sw-1 (thinking on the web)

8/17/2019 SW-1 (Thinking on the Web)

1/264

Today, the world is experiencing the excitement of an historic change. We and

ourselves in the midst of an information revolution, the result of rapid advances

in technology built in great part upon the shoulders of three pivotal pioneers:

Kurt Go¨ del, lan Turing, and Tim !erners"#ee. Through their contributions,

we are witnessing the remar$able refashioning of the %nformation ge, which

began in the &'()s, into the %nformation *evolution as the World Wide Webevolves into a resource with intelligent capabilities.

The contributions of Go¨ del +what is decidable-, Turing +what is

machine intelligence-, and !erners"#ee +what is solvable on the Web- are

important milestones toward ust how much /intelligence0 can be proected onto

the Web.

While the capabilities and scope of today1s World Wide Web are impressive,

its continuing evolution into a resource with intelligent features and

capabilities presents many challenges. The traditional approach of building

information systems has consisted of custom"made, costly database

applications. 2owever, this is changing. %nformation services are beginning to

use generic components and open global standards to offer widely accessible

graphical presentations with easier interaction. s a result, bene3ts are accruing to

transactions over the Web including such areas as: e"commerce, ban$ing,

manufacturing, and education.

t the heart of the %nformation *evolution is the transformation of the world

toward a $nowledge economy with a $nowledge society. 2elping to forge this

transformation is the World Wide Web 4onsortium +W54-, which is wor$ing to

deliver global machine processing built upon layers of open mar$up languages.

The $ey 6uestion is: /2ow far can we go in enhancing the expressive ability of

the Web while still offering tractable solutions0

%n 7art % +what is Web %ntelligence-, we begin with a discussion of the

development of the %nformation ge and how the Web contributes information

services that the bene3t human productivity. Then, the contributions of Go¨ del

in 4hapter 8, Turing in 4hapter 5, and !erners"#ee in 4hapter 9, are

introduced and woven into a portrait of potential intelligent Web capabilities.

!oth abstract and practical 6uestions of intelligence, logic, and solvability

are explored in order to delineate the opportunities and challenges facing thedevelopment of Web capabilities. %n addition, we highlight some of the philo"

sophical issues that underpin the %nformation *evolution with a threaded series

of vignettes or interludes that are presented in between the chapters.

OVERVIEW

%t is widely accepted that the technology of today1s %nformation ge has had a

maor impact on global communications and commerce, and that it will continue

to support maor improvements in human productivity. 2owever, while the World

Wide Web is ma$ing signi3cant contributions to this progress, there remain manychallenges to its further development into a resource with intelligent features.

or the %nformation ge to achieve its full potential in improving human pro"

ductivity, at least two $ey new advances must still be achieved: +1- ubi6uitous

access to transaction applications of all types; and +2- intelligent software appli"

cations enabling automated transactions.

or example, Web


2/264

server framewor$s to successfully communicate complex business logic. The

solution of the W54 to both of these problems is to deliver automatic machine

processing globally through a Web architecture utili=ing layers of open mar$up

languages.

This chapter begins by highlighting what is meant by the concepts of

/thin$ ing0 and /intelligent applications0 on the Web. Then, the developmentof the %nformation ge and the emergence of the Web as an empowering

force for global change is presented. We discuss the forces behind the %nformation

*evolution that are transforming the world1s economic and social systems, and

producing the demand for intelligent features on the Web. >ext are presented

the limitations of today1s Web and the need for intelligent automatic

capabilities through the development of the


3/264

perform tas$s that are sufficiently complex and human"li$e that the term

intelligent may be appropriate. Whereas % can be seen as the science of

machines that behave intelligently +or simulate intelligent behavior-, the concept

of intelligent applications entails the efforts to ta$e advantage of % technologies

to enhance applications and ma$e them act in more intelligent ways.

This brings us to the 6uestion of Web intelligence or intelligent softwareapplications on the Web. The World Wide Web can be described as an

interconnected networ$ of networ$s, but that does not go 6uite far enough. The

present day Web consists not only of the interconnected networ$s, servers, and

clients, but also the multimedia hypertext representation of vast 6uantities of

information distributed over an immense global collection of electronic devices.

With software services being provided over the Web, one can readily see an

analogy to the human +or machine- thin$ing process where information is stored,

accessed, transferred, and processed by electronic patterns in electrical devices

and their interconnections.

2owever, the current Web consists primarily of static data representationsthat are designed for direct human access and use.


4/264

6 ED7FWE*%>G T2E %>F*DT%F> GE

%n &'9, mathematician lan Turing 3rst started to seriously explore the con"

cept of intelligent machines. 2e determined that a computing machine can be

called intelligent if it could deceive a human into believing that it was human. 2is

test H called the Turing Test H consists of a person as$ing a series of 6uestions to both a human subect and a machine. The 6uestioning is done

via a $ey" board so that the 6uestioner has no direct interaction with the subects;

human or machine. machine with true intelligence will pass the Turing Test by

providing responses that are suf3ciently human"li$e that the 6uestioner cannot

determine which responder is human and which is not. We will investigate Turing

and his contributions to the %nformation *evolution in 4hapter 5.

The inventor of the World Wide Web, Tim !erners"#ee, is also the originator

of the proposed next generation Web architecture, the


5/264

T2E %>F*DT%F> GE

around &)) .A., and the %ndustrial *evolution, which began around &)) and

is still continuing to spread across the underdeveloped world even today.

Ten thousand years ago, humans lived in migratory groups and with the

aid of Cexible, rapidly evolving cultures, these loosely organi=ed groups of /hunter gatherers0 were able to adapt to virtually all the climate =ones and

envi" ronmental niches on the planet, from the rctic to temperate =ones to the

tropics. They fed themselves by hunting, herding, 3shing, and foraging. The

essence of hunting and gathering economies was to exploit many resources

lightly rather than to depend heavily on only a few. othing, therefore, could

be allowed to interrupt the harvest. This is due to a very narrow window of opportunity for planting and cultivating. Bnder this $ind of pressure, agricultural

communities became more time"conscious. griculturalists also had to store the

produce of their 3elds for the rest of the year, protect it from moisture, vermin,

and thieves, and learn to distribute supplies so the community could survive and

still have seed for next year1s planting. These conditions created a new $ind of

life style.

While a hunter gather ac6uired resources from &)) acres to produce an

ade" 6uate food supply, a single farmer needed only & acre of land to produce the

e6uivalent amount of food. %t was this &))"fold improvement in land managementthat fueled the agricultural revolution. %t not only enabled far more ef3cient food

production, but also provided food resources well above the needs of subsistence,

resulting in a new era built on trade.

The gricultural *evolution crept slowly across villages and regions, intro"

ducing land cultivation and a new way of life. Auring the long millennia that

this revolution progressed, the world population was divided into two compet"

itive categories: primitive and civili=ed. The primitive tribes continued in the

mode of hunting gathering while the civili=ed communities wor$ed the

land. The civili=ed communities produced foodstuffs for their own use with asurplus to allow for trade.

!ecause farmers consumed what they produced directly and traded their sur"

plus locally, there was a close relationship between production and consump"

tion. 2owever, as trade developed the gricultural *evolution encouraged the


6/264

! ED7FWE*%>G T2E %>F*DT%F> GE

construction of the roads that facilitated the exchange of speciali=ed produce on

an expanding scale until it eventually become global.

