See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2265039

Fuzzy Logic in Artificial Intelligence

Chapter · July 1997

Source: CiteSeer

CITATIONS READS
37 32,005

2 authors:

Erich Peter Klement Wolfgang Slany

Johannes Kepler University of Linz Graz University of Technology
253 PUBLICATIONS 8,854 CITATIONS 193 PUBLICATIONS 2,121 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Erich Peter Klement on 08 September 2014.

The user has requested enhancement of the downloaded file.

Fuzzy Logic in Arti cial Intelligence
Peter Klement Wolfgang Slany?
?ChristianDoppler Laboratory for Expert Systems
E184/2, TU Wien, A-1040 Vienna, Austria, Europe
Phone: +43{1{58801{6141 Fax: +43{1{5055304
URL: http://www.dbai.tuwien.ac.at:8080/sta /slany.html
E-Mail: [email protected]

CD{Technical Report 94/67

Labor

Expertensysteme

Christian Doppler Laboratory

for Expert Systems
Technische Universitat Wien
Institut fur Informationssysteme
Abteilung fur Datenbanken und Expertensysteme
 Fuzzy Logic in Arti cial Intelligence
Erich Peter Klement Wolfgang Slany
Fuzzy Logic Laboratorium Linz Christian Doppler Laboratory for Expert Systems
Johannes Kepler Universitat Technische Universitat Wien
A-4040 Linz, Austria A-1040 Vienna, Austria
E-mail: [email protected] E-mail: [email protected]

June 23, 1994

Abstract
After a basic introduction of fuzzy logic, we discuss its role in
arti cial and computational intelligence. Then we present innovative
applications of fuzzy logic, focusing on fuzzy expert systems, with one
typical example explored in some detail. The article concludes with
suggestions how arti cial intelligence and fuzzy logic can bene t from
each other.

I. Introduction
In 1948, Alan Turing wrote a paper [1] marking the begin of a new era,
the era of the intelligent machine, which raised questions that still remain
unanswered today. This era was heavily in uenced by the appearance of the
computer, a machine that allowed humans to automate their way of thinking.
However, human thinking is not exact. If you had to park your car pre-
cisely in one place, you would have extreme diculties. To allow computers
to really mimic the way humans think, the theories of fuzzy sets and fuzzy
logic were created. They should be viewed as formal mathematical theories
for the representation of uncertainty, which is essential for the management
of real world systems, as it mimics the crucial ability of the human mind to
summarize data and focus on decision relevant information.
1
 Marvin Minsky, one of the founding fathers of arti cial intelligence, once
de ned the latter as
\... the science of making machines do things that would require
intelligence if done by men."
Similarly, Lot A. Zadeh, who in 1965 wrote the founding paper on fuzzy set
theory [2], once described the aim of this theory as being
\the construction of smarter machines."
Zadeh recently coined the term MIQ (machine intelligence quotient) to re-
fer to this particular aspect of the growing number of intelligent consumer
products and industrial systems [3].
Proponents of the so-called `strong' arti cial intelligence believe that
eventually, these machines will be as intelligent as we human beings are
now. Thinking positively about technology, everything that is conceivable
to be solved by arti cial means will eventually be realized if it is interesting
enough. Of course some intellectual processes have been shown to be emer-
gent properties, such as `consciousness'. The concept of emergent properties
of complex systems was rst observed by von Bertalan y [4] in the 1920s in
his study of complex biological systems. He noticed that complex assemblies
of entities organized in particular ways can reveal unique properties not pos-
sessed by the individual entities alone. Emergent properties cease to exist if
the whole is broken into components or if the components are organized in
a di erent way. Additionally, emergent properties cannot be understood by
the study of isolated components. Similar to the notion of a critical mass in
physics, an emergent property will suddenly pop up when a sucient amount
of mass has been accumulated. Contrary to reductionistic approaches, these
approaches normally assume a holistic view of the world, i.e. something com-
plex can be more than simply the accumulation or `sum of its parts'. Of
course, as with the atomic bomb, which was in a certain sense the rst ar-
ti cial application of the physical e ect described above, the ethical aspects
have to be carefully considered. One has to be aware that any technology
can be used for good or for evil. However, not the technology in itself is good
or bad, but instead the humans that use it are so, since technology has so far
been only a tool for human beings. In the case of intelligence, this might be
not true anymore, since advanced intelligence may entail new ethical needs,
2
but these new forms of intelligence have not yet reached a level where ethical
aspects become prevalent.

