Exploring Methods in Artificial Intelligence Development:

lovethreewayAI and Robotics

Oct 20, 2013 (4 years and 8 months ago)


Exploring Methods in Artificial Intelligence Development:

A Lit Review by Ivan Wolyniec

The development of an Artificially Intelligent computer program is, as one would imagine, quite


undertaking. Various methods have been historically employed in the development of the
Artificial Intelligence as we know it today. These range in complexity and age, concepts such as Neural
Networks being a relatively new idea at fifty years of age or Fu
zzy Logic, the underlying theories behind
which being several thousand years old.
It thus requires a comprehension and working, malleable
knowledge of these and many, many more concepts to fit together a working model of an Intelligence.

paramount conc
ept would have to be the
aforementioned Neural Network, which when
referred to in the field of computing rather than biology is known as an Artificial Neural Network. There
has been significant development and theorization in this field since its conceptio
n in 1943 by a pair of
scientists speculating about ho
w neurons


work together to give us the awareness of our
ings to which we are accustomed
. There are

categories of Neural Networks, the most
applicable to the subject at hand and thu
s our focus will be on three in particular, Networks for
Classification, Networks for Data Conceptualization, and Networks for Data Association



The first of which takes in data of various kinds (for this particular project it would b
e words)
and then sorts it into various categories (such as nouns, verbs, etcetera). The second variety of Network,
the Data Conceptualization one, is more straightforward than one is led to believe; it takes the data in and
applies it to different categor
ies (this may seem similar to the first network type, but the latter of the two
is used for response building).The third type of Network essentially checks the data that the program has
for consistency.

There are also a variety of methods that have been p
ublished upon for the express purpose of
honing these special networks and to optimize them such that they build upon one another to increase
For instance, in the eighties a concept was proposed for an Artificial Neural Network that
could recog
nize features about its environment and then selectively execute the code which would benefit
it most under those circumstances, skipping any code which was dubbed usel
ess under those
circumstances (Linsker). Most of the computation
s done by the Neural Net
work are

done i
n a manner
hidden from the user and

running many different pro
cesses and commands

via nodes

at one time, giving
an excellent transition into the next topic.

Expert S
are a good contrast to Neural Networks because the theory behind them is just so
fundamentally different.
The master system assumes that the model works like a kind of a ‘black box’,
that it is one chain process from input to output (Zahedi). It processes
data sequentially, going from one
step to another within one large sequence rather than dividing tasks across a network of mechanisms
designed to
handle smaller tasks in a parallel manner. Because of the computing limi
tations before the
nineties Artifi
cial Intelligences were generally limited to this variety on account of hardware
limitations and relatively simplistic architecture, and most of the early, truly ‘successful’ AIs were Expert
Systems. Generally, Master Systems are limited in their flexibili
ty, often consisting of a series of ‘if’
statements to check if a series of cases is true. This kind of software architecture has generally faded from
relevance due to its inability to evolve as a technology into something that can compete with the
sly discussed Neural Networking

and due to the degree of difficulty it requires to execute properly
in terms of natural language processing.

Fuzzy L
ogic is also a topic quite relevant to the undertaking of Artificial Intelligence
development that also happ
ens to have a fascinating historic precedent. Though proposed to Academia in
1964 by

Dr. Lofti Zadeh, the underlying theory behind Fuzzy Logic had been present since ancient times.
It is important to understand that Fuzzy Logic is inherently different from

probability, another way of
expressing uncertainty, in that it makes different assumptions about the way the world works. Probability
is based, like most of the world today, in the assumption that all facts are binary that was put forth by
Aristotle. That

is, every piece of knowledge is either exactly true or false. About two hundred years
beforehand, Buddha had proposed that a contrary philosophy; that everything contains within it a bit of its
opposite. That an object could be ‘O’ and ‘Not O’ at the same

time without being paradoxical.
logic is the result of that mindset, that there is a ‘truth factor’ that ranges from zero to one, and that it can
be used to convey partial truths, paradoxes, and generalizations (
). This has particularly large
applications to the development of an Artificial Intelligence because of the implications it has for Natural
Language Processing, or the system by which human languages are interpreted by machines. Humans
naturally speak in partial truths and generalizatio
ns, making it very difficult for a binary truth machine to
process. When one says they are ‘kind of’ hungry, the fact that they are hungry has a truth value leaning
towards one but not quite there. Because of the flexibility it affords the program,
Fuzzy L
ogic is
permeating the world of consumer electronics at a brisk pace (Klingenberg).

These are all methods which have long and varied histories in the field of Artificial Intelligence
and are all methods which must be considered for application in such a p
ending project. They all have
merit, and they all warrant investigation into the extremities of their usefulness.

Cited Sources:

Freeman, J. A. and Skapura, D. M. (1991).

Neural Networks: Algorithms, Applications, and
Programming Techniques
Wesley, Reading, MA.

Fatemeh Zahedi
, “
An Introduction to Neural Networks and a Comparison with Artificial Intelligence and
Expert Systems
Vol. 21, No. 2 (Mar.

Apr., 1991), pp. 25

Ralph Linsker, "Self
Organization in a
Perceptual Network,"

, vol. 21, no. 3, pp. 105
117, Mar.
1988, doi:10.1109/2.36

Zadeh, L. (1965):
Fuzzy sets.

Information and Control 8 (1965), 338

Bryan Klingenberg, “Getting Started with Fuzzy Logic”, Calvin College Engineering Department,