This evolutionary transition to an agricultural basis for society was still incom"

plete when, by the end of the seventeenth century, the %ndustrial *evolutionunleashed a new global revolutionary force.


7/264

"T2E %>F*DT%F> GE

/concentration of wor$ers0 is being replaced by Cexible wor$ forces including

/telecommuters.0 nd most importantly, /concentration of stoc$piles0 is being

replaced by /ust"in"time0 inventory and reductions in planning uncertainty.

s a result, production and consumption are continuing to move further apart.or many years, the falling cost of information has shifted power from the hands

of the producers into the hands of the consumer. Even so, the cost of information

has generally changed very slowly. The evolution of information distribution from

writing to the printing press too$ thousands of years. 2owever, once moveable

type was developed, the transition rapidly accelerated. When signi3cant drops in

the cost of information occurred, as a result of the printing press, only certain

types of organi=ations survived. rom the ancient empires to the world1s industrial

giants, leaders have recogni=ed that information is power. 4ontrolling information

means $eeping power.%n fact, it was the high cost of information that made early civili=ations most

vulnerable. %f a temple was sac$ed, it meant the loss of all available $nowledge:

from when to plant crops to how to construct buildings. %nformation was expen"

sive to collect and maintain, and as empires rose and fell, the cost of information

remained high. Empires in 4hina, %ndia, and Europe all used large, expensive

bureaucracies to control information collection and dissemination.

The *oman Empire set the pace of communications by constructing (5,)))

miles of roads, thereby eliminating the traditional dependence on water trans"

portation. The Empire lasted for centuries and spread its administration acrossEurope, West sia, and >orth frica. 4ouriers traveled over *oman roads to the

furthest reaches of the Empire. *ome also moved the management of $nowledge

from the temples to libraries for civil administration and learning. !ut for access

to information resources, one still had to go to the libraries, which meant that

information had limited distribution.

The invention of the printing press enabled common people to gain access to

scienti3c $nowledge and political ideas. !y the sixteenth century, information

moved into the hands of the people and out of the strict control of the state. %n a

similar dramatic change, the invention of the telegraph produced the possibility

for instant widespread dissemination of information, thereby liberating economic

mar$ets.


8/264

# ED7FWE*%>G T2E %>F*DT%F> GE

it remains a strong candidate. %ndeed, service wor$ers today complete $nowl"

edge transactions many times faster through intelligent software using photons

over %7 switching, in comparison to cler$s using electrons over circuit switching

technology ust a few decades ago.!y the mid"twentieth century, the explosion of available information re6uired

greater information management and can be said to have initiated the %nformation

ge. s computer technology offered reduced information costs, it did more than

allow people to receive information. %ndividuals could buy, sell, and even create

their own information. 4heap, plentiful, easily accessible information became as

powerful an economic dynamic as land and energy.

The falling cost of information followed Doore1s law, which said that the price

performance of microprocessors doubled every &J months.


9/264

T2E WF*#A W%AE WE! ##

Fver the past () years, the %nternetMWeb has grown into the global %nformation


10/264

#% ED7FWE*%>G T2E %>F*DT%F> GE

The mid"&'J)s mar$ed a boom in the personal computer and superminicom"

puter industries. The combination of inexpensive des$top machines and powerful,

networ$"ready servers allowed many companies to oin the %nternet for the

3rst time.4orporations began to use the %nternet to communicate with each other and

with their customers. !y &''), the *7>ET was decommissioned, leaving only

the vast networ$"of"networ$s called the %nternet with over 5)),))) hosts.

The stage was set for the 3nal step to move beyond the %nternet, as three

maor events and forces converged, accelerating the development of informa"

tion technology. These three events were the introduction of the World Wide

Web, the widespread availability of the graphical browser, and the unleashing of

commerciali=ation.

%n startling contrast, F#, 4ompuuclear *esearch, 4onseil Europe@en pour la

*echerche >ucle@aire, +4E*>- in etscape >avigator. These browsers are powerful applications thatread the mar$up languages of the Web, display their contents and collect data.

The primary language for formatting Web pages is 2TD#. With 2TD# the

author describes what a page should loo$ li$e, what types of fonts to use, what

color the text should be, where paragraph mar$s come, and many more aspects

of the document. ll 2TD# documents are created by using tags. Tags have

beginning and ending identi3ers to communicate to the browser the beginning

and ending text formatted by the tag in 6uestion.

%n &''5, Darc ndreesen and a group of student programmers at >4ational 4enter for


11/264

#%D%TT%F>< F TFAO1< WE!

#&


12/264

#) ED7FWE*%>G T2E %>F*DT%F> GE

standard form. %n addition, some of today1s basic Web limitations include search,

database support, interoperable applications, intelligent business logic, automa"

tion, security, and trust. s a result, the %nformation *evolution awaits the next

brea$"through to fully open the information Cow.

THE NE*T GENERATION WEB

new Web architecture called the


13/264

4F>4#B #-

Within the next decade, the most prevalent computers may actually be

small mobile wireless devices that combine the capabilities of cell phones,

personal digital assistants +7As-, poc$et"si=ed 74s, and tablets. Their small

si=e, relatively low cost, and wide availability from many manufacturers willensure that many people will have one, or more. The computing environment of

small mobile wireless devices will be very different from today1s predominant

des$top computing environment.

2ow much faster will intelligent applications over wireless Web devices

improve productivity >o one $nows. !ut accessible intelligent Web features

offer a signi3cantly enhanced contribution to an %nformation *evolution.*oughly one"half of today1s world economy involves some related of3ce wor$.

This includes buying and selling transactions, ban$ing applications, insurance,

government, and education forms, and business"to"business transactions. There6uired information processing is currently being done mostly by speciali=ed

humans and secondarily by machines. or the most part, information technology

has been a tool to improve the productivity of the human wor$ force. Even in

that role, the Web is only beginning to scratch the surface of of3ce wor$ and

commercial transactions.

!an$ing, which typically involves straightforward, standardi=ed transactions,

could be one of the 3rst maor areas for widespread small device wireless access.

The ubi6uitous mobile phone is the new contender in 3nancial services and it

carries with it the potential for much broader access. Bnli$e earlier experiments

with smart cards and 74 ban$ing services, mobile devices loo$ li$e a naturalchannel for consumer 3nancial services. Dobile operators have built networ$s

and technology capable of cheap, reliable, and secure, person"to"merchant and

person"to"person payments. Wireless telecommunication can augment the pay"

ment system.


14/264

#6 ED7FWE*%>G T2E %>F*DT%F> GE

we may conclude that: The Web empowers individuals on a global scale, but that

the evolution of the Web re6uires the development of more intelligent features.

The


15/264

Fig01e #.#/ 7rint Gallery. Bsed with permission from D. 4. Escher1s /7rint Gallery0 Q

8))(, The D. 4. Escher 4ompany"2olland, ll rights reserved www.mcescher.com.

http://www.mcescher.com/http://www.mcescher.com/


16/264

INTERL+DE 2#3 THINKING ABO+TTHINKING

Nohn pic$ed up the two double #atte Grandes and wal$ed over to the corner table

near the 3replace where Dary was setting up the chess game. c5, continuing the main line of the Sueen1s 7awn Gambit.