II. Some elements of fuzzy logic

In this section, we brie y outline how fuzzy logic extends classical Boolean
logic (or, equivalently, how fuzzy set theory generalizes Cantorian set theory).
Given a (crisp) universe of discourse X , a fuzzy subset A of X (see, e.g.,
[2, 5]) is characterized by its membership function
A : X ! [0; 1];
where for x 2 X the number A (x) is interpreted as the degree of membership
of x in the fuzzy set A or, equivalently, as the truth value of the statement
`x is element of A'.
The membership function of a fuzzy set is a natural generalization of the
characteristic function of a (classical) subset A of X ,
1A : X ! f0; 1g;
assigning to each element x in X the value 1 whenever x belongs to A, and
the value 0 otherwise.
In order to generalize the set-theoretical operations like intersection and
union (or the corresponding Boolean logical operations conjunction and dis-
junction, respectively), we need triangular norms and conorms [6, 7, 8]: A
triangular norm (t-norm ) is a binary operation on [0; 1], i.e., a function
T : [0; 1]2 ! [0; 1], which is commutative, associative, monotone in both
components, and satis es the boundary condition
T (x; 1) = x:
If T is a t-norm, then the dual triangular conorm (t-conorm ) S : [0; 1]2 !
[0; 1] is de ned by
S (x; y) = 1 ? T (1 ? x; 1 ? y):
There are many, in fact in nitely many, t-norms and t-conorms, only few
of which are used in applications. The most important t-norms, together
with their dual t-conorms, are the following:
3
 Minimum TM, Maximum SM
TM(x; y) = min(x; y); SM(x; y) = max(x; y);
Product TP, Probabilistic Sum SP
TP(x; y) = x
y; SP(x; y) = x + y ? x
y;
Lukasiewicz t-norm TL, Bounded Sum SL
TL(x; y) = max(x + y ? 1; 0); SL (x; y) = min(x + y; 1):
Weakest t-norm TW, Strongest t-conorm SW
(
T (x; y) = min(x; y) if max(x; y) = 1;
W 0 otherwise,
(
SW (x; y) = 1max(x; y) ifotherwise.
min(x; y) = 0;

Given a t-norm T , its dual t-conorm S , and fuzzy subsets A, B of the

universe X , the membership functions of the intersection A \ B , the union
A [ B , and the complement Ac are given by:
A\B (x) = T (A (x); B (x));
A[B (x) = S (A(x); B (x));
A (x) = 1 ? A (x):
c

The values A\B (x), A[B (x), and A (x) describe the truth values of the
c

statements `x is element of A AND x is element of B ', `x is element of A

OR x is element of B ', and `x is NOT element of A', respectively.
Given a t-norm T , its dual t-conorm S and propositions P , Q with truth
values jjP jj and jjQjj, respectively, there are two main extensions of the
Boolean implication P ) Q:
S -implication
jjP )S Qjj = S (1 ? jjP jj; jjQjj);
R-implication
jjP )R Qjj = supf 2 [0; 1] j T ( ; jjP jj)  jjQjjg;
4
 In Boolean logic, i.e., with truth values 0 and 1 only, S - and R-implication
always are equivalent, which is no longer true in fuzzy logic: for the t-norm
TM the corresponding S -implication becomes the Kleene-Dienes implication
and the R-implication is the Godel implication [9] which are quite di erent.
Note, however, that in the case of the Lukasiewicz t-norm TL S - and R-
implication coincide: they both yield the Lukasiewicz implication.
The fact that there is a wide range of possibilities for extending Boolean
operations to fuzzy logic may look disturbing for beginners in the eld. How-
ever, it re ects the richness of fuzzy logic, as it allows for a very sensitive ne
tuning when modeling real world situations.

