wanted to satisfy the pattern of containing all objects classified with fruit.
Maybe the stop sign was chosen because we wanted to fulfill the pattern of having all objects
classified with the color red. Or, it could be that the soccer ball was picked in order to fill the
Relativism is also useful with learning mnemonics. These mnemonic techniques can aid
information that we already know. The memories with more deviant connections can be easier
to recall. Roediger’s article The Effectiveness of Four Mnemonics in Ordering Recall studied
how particular mnemonics can help to improve recall over standard memorization techniques.
Being able to connect terms to images can aid in memory recall. (Roediger, 1980). There are
several different mnemonics. For example, there is the general keyword method where you
connect a term to a strong visual object. And, there is the loci method, also known as the
memory or mind palace method. With this technique, terms are linked to strong visual objects
Memory Alteration and Forgetting: The process of memory alteration and forgetting is
very important. Past memories can be influenced with new information. Machines will need to
be able to forget irrelevant things in order to make room for learning new tasks. The following
example depicts how deviance can be used to decrease the weighted relationship between weaker
connections. A new increased connection between deviant nodes can cause a decreased
connection between other, less deviant, nodes. For instance, the connection between Red and
Balloon may have increased. But, this increase can cause the weaker nodes connected to Red to
61
decrease. This can cause an already weak connection between Red and Banana to decrease
further. In essence, we are forgetting the association between Red and Banana.
Figure 5.5: Semantic network illustrating memory. An increased connection between deviant nodes can cause a
decreased connection between other, less deviant, nodes. The strong weight between the Balloon
node and the Red node increased. However, the weaker connection between the Red node and the
Banana node decreased.
62
CHAPTER 6 - FUTURE WORK
63
CHAPTER 7 - CONCLUSION
In conclusion, here is a recap of what the thesis covered. First, we addressed some of
the main issues for general artificial intelligence. Second, an overview was given for
absolutism and relativism. We introduced some examples of how these concepts relate to
machine learning algorithms. Then, we argued the thesis statement that classification requires
relativism to be useful. The concept of relativity can be useful for: (1) axiology, (2) defining
The goal of this thesis is to show how relativism can be used as a strategy to help solve
some of the challenges for general artificial intelligence. There are cases in the domain of
artificial intelligence where the method of relativism can be useful. Some cases are
straightforward while others are not as clear. We started with the more established axiology
examples of ethics and aesthetics in which relativism can be useful. And, then worked our way
to the other instances in which the method of relativism can be useful for defining things and can
64
REFERENCES
Battistella, E. (2005). “Bad Language: Are Some Words Better than Others?” Oxford
9780195172485.001.0001.
Brumfiel, G. (August 2009). “This Kilogram Has A Weight-Loss Problem.” National Public
Calder, T. (2015). "The Concept of Evil." The Stanford Encyclopedia of Philosophy. Edward
Artificial Intelligence: What are Progressive Neural Nets?" TechWorld. Article No:
3646730. <http://www.techworld.com/big-data/inside-deepminds-latest-attempts-
Emspak, J. (April 2013). "Speed of Light May Not Be Constant, Physicists Say." Live-Science.
<http://www.livescience.com/29111-speed-of-light-not-constant.html>. Accessed
August 2016.
65
Farmer, M. (2005). “Evidential Reasoning for Control of Smart Automotive Air-bag
Fildes, J. (November 2007). “Getting the Measure of a Kilogram.” BBC News. Science and
Findlay, M. (November 2012). “The Value of Art: Money, Power, Beauty.” Artnet News.
Freeland, C. (April 2002). “But Is It Art? An Introduction to Art Theory.” Oxford University
Press. Oxford, New York. Book. ISBN: 9780191593772. Call No: N71.F653 2002.
Grant, A. (February 2015). “Speed of Light Not So Constant After All.” Science-News
<https://www.sciencenews.org/article/speed-light-not-so-constant-after-all>. Accessed
August 2016.
Grant, D. (May 2011). “Is an Artist Only Appreciated After He Dies?” The Huffington Post.
<http://www.huffingtonpost.com/daniel-grant/is-an-artist-only-appreci_b_512932.html>.
66
Harper, D. (2016). “Online Etymology Dictionary: Bad.” Etymonline. Retrieved July 2016.
<http://www.etymonline.com/index.php?term=bad>.
Harris, K. (July 2014). “10 English Words that Mean Something Else in Other Languages.”
else-languages>.
Harris, W., Freudenrich, C. (July 2011). “How Light Works.” How Stuff Works: Science.
Mackie, J. (1955). "Evil and Omnipotence." Mind. Volume 64, Issue 254. P.200-212.
Maier, R., Hadrich, T., Peinl, R. (2009). “Enterprise Knowledge Infrastructures.” Dordrecht:
Mancas, M., Ferrera, V., Riche, N., Taylor, J. (2016). “From Human Attention to Computational
Attention: A Multidisciplinary Approach.” Springer New York. New York, NY. p. 1-3,
Nunamaker, J., Romano, N., Briggs, R. (2001). “A Framework for Collaboration and
67
Pashler, H. (1998). “Attention.” Hove: Psychology Press. East Sussex, UK. p.191, 223, 257.
DOI: 10.1037/0278-7393.6.5.558.
Experiments.” 3rd Ed. New York: McGraw-Hill. p.325-367, 536, 480. ISBN:
9780072991970.
“What is Metric? Making the First International Kilograms and Metres.” (2013). UK Metric
Zimbardo, P. (2008). “The Lucifer Effect: Understanding How Good People Turn Evil.”
Zimmermann, H. (2001). “Fuzzy Set Theory - And Its Applications.” 4th Ed. Springer Science
68
REFERENCES - IMAGES
“‘Composition VII’ by Theo van Doesburg.” (March 2008). [Image]. Retrieved November
Composition_VII_(the_three_graces).jpg>, <http://creativecommons.org/publicdomain>.
“‘Girl with a Pearl Earring’ by Vermeer.” (October 2014). [Image]. Retrieved November 2016
from: <https://commons.wikimedia.org/wiki/File:Meisje_met_de_parel.jpg>,
<http://creativecommons.org/publicdomain>.
Miranda, Carolina. (September 2014.) “'Levitated Mass,' Doug Pray's rock doc on
LACMA's epic boulder.” Los Angeles Times. Arts & Culture section.
<http://www.latimes.com/entertainment/arts/miranda/la-et-cam-levitated-mass-
documentary-michael-heizer-rock-lacma-20140903-column.html>.
“‘The Gleaners’ by Millet.” (July 2012). [Image]. Retrieved November 2016 from:
<https://commons.wikimedia.org/wiki/File:Jean-Fran%C3%A7ois_Millet_-_Gleaners_-
_Google_Art_Project_2.jpg>, <http://creativecommons.org/publicdomain>.
Unna, J. (July 2008). “‘David’ by Michelangelo.” [Image]. Retrieved November 2016 from:
<https://commons.wikimedia.org/wiki/File:'David'_by_Michelangelo_JBU0001.JPG>,
<https://creativecommons.org/licenses/by/3.0>.
69
“‘White Paintings’ by Robert Rauschenberg.” [Image]. Retrieved November 2016 from:
Bershidsky, Leonid. “Why Pay $15 Million for a White Canvas?” Bloomberg.
levitated-mass-documentary-michael-heizer-rock-lacma-20140903-column.html>.
70