/Oou1ve made my point,0 he exclaimed, /Aeep !lue did not ma$e its own deci"

sions before it moved. ll it did was accurately execute, the very sophisticated

udgments that had been preprogrammed by the human experts.0


17/264

%>TE*#BAE ?&: T2%>K%>G !FBT T2%>K%>G #"

/#et1s loo$ at it from another angle,0 Dary said as she moved >fI. /Duch li$e

a computer, Kasparov1s brain used its billions of neurons to carry out hundreds

of tiny operations per second, none of which, in isolation, demonstrates intelli"

gence. %n totality, though, we call his play Rbrilliant1. Kasparov was processinginformation very much li$e a computer does. Fver the years, he had memori=ed

and pre"analy=ed thousands of positions and strategies.0

/% disagree,0 said Nohn 6uic$ly moving e5. /Aeep !lue1s behavior was merely

logic algebra H expertly and 6uic$ly calculated, % admit. 2owever, logic

estab" lished the rules between positional relationships and sets of value"data.

fun" damental set of instructions allowed operations including se6uencing,

branching, and recursion within an accepted hierarchy.0

Dary grimaced and held up her hands, />o lectures please.0 Doving to eI she

added, / perfectly reasonable alternative explanation to logic methods is touse heuristics methods, which observe and mimic the human brain. %n particu"

lar, pattern recognition seems intimately related to a se6uence of uni6ue images

connected by special relationships. 2euristic methods seem as effective in pro"

ducing % as logic methods. The success of Aeep !lue in chess programming is

important because it employed both logic and heuristic % methods.0

/>ow who1s lecturing,0 responded Nohn, ta$ing Dary1s pawn with his bishop. /%n

my opinion, human grandmasters do not examine 8)),))),))) move se6uences

per second.0

Without hesitation Dary moved c( and said, /2ow do we $now Nust because

human grandmasters are not aware of searching such a number of positions

doesn1t prove it. %ndividuals are generally unaware of what actually does go on

in their minds. 7atterns in the position suggest what lines of play to loo$ at,

and the pattern recognition processes in the human mind seem to be invisible to

the mind.0

Nohn said, /Oou mean li$e your playing the same Sueen1s Gambit ccepted line

over and over again0 as he castled.

%gnoring him, Dary moved aI and said, /


18/264


19/264

%GO4 DEL3 WHAT IS

DECIDABLE5

OVERVIEW

%n 4hapter &, we suggested that small wireless devices connected to an intelligent

Web could produce ubi6uitous computing and empower the %nformation *evolu"

tion. %n the future,


20/264

%#


21/264

GF¨ AE#: W2T %


22/264

%&72%#FA DT2EDT%4# #FG%4

!y logic then, we mean the study and application of the principles of reason"

ing, and the relationships between statements, concepts, or propositions. #ogic

incorporates both the methods of reasoning and the validity of the results. %n

common language, we refer to logic in several ways; logic can be considered asa framewor$ or system of reasoning, a particular mode or process of reasoning,

or the guiding principles of a 3eld or discipline. We also use the term /logical0 to

describe a reasoned approach to solve a problem or get to a decision, as opposed

to the alternative /emotional0 approaches to react or respond to a situation.

s logic has developed, its scope has splintered into many distinctive branches.

These distinctions serve to formali=e different forms of logic as a science.

The distinctions between the various branches of logic lead to their limitations

and expressive capabilities, which are central issues to designing the


23/264

GF¨ AE#: W2T %


24/264

%-72%#FA DT2EDT%4# #FG%4

*ussell1s paradox represents either of two interrelated logical contradictions.

The 3rst is a contradiction arising in the logic of sets or classes.


25/264

GF¨ AE#: W2T %i


26/264

%72%#FA DT2EDT%4# #FG%4

Go¨ del is best $nown for his &'5& proof of the /%ncompleteness

Theorems.0 2e proved fundamental results about axiomatic systems showing

that in any axiomatic mathematical system there are propositions that cannot

be proved or disproved within the axioms of the system. %n particular, theconsistency of the axioms cannot be proved.

This ended &)) years of attempts to establish axioms and axiom"based logic

systems that would put the whole of mathematics on this basis. Fne maor

attempt had been by !ertrand *ussell with rinci!ia Mathematica +&'&)

&'&5-. nother was 2ilbert1s formalism, which was dealt a severe blow by

Go¨ del1s results. The theorem did not destroy the fundamental idea of

formalism, but it did demonstrate that any system would have to be more

comprehensive than that envisaged by 2ilbert. Fne conse6uence of Go¨ del1s

results implied that a computer can never be programmed to answer allmathematical 6uestions.

%n &'5(, Go¨ del proved important results on the consistency of the

axiom of choice with the other axioms of set theory. 2e visited Go¨

ttingen in the summer of &'5J, lecturing there on his set theory research and

returned to Xienna to marry dele 7or$ert in &'5J.

fter settling in the Bnited


27/264

? @ ? ∨ @

T

T

T

T

T

T

T

GF¨ AE#: W2T %


28/264

%"72%#FA DT2EDT%4# #FG%4

languages are by their nature very restrictive and as a result many 6uestions

cannot be discussed using them. Fn the other hand, F#s have precise grammars.

7redicate calculus is 6uanti3cational and based on atomic formulas that are

propositional functions and modal logic. %n 7redicate calculus, as in grammar, asubect is what we ma$e an assertion about, and a predicate is what we assert

about the subect.

A0t9;ated In=e1ene =91 FOL

utomated inference using F# is harder than using 7ropositional #ogic because

variables can ta$e on potentially an in3nite number of possible values from their

domain. 2ence, there are potentially an in3nite number of ways to apply the

Bniversal"Elimination rule of inference.

Godel1s 4ompleteness Theorem says that F# is only semidecidable. That is,

if a sentence is true given a set of axioms, there is a procedure that will determine

this. 2owever, if the sentence is false, then there is no guarantee that a procedure

will ever determine this. %n other words, the procedure may never halt in this

case. s a result, the Truth Table method of inference is not complete for F#

because the truth table si=e may be in3nite.

>atural deduction is complete for F#, but is not practical for automated

inference because the /branching factor0 in the search process is too large. This

is the result of the necessity to try every inference rule in every possible way

using the set of $nown sentences.#et us consider the rule of inference $nown as Dodus 7onens +D7-. Dodus

7onens is a rule of inference pertaining to the %MT2E> operator. Dodus 7onens

states that if the antecedent of a conditional is true, then the conse6uent must

also be true:

% D7 & Given the statements p and i' ! then (, infer 6$

The Generali=ed Dodus 7onens +GD7- is not complete for F#. 2owever, GD7

is complete for Knowledge !ases +K!s- containing only 2orn clauses.nother very important logic that will be discussed in detail in 4hapter J, is

2orn logic. 2orn clause is a sentence of the form:

% #&% 7&%#&∧

78%#&∧

$ $ $∧

7n%#&& ⇒ S%#&

where there are ) or more 7i1s, and the 7i1s and S are positive +i.e., unnegated-

literals.

2orn clauses represent a subset of the set of sentences representable in F#.

or example: 7+a- v S+a- is a sentence in F#, but is not a 2orn clause.

>atural deduction using GD7 is complete for K!s containing only 2ornclauses. 7roofs start with the given axiomsMpremises in K!, deriving new sen"

tences using GD7 until the goalM6uery sentence is derived. This de3nes a for"

ward chaining inference procedure because it moves /forward0 from the K! to

the goal.