III. Fuzzy logic, arti cial intelligence, and

computational intelligence
It is important to note that the term fuzzy logic is used in two distinct senses.
In its narrower sense, fuzzy logic is only one branch of fuzzy set theory. Fuzzy
set theory was invented by Zadeh to be able to better represent such everyday
notions as the set of `tall persons'. Of course, this set is de ned vaguely, and
persons will more or less be a member of it, i.e. member to a certain degree.
Fuzzy logic in this narrow sense deals in a natural way with the representation
and inference from such vaguely formulated or uncertain knowledge, similarly
to classical logic which deals with crisp knowledge where statements can only
be either true or false (well, almost, at least if you do not count the ndings of
Kurt Godel). In recent years, however, it has become increasingly common to
employ the term fuzzy logic in a much broader sense, making the di erence
between the notions of fuzzy set theory and fuzzy logic vanish. To avoid
confusion, we follow the trend to use fuzzy logic in its general sense.
James Bezdek, editor in chief of the IEEE transactions on fuzzy systems,
de ned fuzzy logic in a delightful essay [10] to be one part of `computational
intelligence', altogether with such research areas as neural networks, evolu-
tionary computation, and genetic algorithms. Bezdek contrasts the ABC's on
intelligence: arti cial, biological and computational. In the strictest sense,
computational intelligence \depends on numerical data supplied by manufac-
turers and [does] not rely on `knowledge'." Arti cial intelligence, on the other
hand, uses what Bezdek calls `knowledge tidbits'. Heuristically constructed

5
arti cial intelligence such as an expert system is an example. Practicing
knowledge engineers and neural smiths know the distinction is at times not
precise. Expert extraction of feature data for training a layered perceptron
certainly falls in the area of arti cial intelligence. Using these features to
train the layered perceptron is primarily computational. Fuzzy inference
engines crafted by experts fall into the de nition of arti cial intelligence.
Algorithmic tuning of the engine with raw data, however, is computational
intelligence.
Even though the boundary between computational intelligence and arti -
cial intelligence is not distinct, we can, making certain assumptions, monitor
the volume of research activity in each. Indeed, the separate identities of
computational intelligence and arti cial intelligence are con rmed by inspec-
tion of the recent volume of publishing and patent activity [11].
However, the term `Computational Intelligence' itself is not undisputed,
since it had already been widely used to mean arti cial intelligence before
it was rede ned by Bezdek, see for example the journal `Computational In-
telligence', published since 1985, the conference `Computational Intelligence'
taking place annually since 1988, and numerous other publications and or-
ganizations using the term in this traditional sense.
In both cases, arti cial intelligence as well as fuzzy logic, one tries in
some sense to imitate life in its problem-solving capability. The ways how
to achieve this goal are di erent in many respects, but there are also many
common points where the two elds overlap: Robert Marks [11] counted 4811
entries on fuzzy logic in the INSPEC data base from 1989 to 1993, containing
citations from over 4000 selected journals, books, conference proceedings and
technical reports | \22% of them [were] cross categorized in the expert
system category, and 12% with neural networks." Based on various `bean
countings', Marks concludes that the overlapping areas cover, depending on
the way to count, from 14% to 33%.
It should not be left untold that there has been a lot of scienti c antago-
nism between fuzzy logic and arti cial intelligence, and, accordingly, skeptics
on both sides exist and treat the other side with reservation, if not with open
hostility. There are many reasons for this, e.g. some critics of fuzzy logic
credit the word `fuzzy' for being too controversial and misleading in itself,
others maintain that anything that can be done with fuzzy logic and fuzzy
set theory can be done equally well with classical logic and probability the-

6
ory [12]1, and still others insist on denying fuzzy logic the status of a logic
itself [13]. Of course these claims were refuted [14, 15, or see discussions in
the archives of the news-groups mentioned later in this article]. Fuzzy logic
in its narrow sense is simply a logic of fuzziness, not a logic which itself is
fuzzy. Just as the laws of probability are not random, so the laws of fuzziness
are not vague.
On the other hand, critics of arti cial intelligence have observed that the
sometimes over-ambitious predictions made in the past did not come true.
Some even go as far as to deny that there has been even one successful ex-
pert system implemented that really became used. Others believe that the
aim to create arti cial intelligence is useless and impossible on philosophical
grounds. However, such views are likely to become muted with the passage
of time and a better understanding of the basic ideas underlying the theories
of both arti cial intelligence and fuzzy logic. We observe nevertheless that,
nurtured by the current success of fuzzy logic in the real world, dangerously
unrealistic predictions and claims appear again. Bart Kosko, a respected
scholar in the eld and author of a best-selling textbook on `Neural Net-
works and Fuzzy Systems' [16] for instance predicts for the next few decades
fuzzy logic based natural language understanding, machines that write inter-
esting novels and screenplays in a selected style such as Hemingway's, or even
sex robots with a humanlike repertoire of behavior [14]. Some researchers
suggest however that as attempts are made to make fuzzy systems larger,
they will encounter similar diculties as conventional reasoning methodolo-
gies. Fuzzy logic is certainly not a philosopher's stone solving all problems
that confront us today. But it has a considerable potential for practical ap-
plications. The management of uncertainty will be of growing importance.
This uncertainty can have various reasons, ranging from uncertainty due to
the lack of knowledge or evidence, due to an abundance of complexity and
information, to uncertainty due to the fast and unpredictable development
of scienti c, political, social, and other structures nowadays.
1But Cheeseman, the author of [12], also rejects nonmonotonic reasoning, default logic,
and Dempster Shafer's theory, arguing that probabilities are better suited to model the
world, and that the other methods are at most harmless if not outright wrong.