29/264

GF¨ AE#: W2T %


30/264

H%#FA DT2EDT%4# #FG%4

irst"Frder #ogic

Aescription#ogic

2orn #ogic7rograms

Aescription#ogic

7rograms

Fig01e %.%/ This diagram shows the relationship of A# and 2orn #ogic as subsets

of F#.

Re


31/264

order syntax, variables are allowed to appear in places where normally predicate

or function symbols appear.

7redicate calculus is the primary example of logic where syntax and semantics

are both 3rst order. There are logics that have higher order syntax, but 3rst order semantics. Bnder a higher order semantics, an e6uation between predicate +or function- symbols is true, if and only if logics with a higher order semantics and

higher order syntax are statements expressing trust about other statements.

To state it another way, higher order logic is distinguished from F# in severalways. The 3rst is the scope of 6uanti3ers; in F#, it is forbidden to 6uantify over

predicates. The second way in which higher order logic differs from F# is in

the constructions that are allowed in the underlying type theory. higher order predicate is a predicate that ta$es one or more other predicates as arguments. %n

general, a higher order predicate of order n ta$es one or more +n Y &-th"order predicates as arguments +where n ) &-.

Re018i9n The91y

*ecursion is the process a procedure goes through when one of the steps of the

procedure involves rerunning a complete set of identical steps. %n mathematicsand computer science, recursion is a particular way of specifying a class of

obects with the help of a reference to other obects of the class: recursive

de3nition de3nes obects in terms of the already de3ned obects of the class.

recursive process is one in which obects are de3ned in terms of other obects of the same type. !y using a recurrence relation, an entire class of obects can be

built up from a few initial values and a small number of rules.The ibonacci numbers +i.e., the in3nite se6uence of numbers starting ), &, &,

8, 5, (, J, &5, $ $ $ , where the next number in the se6uence is de3ned as the sum

of the previous two numbers- is a commonly $nown recursive set.The following is a recursive de3nition of person1s ancestors: Fne1s parents

are one1s ancestors +base case-. The parents of any ancestor are also ancestors

of the person under consideration +recursion step-.Therefore, your ancestors include your parents, and your parents1 parents

+grandparents-, and your grandparents1 parents, and everyone else you get by

successively adding ancestors.%t is convenient to thin$ that a recursive de3nition de3nes obects in terms of a

/previously de3ned0 member of the class. While recursive de3nitions are usefuland widespread in mathematics, care must be ta$en to avoid self"recursion, in

which an obect is de3ned in terms of itself, leading to an in3nite nesting +see

ig. &"&: /The 7rint Gallery0 by D. 4. Escher is a visual illustration of self"recursion-. +igs. 8"9, 5"8, 9"8, (", I"9, "5, J"( are a progression of images

that illustrate recursion-.

KNOWLEDGE REPRESENTATION

#et us de3ne what we mean by the fundamental terms /data,0 /information,0/$nowledge,0 and /understanding.0 n item of data is a fundamental element of


32/264

K>FW#EAGE *E7*ETT%F> &&

an application. Aata can be represented by populations and labels. Aata is raw;

it exists and has no signi3cance beyond its existence. %t can exist in any form,

usable or not. %t does not have meaning by itself.

%nformation on the other hand is an explicit association between items of data. ssociations represent a function relating one set of things to another set

of things. %nformation can be considered to be data that has been given mean"

ing by way of relational connections. This /meaning0 can be useful, but does

not have to be. relational database creates information from the data stored

within it.

Knowledge can be considered to be an appropriate collection of information,

such that it is useful. Knowledge"based systems contain $nowledge as well as

information and data. rule is an explicit functional association from a set of

information things to a speci3c information thing. s a result, a rule is $nowl"edge.

%nformation can be constructed from data and $nowledge from information to

3nally produce understanding from $nowledge. Bnderstanding lies at the highest

level. Bnderstanding is an interpolative and probabilistic process that is cogni"

tive and analytical. %t is the process by which one can ta$e existing $nowledge

and synthesi=e new $nowledge. Fne who has understanding can pursue useful

actions because they can synthesi=e new $nowledge or information from what

is previously $nown +and understood-. Bnderstanding can build upon currently

held information, $nowledge, and understanding itself.rti3cial %ntelligence systems possess understanding in the sense that they are

able to synthesi=e new $nowledge from previously stored information and $nowl"

edge. n important element of % is the principle that intelligent behavior can be

achieved through processing of symbolic structures representing increments of

$nowledge. This has produced $nowledge"representation languages that allow the

representation and manipulation of $nowledge to deduce new facts from the exist"

ing $nowledge. The $nowledge"representation language must have a well"de3ned

syntax and semantics system while supporting inference.

Three techni6ues have been popular to express $nowledge representation and

inference: +1 - logic"based approaches, +2 - rule"based systems, and +* -

frames and semantic networ$s.

#ogic"based approaches use logical formulas to represent complex relation"

ships. They re6uire a well"de3ned syntax, semantics, and proof theory. The

formal power of a logical theorem proof can be applied to $nowledge to derive

new $nowledge. #ogic is used as the formalism for programming languages

and databases. %t can also be used as a formalism to implement $nowledge

methodology. ny formalism that admits a declarative semantics and can be

interpreted both as a programming language and a database language is a $nowl"

edge language. 2owever, the approach is inCexible and re6uires great precision

in stating the logical relationships. %n some cases, common sense inferences

and conclusions cannot be derived, and the approach may be inef3cient, espe"

cially when dealing with issues that result in large combinations of obects or

concepts.


33/264

GF¨ AE#: W2T % or other conditional rules. This approach

is more procedural and less formal in its logic. s a result, reasoning can be

controlled through a forward or bac$ward chaining interpreter.rames and semantic networ$s capture declarative information about related

obects and concepts where there is a clear class hierarchy and where the principle

of inheritance can be used to infer the characteristics of members of a subclass.

The two forms of reasoning in this techni6ue are matching +i.e., identi3cation

of obects having common properties-, and property inheritance in which prop"

erties are inferred for a subclass. rames and semantic networ$s are limited to

representation and inference of relatively simple systems.

%n each of these approaches, the $nowledge"representation component +i.e.,

problem"speci3c rules and facts- is separate from the problem"solving and infer"ence procedures.

or the 0 statements. The rules of inference are

applied to determine whether the axioms are suf3cient to ensure the truth of the

goal statement. The execution of a logic program corresponds to the constructionof a proof of the goal statement from the axioms.


34/264

*T%%4%# %>TE##%GE>4E &-

%n the logic programming model, the programmer is responsible for specifying

the basic logical relationships and does not specify the manner in which the

inference rules are applied. Thus

#ogic Z 4ontrol [ lgorithms

The operational semantics of logic programs correspond to logical inference. The

declarative semantics of logic programs are derived from the term model. The

denotation of semantics in logic programs are de3ned in terms of a function that

assigns meaning to the program. There is a close relation between the axiomatic

semantics of imperative programs and logic programs.The control portion of the e6uation is provided by an inference engine whose

role is to derive theorems based on the set of axioms provided by the programmer.The inference engine uses the operations of resolution and uni3cation to construct

proofs. aulty logic models occur when the essential problem has not been clearly

stated or de3ned.