7
IV. Applications of fuzzy logic
The applications of fuzzy technologies fall mainly into two categories: fuzzy
control applications, which are often rather simple but very ecient fuzzy
rule-based systems, such as autofocusing systems in cameras, washing ma-
chines, automobile transmissions, subway control, or even handwriting recog-
nition. In these applications, fuzzy logic is used as a powerful knowledge
representation technique that allows to hide unessential details and to han-
dle uncertain data. However, their eciency depends also heavily on the
use of sensors and e ectors, thus their success should really be explained by
the interaction of these various parts. The second category consists of those
much more complex systems that aim at supporting or even replacing a hu-
man expert. Such applications are exempli ed by medical diagnosis systems,
securities funds and portfolio selection systems, trac control systems, fuzzy
expert systems, and fuzzy scheduling systems. In this second category, there
are still many problems that remain to be addressed, and there is an equally
pressing need for a better understanding of how to deal with knowledge-based
systems in which knowledge is both uncertain and imprecise.
Areas where fuzzy logic and arti cial intelligence meet in current research
include: fuzzy expert systems (e.g., for medical diagnosis or intelligent tutor-
ing systems), theoretical investigations (e.g., combinations of fuzzy logic with
modal logics and other forms of defeasible reasoning, i.e. based on question-
able knowledge; this also includes investigations into fuzzy logic programming
languages such as fuzzy extensions of PROLOG), machine learning (e.g.,
combinations of fuzzy logic with neural networks, genetic algorithms, asso-
ciative memories, symbolic learning methods such as case based reasoning),
robotics (involving motion control and planning capabilities, e.g. when ying
a fully automated helicopter or driving a car on a freeway), pattern matching
(e.g., face recognition), fuzzy deductive databases (e.g., to ease data retrieval
in geographic information systems), or constraint satisfaction problem solv-
ing methods (applied for example in manufacturing process scheduling [17],
or in bridge design).