7rogram developers wor$ carefully to construct logic models to avoid logic

conCicts, recursive loops, and paradoxes within their computer programs. s

a result, programming logic should lead to executable code without paradox

or conCict, if it is Cawlessly produced. >evertheless, we $now that /bugs0 or

programming errors do occur, some of which are directly or indirectly a result

of logic conCicts.

s programs have grown in si=e from thousands of line of code to millions of

lines, the problems of bugs and logic conCicts have also grown. Today, programs,

such as operating systems, can have )8( million lines of codes and are considered

to have hundreds of thousands of bugs, most of which are seldom encountered

during routine program usage.4on3ning logic issues to beta testing on local servers allows programmers

reasonable control of conCict resolution. >ow, consider applying many lines of

application code logic to the


35/264

GF¨ AE#: W2T %ET and


36/264

&T2E T%4 WE!

an owner and a ban$ holder1s portfolio is the sum of his accounts; and who can

access and update these obects.

s an example, consider an online store that provides real"time pricing and

availability information. The site will provide a form for you to choose a product.When you submit your 6uery, the site performs a loo$up and returns the results

embedded within an 2TD# page. The site may implement this functionality in

numerous ways.

The Web server delegates the response generation to a script; however, the

business logic for the pricing loo$up is included from an application server. With

that change, instead of the script $nowing how to loo$ up the data and formulate

a response, the script can simply call the application server1s loo$up service.

The script can then use the service1s result when the script generates its 2TD#

response.The application server serves the business logic for loo$ing up a product1s

pricing information. That functionality does not say anything about display or

how the client must use the information. %nstead, the client and application server

send data bac$ and forth. When a client calls the application server1s loo$up

service, the service simply loo$s up the information and returns it to the client.

!y separating the pricing logic from the 2TD# response"generating code, the

pricing logic becomes reusable between applications. second client, such as a

cash register, could also call the same service as a cler$ chec$ing out a customer.

*ecently, ePtensible Dar$up #anguage +PD#- Web


37/264

+ i g n a t u r e

, n c r " ! t i o n

GF¨ AE#: W2T %


38/264

parties who have agreed to the de3nitions beforehand, their lac$ of semantics pre"

vents machines from reliably performing this tas$ with new PD# vocabularies.

%n addition, the ontology of *A and *A


39/264

this tas$. more detailed presentation on paradoxes on the Web and what is

solvable on the Web will be provided in the next few chapters.

In=e1ene Engine8 =91 the Se;anti We>

%nference engines process the $nowledge available in the


40/264

EPE*4%


41/264

Fig01e %.)/ Escher and the Aroste effect +http:MMescherdroste.math.leidenuniv.nlM-

http://escherdroste.math.leidenuniv.nl/http://escherdroste.math.leidenuniv.nl/


42/264

INTERL+DE 2%3 TR+TH AND BEA+T'

s Nohn passed with a sour loo$ on his face, Dary loo$ed up from her text boo$

and as$ed, /Aidn1t you enoy the soccer game0

/2ow can you even as$ that when we lost0 as$ed Nohn gloomily.

/% thin$ the team performed beautifully, despite the score,0 said Dary.

This instantly frustrated Nohn and he said, /Ao you $now Dary that sometimes %

3nd it disarming the way you express obects in terms of beauty. % 3nd that simply

accepting something on the basis of its beauty can lead to false conclusions0

Dary reCected upon this before offering a gambit of her own, /Well Nohn, do

you $now that sometimes % 3nd that relying on obective truth alone can lead to

unattractive conclusions.0

Nohn became Custered and reCected his dismay by demanding, /Give me an

example.0

Without hesitation, Dary said, /7erhaps you will recall that in the late &'8)s,

mathematicians were 6uite certain that every well"posed mathematical 6uestion

had to have a de3nite answer H either true or false. or example, suppose

they claimed that every even number was the sum of two prime numbers,0

referring to Goldbach1s 4onecture, which she had ust been studying in her

text boo$. Dary continued, /Dathematicians would see$ the truth or falsity of the claim by examining a chain of logical reasoning that would lead in a 3nite

number of steps to prove if the claim were either true or false.0

/


43/264

)) %>TE*#BAE ?8: T*BT2 >A !EBTO

/Than$ you for that clear explanation,0 said Nohn. /!ut isn1t such a fact simply a

translation into mathematic terms of the famous #iar1s 7aradox: RThis statement

is false.1 0

/Well, % thin$ it1s a little more complicated than that,0 said Dary. /!ut Go¨del did identify the problem of self"reference that occurs in the #iar1s 7aradox.

>evertheless, Go¨ del1s theorem contradicted the thin$ing of most of the

great mathematicians of his time. The result is that one cannot be as certain as

the mathematician had desired.


44/264

&T+RING3 WHAT IS MACHINEINTELLIGENCE5

OVERVIEW

Web intelligence is an issue of philosophy as much as application. %t has beensuggested that the next generation of Web architecture, the


45/264

)6 TB*%>G: W2T %< D42%>E %>TE##%GE>4E

war as %ntelligent Dachinery, and are now referred to as %. Key to this 3eld of

study is the de3nition of what is meant by the terms /thin$ing0 and /intelligence.0

Thin$ing is often ambiguously de3ned, but generally can be applied to a

complex processes that uses concepts, their interrelationships, and inference to produce new $nowledge. We can extend the concept of thin$ing and identify an

intelligent individual as one who is capable of accurate memory recall or able to

apply logic to extend $nowledge.

%t is possible to extend the description of intelligence to nonhuman entities as

well, such as in %. !ut we fre6uently mean something different than in the case

of human intelligence. or example, while one might be 6uite impressed with the

intelligence of a child prodigy who can perform dif3cult arithmetic calculations

6uic$ly and accurately, a computer that could perform the same calculations

faster and with greater accuracy would not be considered intelligent.While it is still not possible to resolve controversial differences of opinion over

the nature of human intelligence, it is possible to recogni=e certain attributes that

most would agree reCect the concept. These include such elements as: the ability

to learn; the ability to assimilate; the ability to organi=e and process information;

and the ability to apply $nowledge to solve complex problems. !y extension then,

many of these attributes of human intelligence can be traced into the various areas

of research in the 3eld of arti3cial intelligence. rti3cial intelligence addresses

the basic 6uestions of what it means for a machine to have intelligence.

%n &'9, shortly after the end of World War %%, English mathematician lan

Turing 3rst started to seriously explore the idea of intelligent machines. !y&'(I, Nohn Dc4arthy of D%T coined the term rti3cial %ntelligence, and by

the late &'()s, there were many researchers in %, most basing their wor$ on

programming computers. Eventually, % became more than a branch of science:

it expanded far beyond mathematics and computer science into 3elds such as

philosophy, psychology, and biology.

ALAN T+RING

lan Dathison Turing was one of the great pioneers of the computer 3eld. 2e

designed /The Turing machine0 and /Turing1s test.0 s a mathematician he

applied the concept of the algorithm to digital computers. 2is research into

the relationships between machines and >ature created the 3eld of %. 2is

insights opened the door into the information age.

Turing was born in #ondon, B.K., on Nune 85, &'&8 +see ig. 5"&-. 2e had

a dif3cult childhood, and was separated from his parents for long periods of

time. 2e struggled through his school years, but he excelled in mathematics.

2e entered King1s 4ollege, 4ambridge, in &'5& to study mathematics.2e too$ an early interest in the wor$s of von >eumann, Einstein, Edding"

ton, and *ussell1s Introduction to Mathematical hiloso!h".

!y &'55, Turing1s interest in mathematical logic intensi3ed. 2e suggested

that a purely logic"based view of mathematics was inade6uate; and that


46/264

W2T %< D42%>E %>TE##%GE>4E )

Fig01e &.#/ 7hoto of lan Turing.

mathematical propositions possessed a variety of interpretations. Turing1s

achievements at 4ambridge were primarily centered on his wor$ in proba" bility theory. 2owever, he began to focus his attention on the 6uestion of

mathematical decidability. %n &'5I, he published the paper n 1om!utable

3umbers, 5ith an 6!!lication to the ntscheidungs!roblem$ Turing introduced

the idea of a computational machine, now called the /Turing machine,0 which

in many ways became the basis for modern computing +see %nterlude ?5-. The

Turing machine was an abstract device intended to help investigate the extent

and limitations of computation. Turing machine is a /state machine0 that

can be considered to be in any one of a 3nite number of states. %nstructions

for a Turing machine consist of speci3ed conditions under which the machinewill transition between one state and another using a precise, 3nite set of rules

+given by a 3nite table- and depending on a single symbol it read from a tape,

representing the state of the machine.

Turing machine includes a one"dimensional tape, theoretically of in3nite

length, divided into cells. Each cell contains one symbol, either /)0 or /&.0 The

machine has a read write head to scan a single cell on the tape. This read

write head can move left and right along the tape to scan successive cells.

The action of a Turing machine is determined completely by +1- the current

state of the machine; +2- the symbol in the cell currently being scanned by the

head; and +*- a table of transition rules, which serves as the /program0 for

the machine. %f the machine reaches a situation in which there is not exactly

one transition rule speci3ed, then the machine halts.

%n modern terms, the tape represents the memory of the machine, and the

read write head represents the memory bus through which data is

accessed


47/264

)! TB*%>G: W2T %< D42%>E %>TE##%GE>4E

+and updated- by the machine. Two important factors are +1- the machine1s

tape is in3nite in length; and +2- a function will be Turing"computable if

there exists a set of instructions that will result in the machine computing

the function regardless of how long it ta$es; that is, the function will besuccessfully computed in a 3nite number of steps.

These two assumptions ensure that no computable function will fail to

be computable on a Turing machine +i.e., Turing computable- solely because

there is insuf3cient time or memory to complete the computation.

Turing de3ned a computable number as a real number whose decimal

expansion could be produced by a Turing machine. 2e showed that, although

only countable many real numbers are computable, most real numbers are not

computable +e.g., irrational numbers-. 2e then described a number that is not

computable and remar$ed that this seemed to be a paradox since he appearedto have described, in 3nite terms, a number that cannot be described in 3nite

terms. 2owever, Turing understood the source of the apparent paradox. %t

is impossible to decide +using another Turing machine- whether a Turing

machine with a given table of instructions will output an in3nite se6uence

of numbers. Turing1s paper contains ideas that have proved of fundamental

importance to mathematics and to computer science ever since.

%n &'5', he was recruited to wor$ for the !ritish Government brea$ing

the German Enigma codes. Together with W. G. Welchman, Turing devel"

oped the !ombe, a machine based on earlier wor$ by 7olish mathemati"

cians, which, from late &'9), was decoding messages sent by the Enigma

machines of the German #uftwaffe. !y the middle of &'9&, Turing1s sta"

tistical approach, together with captured information, had led to the German

>avy messages being decoded, using the 3rst practical programmed computer,

called 4olossus.

%n Darch &'9I, Turing submitted a design proposing the utomatic 4omput"

ing Engine +4E-. Turing1s design was a prospectus for a modern computer.

%n &'(), Turing published om!uting Machiner" and Intelligence in the

ournal Mind . %t is a remar$able wor$ on 6uestions that would become increas"

ingly important as the 3eld of computer science developed. 2e studied prob"lems that today lie at the heart of arti3cial intelligence. %n his &'() paper, he

suggested what has become $nown as a Turing1s test, still the acid test for

recogni=ing intelligence in a machine. Turing died of cyanide poisoning, an

apparent suicide, in &'(9.

T+RING TEST AND THE LOEBNER PRIE

%n Turing1s seminal wor$ entitled om!uting Machiner" and Intelligence more

than () years ago, he suggested that a computer can be called intelligent if itcould deceive a human into believing that it was human. 2is test H called

the Turing test H consists of a person as$ing a series of 6uestions to both a

human subect and a machine +see %nterlude ?9-. The 6uestioning is done via a

$eyboard so that the 6uestioner has no direct interaction. machine with true

intelligence


48/264

*T%%4%# %>TE##%GE>4E )"

will pass the Turing test by providing responses that are suf3ciently human"li$e

that the 6uestioner cannot determine which responder is human. scaled down

version of the Turing test, $nown as the #oebner 7ri=e, re6uires that machines

/converse0 with testers only on a limited topic in order to demonstrate their intelligence.

OHN SEARLE(S CHINESE ROOM

While % enthusiasts have pursued and promoted technologies related to machine

intelligence, doubts about the possibility of producing intelligent machines have

also continued over the years. %n &'J), Nohn


49/264

-$ TB*%>G: W2T %< D42%>E %>TE##%GE>4E

Today, % still means different things to different people.


50/264

*E%>G W%T2 T%4 >ETWF*K< -#

incorporate the soft"computing paradigm as well. The bene3t of such a step

will be the creation of adaptive software. This would imply that soft"computing

applications will have the ability to adapt to changing environments and input.

While the


51/264

-% TB*%>G: W2T %< D42%>E %>TE##%GE>4E

with concepts li$e /some of 0 and /all of ,0 which are $nown as 6uanti3ers.

They are impossible to represent in normal semantic nets, and extensions have

been proposed to deal with the problem. 2owever, there are still limitations to

semantic nets and the only way to get around these is to use a more complexcalculus, such as that offered by frames.

rames offer a more highly structured hierarchy of nodes. !asically, each

level in the hierarchy is of a given type, and there are certain options for 3lling

the slot of this node type. %n this way, speci3c $nowledge can be represented

and manipulated. lot of wor$ has been done in the 3eld of hypertext networ$

organi=ation. dvocates of hypertext suggest that the ideas relevant to a subect

can be represented best by a simple associative networ$, or web, with multiple

interrelationships speci3ed. The idea is that hypertext representation mimics the

associative networ$s of the human brain a notion derived directly from Xannevar !ush1s seminal &'9( 6tlantic Monthl" article entitled /s We Day Thin$.0

%t has long been recogni=ed that hypertext structures mimic semantic networ$s.

odes

represent concepts, and lin$s represent relationships between them. hypertext

system with arbitrary lin$ types corresponds to a free semantic net. %f the hyper"

text system allows ust a limited number of lin$ types, the underlying semantic

net is restricted.


52/264

4FD7BTT%F># 4FD7#EP%TO -&

%n computational complexity, a single problem is in reality a class of related6uestions, where each 6uestion can be considered a 3nite"length string. or

example, a problem might be to determine the prime factors of a number; this

problem could consist of many or even an in3nite number of instances of the problem.

The time complexity of a problem is characteri=ed by the number of stepsthat it ta$es to solve an instance of the problem as a function of the si=e of theinput +usually measured in bits-, using the most ef3cient algorithm. %f a probleminstance involves an input value that is n bits long, and the problem can be solved

in n8 steps, we say the problem has a time complexity of n8 . The actual number of steps will depend on exactly what machine computational process is beingused. >evertheless, we generally use the terminology that if the complexity is of

order n

8

\or F+n

8

-] on one typical computer, then it will also have complexityF+n8 - on other computers.