8
V. Fuzzy expert systems
Let us take a closer look at fuzzy expert systems as the archetypical spin
o coming from the combination of techniques from fuzzy logic and arti cial
intelligence. Classical expert systems are computer programs that emulate
the reasoning of human experts or perform in an expert manner in a domain
for which no human expert exists. This could be due to a dangerous working
environment or simply because of a domain that is to large for one human
being. These expert systems typically reason with uncertain and imprecise
information, using various methods besides fuzzy logic to handle them. There
are many sources of imprecision and uncertainty. The knowledge that the
expert systems embody is often not exact, in the same way as a human's
knowledge is imperfect. Given facts or user-supplied information are also
often uncertain.
An expert system is typically made up of at least three parts: an inference
engine, a knowledge base, and a working memory. The inference engine uses
the domain knowledge together with acquired information about a problem to
provide an expert solution. The knowledge base contains the expert domain
knowledge for use in problem solving, very often in form of explicit facts and
IF-THEN rules.
A fuzzy expert system, usually, is an expert system that uses a collection
of fuzzy membership functions and rules to reason about data. The rules in
a fuzzy expert system are typically of a form similar to the following:
IF heat is low AND pressure is high THEN valve is closed
where `heat' and `pressure' are (linguistic) input variables, i.e., names for
known data values, `valve' is a (linguistic) output variable, i.e., a name for
a data value to be computed, low is one of the possible linguistic values of
the variable `heat' described by membership function of the corresponding
fuzzy set, high is a linguistic value of the variable `pressure', and closed is a
linguistic value of the variable `valve'. The antecedent (the rule's premise)
describes to what degree the rule applies, while the conclusion (the rule's
consequent) assigns a fuzzy set or, if defuzzi cation takes place, a crisp value
to each of the output variables. Most tools for working with fuzzy expert
systems allow for more than one conclusion per rule. The set of rules in a
fuzzy expert system is known as the rulebase or knowledge base.
The general inference process proceeds in three (or four) steps.
9
1. In the fuzzi cation step, the linguistic terms de ned through their asso-
ciated fuzzy membership functions are matched with the actual values
of the input variables, to determine the degree of truth for each rule's
premise.
2. In the inference step, the truth values for the premises are propagated to
the conclusion part of each rule. This results for each rule in one fuzzy
subset that is assigned to an output variable. Usually, only minimum
or product are used as inference methods. In minimum inferencing, the
output membership function is clipped o at the height corresponding
to the rule premise's computed degree of truth. In product inferenc-
ing, the output membership function is scaled by the rule premise's
computed degree of truth.
3. In the composition step, all fuzzy subsets assigned to a given output
variables are combined to form a single fuzzy subset for each output
variable. Also in this case, two ways of composition dominate in most
applications, namely, max and bounded sum. In max composition, the
combined output fuzzy set is constructed by taking the pointwise max-
imum over all of the membership functions of the fuzzy sets assigned to
the output variable by the inference rule. In bounded sum composition,
the combined output fuzzy set is constructed by taking the pointwise
sum (cut o at level 1 whenever it would be exceeded) of all the mem-
bership functions of the fuzzy sets assigned to the output variable by
the inference rule.
4. The optional defuzzi cation step is used when it is useful and/or nec-
essary to convert the output fuzzy set into a crisp value by choosing a
crisp number which is representative for the fuzzy output set. This is
almost always the case in technical control problems where the output
must be crisp, less often in expert systems where output fuzzy sets are
sometimes linguistically approximated. There are at least 30 di er-
ent defuzzi cation methods. Two of the more common techniques are
the center of gravity and center of maxima methods. In the center of
gravity method, the crisp value of the output variable is computed by
nding the variable value of the center of gravity of the membership
function for the fuzzy value. In the center of maxima method, the

10
 midpoint of the region where the fuzzy set assumes its maximum truth
value is chosen as the representative crisp value for the output variable.
It should be mentioned that min and product are two out of an in nite
number of possible multiple-valued generalizations of the logical operation
`and' (conjunction) in Boolean (two-valued) logic.

VI. A typical example

To cite one of the most prominent and successful fuzzy expert systems, we
have to refer to a very long ranging project initiated as early as 1976 by
Klaus-Peter Adlassnig and resulting in a system in use today. `CADIAG-2',
which is currently evolving to become `CADIAG-3', is a medical diagnosis
system based on fuzzy expert system technology ([3] contains a recent paper
about this very large project which has resulted in an enormous amount of
publications; [5] also contains a description of CADIAG-2). A typical rule
of this system looks as follows (the rule has been slightly simpli ed for this
example):
IF fever is frequent AND
high fever is frequent AND
knee dropsy is rare AND
carditis is very-rare AND
articular pain is almost-always AND
erythema is frequent AND
previous tonsillitis is very-frequent AND
staphylokokkus is never AND
increased AST is almost-always
THEN rheumatic fever is plausible
Here one can see again the two key concepts which play a central role in the
application of fuzzy logic in expert systems. The rst is that of a linguistic
variable such as `high fever', that is, a variable whose values are words or sen-
tences in a natural or synthetic language such as frequent or rare. The other
is that of a fuzzy IF-THEN rule in which the antecedent and consequent are
propositions containing linguistic variables. The essential function served by
linguistic variables is that of granulation of variables and their dependencies.
11
In e ect, the use of linguistic variables and fuzzy IF-THEN rules results |
through granulation | in soft data compression which exploits the tolerance
for imprecision and uncertainty. Of course, the e ective membership func-
tions represented by words such as very-rare have also to be determined and
must be known at inference time to the inference engine.
For a detailed account of what expert systems in general and fuzzy expert
systems in particular are and how they work, we refer to [18, 19, 5].