Dei8i9n P19>o. or example, the problem to determine whether or not a number is prime can be stated as: given an

integer written in binary, return whether it is a prime number or not. decision

problem can be considered to be e6uivalent to a language, in the sense that it can

be considered to be a set of 3nite"length strings. or a given decision problem,the e6uivalent language is the set of all strings for which the answer is OE


53/264

-) TB*%>G: W2T %< D42%>E %>TE##%GE>4E

general only problems that have polynomial"time solutions are solvable for more

than the smallest inputs. 7roblems that are $nown to be intractable include those

that are exponential"time"complete. %f >7 is not the same as 7, then the >7"

complete problems are also intractable.To see why exponential"time solutions are not usable in practice, consider a

problem that re6uires 8n operations to solve +where n is the si=e of the input-.or a relatively small input si=e of n [ &)), and assuming a computer that can

perform &)&) operations per second, a solution would ta$e about 9 ^ &)&8 years,far longer than the age of the universe.

DESCRIPTION LOGIC

The 6uestion of how best to represent $nowledge in a database or informa"

tion system has long been recogni=ed as a $ey issue in %. The main research

effort in Knowledge *epresentation +K*- is centered on theories and

systems for expressing structured $nowledge and for accessing and reasoning

with it. Aescription #ogics +A#- is an important powerful class of logic"based

$nowl" edge representation languages.

The basic components of A# include a family of $nowledge

representation formalisms designed for the representation of and reasoning about

semantic net" wor$s and frame"based systems. Aescription #ogics are alsoclosely related to Dodal #ogic +an extension of propositional calculus, which

uses operators that express various /modes0 of truth, such as: necessarily ,

possibly , probably , it has always been true that , it is permissible that ,

it is believed that and A# +a multimodal logic in which there are explicit

syntactic constructs, called programs, whose main role is to change the values of

variables, thereby changing the truth values of formulas based on program

representation-, and it has turned out that A#s are also well suited to the

representation of and reasoning about database conceptual models, information

integration, and ontologies. variety of different A#s exist with different

expressive power and different computational complexity for the corresponding

inference problems.

The


54/264

%>E*E>4E E>G%>E< --

re6uiring everyone to share exactly the same de3nition of common concepts.

!ut central control is stiCing, and increasing the si=e produces complexity that

rapidly becomes unmanageable. These systems limit the 6uestions that can be

as$ed reliably. %n avoiding the problems, traditional $nowledge representationsystems narrow their focus and use a limited set of rules for ma$ing inferences.

7ossible inference engines for the


55/264

-6 TB*%>G: W2T %< D42%>E %>TE##%GE>4E

about the information in the $nowledge base, and for formulating conclusions.%nference engines process $nowledge available in the ormally, an agent willhave a repertoire of actions available to it.


56/264

4F>4#B -

%n the &'J)s, obect"oriented programming made it easier to reorgani=e for

changes, because functionality was split up into separate classes. typical appli"

cation was a des$top publishing system using user"initiated events +mouse clic$s

or menus-. The problem with using existing software, however, is that it ta$estoo much time and money to modify, and it is brittle when used in situations for

which it was not explicitly designed. daptive software design methodologies

can help alleviate this problem.

primary element in adaptive software is reali=ing that optimi=ation of

structured programs is not the only solution to increasingly complex problems.

The optimi=ation approach is based on maintaining control to impose order on

uncertainty. %mposed order is the product of rigorous engineering discipline and

deterministic, cause"and"effect"driven processes. The alternative idea is one of

an adaptive mindset, of viewing organi=ations as complex adaptive systems, andof creating emergent order out of a web of interconnected components.

4omplexity involves the number of interacting agents and the speed of agent

interaction. or software products, the need for adaptive development arises when

there are a great many independent operators H developers, customers,

vendors, competitors, stoc$holders H interacting with each other, fast enough

that linear cause and effect rules are no longer suf3cient for success.


57/264

-! TB*%>G: W2T %< D42%>E %>TE##%GE>4E

intelligence.


58/264

Fig01e &.%/ This is a /completed0 representation of Escher1s 7rint Gallery, which in"

cludes the 3lled"in hole in igure &"&, created +http:MMescherdroste.math.leidenuniv.nlM- by

2endri$ #enstra and !art de


59/264

INTERL+DE 2&3 COMP+TINGMACHINES

2aving had the 3nal word in their last discussion, Nohn was feeling a little smug

as he listened to his ipod. Dary sat down next to him on the library steps. Their last class had been on computer design and they were both thin$ing about ust

how far the new technology could evolve.

Nohn said, /s you suggested earlier, Go¨ del was concerned that a logic

system had to be consistent and then he determined that no logic system can prove itself to be consistent.0

/True,0 replied Dary. /!ut it was Turing who built on Go¨ del1s 3ndings.


60/264

%>TE*#BAE ?5: 4FD7BT%>G D42%>E< 6#

operations represent a program. #et1s imagine that % want to use the machine toadd two numbers & and 8 together. The computing machine would begin with

placing a R&1 in the 3rst location and a R81 in the second location and then the

computer consults a program for how to do addition. The instructions would saygather the numbers from the two locations and perform a summing operation

to yield the sum of the two numbers and place the resultant number R51 in thethird location. This process could be considered to mimic the operations a human

would perform.0

Nohn replied solemnly, /


61/264

6% %>TE*#BAE ?5: 4FD7BT%>G D42%>E<

answer all those Trivial 7ursuit 6uestions. Why can1t that computer, be considered

intelligent.0

Nohn said, /Well human thin$ing involves complicated interactions within the

biological components of the brain. %n addition, the processes of communicationand learning are also important elements of human intelligence.0

Dary replied, /!y mentioning intelligent communication you have led us to

Turing1s test for machine intelligence.0

Nohn said, /F$, but please, let1s tal$ about that tomorrow.0


62/264

)BERNERS.LEE3 WHAT ISSOLVABLE ON THE WEB5

OVERVIEW

When Tim !erners"#ee was developing the $ey elements of the World Wide Web,he showed great insight in providing 2ypertext Dar$up #anguage +2TD#- as a

simple easy"to"use Web development language. s a result, it was rapidly and

widely adopted. To produce Web information re6uired s$ills that could be learned

with a high school level education. 4onse6uently, personal computing merged

with global networ$ing to produce the World Wide Web.

The continuing evolution of the Web into a resource with intelligent features,

however, presents many new challenges. The solution of the World Wide Web

4onsortium +W54- is to provide a new Web architecture that uses additional lay"

ers of mar$up languages that can directly apply logic. 2owever, the addition of ontologies, logic, and rule systems for mar$up languages means consideration of

extremely dif3cult mathematic and logic conse6uences, such as paradox, recur"

sion, undecidability, and computational complexity on a global scale. Therefore,

it is important to 3nd the correct balance between achieving powerful reasoning

with reasonable complexity on the Web. This balance will decide what is solvable

on the Web in terms of application logic.