VII. Can fuzzy logic help to reduce compu-

tational complexity?
In the nal `Future perspectives' section of his book [5], Hans-Jurgen Zim-
mermann, who is also editor in chief of the in uential journal Fuzzy Sets and
Systems, points out an important potential bene t that fuzzy logic could
bring to the eld of arti cial intelligence. In particular, for many problems
encountered in sub elds of arti cial intelligence, such as logic programming,
deductive databases, or constraint satisfaction, complexity analysis results
range from intractable to undecidable, i.e. these problems are computation-
ally very hard to solve. It is very easy to encounter problems not solvable
with current computing power. Zimmermann therefore raises the question
whether fuzzy logic can help to solve such large and complex problems. At
the time of the writing of his book, the answer was still `no'.
We shortly restate here some common notions from complexity theory
to analyze whether another answer can be found. Many of the complex
problems most commonly encountered in the mentioned sub elds of arti cial
intelligence have been proven to be so-called NP hard problems, referring to
their combinatorially explosive need for computing resources. Especially the
run-time behavior of algorithms that solve these hard problems is at least
exponential in the size of the problem description. NP hardness formally
implies that these problems are `at least as hard' as so-called NP complete
problems. According to Garey and Johnson [20], NP complete problems are
known to be
\: : : `just as hard' as a large number of other problems that are
widely recognized as being dicult and that have been confounding
the experts for years.

12
 : : : the knowledge that [a problem] is NP complete does provide
valuable information about what lines of approach have the poten-
tial of being most productive. Certainly the search for an ecient,
exact algorithm should be accorded low priority."
The main result is that an exact and ecient solution for NP hard problems
has eluded many researchers until now. These problems have therefore been
termed intractable. It is however possible to relax one of two criteria, either
exactness or eciency, in which case the other criterion can be ful lled in
many cases. One suggestion could be to relax the problem somewhat in its
unimportant characteristics, i.e. to model a simpli ed version that does yield
an acceptable solution eciently. This is often sucient for real world opti-
mization problems. Another suggestion is indicated by the second sentence
in above quotation, which is worth some more investigation. In particular, it
is interesting to know that NP complete problems can be solved in polyno-
mial time by a nondeterministic computer. The scenario is often such that a
solution has to be `guessed', for instance by consulting an `oracle', followed
by calling a polynomial algorithm to check whether the guessed solution is
correct. This would suggest that an algorithm that intelligently `guesses' a
complete instantiation and then checks whether it is a solution could be used
to construct an algorithm that nds a `reasonably good' solution for `almost
all' problems.
The following paragraph provides a little more background about the
introduced notions. It is an open problem of complexity theory whether
NP is equal to P, P being the problems solvable in polynomial time, i.e.
the tractable problems. Indeed, many researchers even believe that this
question is currently to be the most important open question in computer
science as a whole. However, most researchers think that P and NP are
di erent. This would mean that NP complete problems would remain, at
least in the worst case, intractable, i.e. their execution time grows more than
polynomially when the structural parameters grow linearly. In such a case,
doubling the speed of the computer does not really help since only negligible
larger problems will be solvable by that computer, which is usually by far
not enough. The NP complete problems are characterized by the fact that
they are NP problems and that all other NP problems can be polynomially
reduced to these NP complete problems. This means that NP complete
problems are at least as dicult as any other NP problem. To prove that a

13
problem A is NP complete, it is sucient to show that A belongs to NP and
that one other problem B known to be NP complete can be polynomially
transformed into A. A general search problem H belongs to the NP hard
problems if and only if there exists a polynomial time algorithm for some
decision problem C known to be NP complete, assuming that H could be
used arbitrarily often for further computations at a computational cost of
one unit-time interval by the polynomial time algorithm solving C (i.e., C
must be polynomial time Turing-transformable into H ).
Therefore, the answer to the question posed by Zimmermann has to re-
main negative on theoretical grounds. On the contrary, the computational
complexity will often increase when a problem is fuzzi ed, since in general
the search space will be enlarged.
However, knowledge modeled using fuzzy logic can easily approximate
a problem in its important characteristics. In particular, problems encoun-
tered in the real world, such as in manufacturing, exhibit a tendency to obey
certain intangible rules, thus allowing the approximation to be done success-
fully. The same would not be possible for random data, representing the
worst case that must also be correctly handled by exact (but much slower)
algorithms. Additionally, some heuristics such as described in [17] match par-
ticularly well with fuzzy knowledge representation schemes, and thus help to
eciently solve large and complex problems. In this connection, a `solution'
to a problem is often understood pragmatically as the result of a search for
the best solution that can be found using all the available resources such as
available computers, available time, and available algorithms. So, to provide
a more positive answer to the question raised by Zimmermann, we conclude
that fuzzy logic can help to reduce complexity if the problem to be solved
can be adequately approximated, which is true for many real world problems.

VIII. Synergy e ects

To emphasize again in what respect arti cial intelligence and fuzzy logic can
mutually bene t from each other, we want to point out that all complex
systems and machines that where built so far required more than just one
basic technology in order to be successful. In a large measure, techniques from
arti cial intelligence and from fuzzy logic are complementary rather than
competitive. We believe that it is possible to fruitfully combine techniques

14
from both elds in many areas. The resulting hybrid systems will be more
and more important in the future. Following the line of reasoning given at
the begin concerning emergent properties, the synergy e ect resulting in this
combination is necessary to achieve the ultimate goal of creating machines
that act more and more intelligently for the bene t of mankind.

IX. Further readings :::

For readers interested in gaining a better understanding of one of the two

elds, fuzzy logic and arti cial intelligence, we would like to refer to some
good introductory texts such as Winston's book on arti cial intelligence [21],
or, more recently, McNeill and Freiberger's book on fuzzy logic [14]. For those
wanting to dig deeper or to answer more elaborate questions, we recommend
to consult some of the following texts and media (the list could of course be
much longer, but we limit ourselves to the most accessible items):
 The excellent `Encyclopedia of Arti cial Intelligence' edited by
Shapiro [19] covers almost all possible subjects related to this eld,
including numerous articles on fuzzy logic.
 The `Readings in Fuzzy Sets for Intelligent Systems' [22] to rapidly
nd the most in uencing articles published in this eld, as well as the
`Selected Papers by L. A. Zadeh' [23].
 The internet news-groups comp.ai and comp.ai.fuzzy, also accessible
electronically via various mailing lists and blackboards, including their
respective frequently-asked-questions (with answers) lists, which con-
tain pointers to other electronic sources of information such as world-
wide-web-servers, pointers to the most important conferences, major
journals, scienti c societies, research centers, major scienti c projects,
book-lists, as well as names of persons-to-know and companies related
to the respective elds. These news-groups are also forums to discuss
all topics related to the two elds, and are equipped with searchable
archives extending over several years [15, 24].
For readers searching references covering primarily the intersection of ar-
ti cial intelligence and fuzzy logic, we have compiled a list of some important
15
textbooks [25, 18, 26, 27, 28] and conference proceedings [29, 3] in the bibli-
ography.

References
[1] Alan M. Turing. Intelligent machinery. In D. C. Ince, editor, Mechanical
Intelligence, Collected Works of A. M. Turing. North-Holland, 1992.
Original paper appeared in B. Meltzer and D. Michie (Editors), Machine
Intelligence, 5:3{23, 1969, Edinburgh University Press, but was actually
written as early as in 1948.
[2] Lot A. Zadeh. Fuzzy sets. Information and Control, New York: Aca-
demic Press., 8:338{353, 1965. Republished in [23].
[3] Erich Peter Klement and Wolfgang Slany, editors. Fuzzy Logic in
Arti cial Intelligence. Proceedings of the 8th Austrian Arti cial Intel-
ligence Conference, FLAI'93, Linz, Austria, June 1993, volume 695
of Lecture Notes in Arti cial Intelligence. Springer Verlag Berlin Hei-
delberg, 1993. A conference report is available on the net: URL:
ftp://mira.dbai.tuwien.ac.at/pub/slany/ ai.txt.
[4] L. von Bertalan y. The organism considered as a physical system. In
L. von Bertalan y, editor, General system theory. Braziller, New York,
1968. Republished from 1940.
[5] Hans-Jurgen Zimmermann. Fuzzy Set Theory | and Its Applications.
Kluwer Academic Publishers, 2nd, revised edition, 1991.
[6] Erich P. Klement. Operations on fuzzy sets and fuzzy numbers related
to triangular norms. In Proceedings of the Eleventh International Sym-
posium on Multiple-Valued Logic, pages 218{225, Norman, 1981. IEEE,
New York.
[7] B. Schweizer and A. Sklar. Probabilistic Metric Spaces. North-Holland,
Amsterdam, 1983.
[8] D. Butnariu and E. P. Klement. Triangular Norm-Based Measures and
Games with Fuzzy Coalitions. Kluwer, Dordrecht, 1993.

16
 [9] Rudolf Kruse, Jorg Gebhardt, and Frank Klawonn. Foundations of Fuzzy
Systems. John Wiley and Sons Ltd., Chichester, 1994.
[10] James C. Bezdek. On the relationship between neural networks, pattern
recognition and intelligence. The International Journal of Approximate
Reasoning, 6:85{107, 1992.
[11] Robert J. Marks II. Intelligence: Computational versus arti cial. IEEE
Transactions on Neural Nets, September 1993.
[12] Peter Cheeseman. In defense of probability. In Proceedings of the Ninth
International Joint Conference on Arti cial Intelligence, pages 1002{
1009, Los Angeles, Ca., August 1985. Morgan Kaufmann Publishers,
Inc.
[13] Charles Elkan. The paradoxical success of fuzzy logic. In Pro-
ceedings of the Eleventh National Conference on Arti cial Intelli-
gence, AAAI'93, pages 698{703, Washington D.C., July 1993. URL:
ftp://cs.ucsd.edu/pub/elkan/paradoxicalsuccess.ps.Z; Responses: URL:
ftp://ftp.cs.cmu.edu/user/ai/areas/fuzzy/doc/elkan/response.txt.
[14] Daniel McNeill and Paul Freiberger. Fuzzy Logic. Simon & Schuster,
1993.
[15] Erik Horstkotte, Cli Joslyn, and Mark Kantrowitz. comp.ai.fuzzy
faq: Fuzzy logic and fuzzy expert systems 1/1 [monthly post-
ing]. URL: ftp://ftp.cs.cmu.edu/afs/cs.cmu.edu/project/ai-repository/-
ai/pubs/faqs/fuzzy/fuzzy.faq, January 1994.
[16] Bart Kosko. Neural Networks and Fuzzy Systems. Prentice-Hall, Engle-
wood Cli s, NJ, 1992.
[17] Wolfgang Slany. Scheduling as a fuzzy multiple criteria opti-
mization problem. CD-Technical Report 94/62, Christian Doppler
Laboratory for Expert Systems, Technical University of Vi-
enna, 1994. Submitted to Fuzzy Sets and Systems. URL:
ftp://mira.dbai.tuwien.ac.at/pub/slany/cd-tr9462.ps.Z.
[18] A. Kandel, editor. Fuzzy Expert Systems. CRC Press, Boca Raton, Ca.,
1992.
17
 [19] Stuart C. Shapiro, editor. Encyclopedia of Arti cial Intelligence. John
Wiley & Sons, Inc., 2nd enlarged and revised edition, 1992.
[20] Michael R. Garey and David S. Johnson. Computers and Intractability:
A Guide to the Theory of NP-Completeness. Freeman and Co., 1979.
[21] Patrick Henry Winston. Arti cial Intelligence. Addison-Wesley Pub-
lishing Company, Inc., 1977.
[22] Didier Dubois, Henri Prade, and Ronald R. Yager, editors. Readings in
Fuzzy Sets for Intelligent Systems. Morgan Kaufmann, 1993.
[23] R. R. Yager, S. Ovchinnikov, R. M. Tong, and H. T. Nguyen, editors.
Fuzzy Sets and Applications: Selected Papers by L. A. Zadeh. John
Wiley & Sons, 1987.
[24] Mark Kantrowitz. comp.ai faqs: 6 parts [monthly posting]. URL: ftp://-
ftp.cs.cmu.edu/afs/cs.cmu.edu/project/ai-repository/ai/pubs/faqs/ai/ai [1-
6].faq and other directories at the same site, January 1994.
[25] Didier Dubois and Henri Prade. Possibility Theory: An Approach to
Computerized Processing of Uncertainty. Plenum Press, N.Y., 1988.
[26] R. Kruse, E. Schwecke, and J. Heinsohn. Uncertainty and Vagueness in
Knowledge Based Systems { Numerical Methods. Springer Verlag, 1991.
[27] Constantin V. Negoita. Expert Systems and Fuzzy Systems. Benja-
min/Cummings, 1985.
[28] Maria Zemankova-Leech and Abraham Kandel. Fuzzy relational data
bases { a key to expert systems. Verlag TU V Rheinland, 1984.
[29] D. Heckerman and A. Mamdani, editors. Uncertainty in Arti cial In-
telligence (Proceedings of the 9th Conference). Morgan Kaufmann Pub-
lishers, 1993.

18

View publication stats