This chapter will brieCy review !erners"#ee1s contribution in developing the

Web. Then, we loo$ at the impact of adding formal logic to Web architecture

and present the new mar$up languages leading to the future Web architecture: the


63/264

6) !E*>E* in &'J&. %n &'J9, he returned to

4E*>, to wor$ on distributed real"time systems for scienti3c data ac6uisition

and system control. Auring this time at 4E*>, !erners"#ee began to

conceive of a different type of En6uire system. The %nternet was &( years

old and had proven to be


64/264

6-T2E WF*#A W%AE WE!

Fig01e ).#/ 7hoto of Tim !erner"#ee.

a reliable networ$ing system, but it was still cumbersome to use. !erners"

#ee was loo$ing at ways to simplify the exchange of information. !erners"

#ee began to imagine a system that would lin$ up all the computers of his

colleagues at 4E*>, as well as those of 4E*>1s associates in laboratories

around the world.

%n &'J', !erners"#ee with a team of colleagues developed 2TD#, an

easy"to"learn document coding system that allows users to clic$ onto a lin$

in a document1s text and connect to another document. 2e also created an

addressing plan that allowed each Web page to have a speci3c location $nown

as a B*#. inally, he completed 2TT7 a system for lin$ing these documents

across the %nternet. 2e also wrote the software for the 3rst server and the 3rst

Web client browser that would allow any computer user to view and navigate

Web pages, as well as create and post their own Web documents.

%n the following years, !erners"#ee improved the speci3cations of B*#s,

2TT7, and 2TD# as the technology spread across the %nternet.

While many early Web developers became %nternet entrepreneurs, !erners"

#ee eventually chose an academic and administrative life. 2e left 4E*> in

the early &'')s and spent research stints at various laboratories, includingPerox1s 7alo lto *esearch 4enter +7*4-, in 4alifornia, and the #aboratory

for 4omputer


65/264

66 !E*>E*


66/264

6T2E WF*#A W%AE WE!

Web

The Web has changed from its original structure of a distributed, high"reliability,

open system without a superimposed logic or metadata. Today, the basic infor"mation is still displayed as a distributed open system, but the development of

portals, such as, Oahoo, Google F#, and D has focused Web entry and led

to controlling traf3c to partisan sites. %n addition, business logic has migrated

primarily into two segregated server framewor$s: active server pages and ava

server pages. The result has produced a decentrali=ed Web system with critical

proprietary portal"centric nodes and framewor$s.

%n the future, we can expect signi3cant improvements, such as increased aver"

age bandwidth, the use of open standards to facilitate advanced mar$up languages,

the application of metadata, and the use of inference search.The paradigm of the Web is centered on the client server interaction, which is

a fundamentally asymmetric relationship between providers, who insert content

into the Web hypertext +server- and users who essentially read texts or provide

answers to 6uestions by 3lling out forms +clients-. The hyperlin$s of the Web

represent structures of meaning that transcend the meaning represented by indi"

vidual texts. t present, these Web structures of meaning lac$ longevity and can

only be blindly used, for example by search engines, which at best optimi=e

navigation by ta$ing into account the statistical behavior of Web users.

%n effect, the Web has developed as a medium for humans without a focus on

data that could be processed automatically. 2ypertext Dar$up #anguage is notcapable of being directly exploited by information retrieval techni6ues, hence the

Web is restricted to manual $eyword searches.The problem at present is that there is no way to construct complex networ$s

of meaningful relations between Web contents. %n fact, the providers have no


67/264

6! !E*>E*


68/264

6"T2E T%4 WE! *FAD7

indexes that contain very complete lists of all occurrences of a given term, and

then use logic to weed out all but those that can be of use in solving the given

problem.

%f the


69/264

$ !E*>E*


70/264

#T2E T%4 WE! *FAD7

Dost of FW#s power comes from primitives for expressing classi3cations.

The FW# provides a toolbox of class expressions, which bring the power of

mathematical logic and set theory to the tric$y and important tas$ of mapping

ontologies through classi3cations.

We> Ont9


71/264

% !E*>E*


72/264

&T2E T%4 WE! *FAD7

Evaluating different $nowledge"representation methodologies is highly depen"

dent on the issue of scaling semantic applications for the Web. The complexity

introduced with each methodology will have to be closely analy=ed.

SWRL and R0


73/264

) !E*>E*


74/264

#FG%4 F> T2E T%4 WE! -

Se;anti We> Se17ie8

Bsing the new generation of Web mar$up languages, such as FW#, an ontology

for Web


75/264

6 !E*>E*i


76/264

EPE*4%


77/264

Fig01e ).%/ ooming into igure 5"8 to create a blowup by a factor 8.

+http:MMescherdroste.math.leidenuniv.nlM -.


78/264

INTERL+DE 2)3 T+RING(S TEST

The next day, Dary was once again sitting on the library steps when Nohn came

by and oined her to resume their conversation.

/Well, after our last discussion on Turing1s machine you should already have

considered the next step.0

Nohn said, /!y the next step, % expect you mean determining ust how intelligent

the Turing machine could become0

Dary said, /Oes, and Turing was obliging in suggesting a test to evaluate ust

such a case. The Turing test is a behavioral approach to determining whether or

not a machine is intelligent.0

Nohn said, /nd can you state the conditions of the test0

Dary said, /Ff course. Friginally, lan Turing proposed that conversation was

the $ey to udging intelligence. %n his test, a udge has conversations +via teletype-

with two subects, one human, the other a machine. The conversations can be

about anything, and would proceed for a set period of time +e.g., & h-. %f, at the

end of this time, the udge cannot distinguish the machine from the human on

the basis of the conversation, then Turing argued that we would have to say that

the machine was intelligent.0

Nohn said, /There are a number of different views about the utility of the Turingtest.


79/264

!$ %>TE*#BAE ?9: TB*%>G1< TE


80/264

PART II

WEB ONTOLOG' AND LOGIC

!efore we can achieve anything approaching arti3cial intelligence or Rthin$ing1on the Web, the next generation Web architecture must be able to support the

basic elements of logic and automation.

%n 7art %%, Web Fntology and #ogic are presented: the solution of the World

Wide Web 4onsortium +W54- to deliver


81/264

!#


82/264

!% WE! F>TF#FGO >A #FG%4

4hapter ' presents the current state of development for the


83/264

-RESO+RCE DESCRIPTIONFRAMEWORK

OVERVIEW

The ePtensible Dar$up #anguage +PD#- is a universal meta"language for de3n"ing mar$up. %t provides a uniform framewor$ for exchanging data between

applications. %t builds upon the original and most basic layer of the Web, 2yper"

text Dar$up #anguage +2TD#-. 2owever, PD# does not provide a mechanismto deal with the semantics +the meaning- of data.

*esource Aescription ramewor$ +*A- was developed by the World Wide

Web 4onsortium +W54- for Web"based metadata in order to build and extendPD#. The goal of *A is to ma$e wor$ easier for autonomous agents and

automated services by supplying a rudimentary semantic capability.

The *A is a format for data that uses a simple relational model that allows

structured and semistructured data to be mixed, exported, and shared across differ"ent applications. %t is a data model for obects and relationships between them and

is constructed with an obect"attribute"value triple called a statement. While PD#

provides interoperability within one application +e.g., producing and exchanging

ban$ statements- using a given schema, *A provides interoperability across

applications +e.g., importing ban$ statements into a tax calculating program-.

This chapter highlights some basic characteristics of 2TD# and PD#. Then,

we introduce *A and present fundamental concepts, such as resources, prop"

erties, and statements. We de3ne the subect, predicate, and obect as the *A

triplet and illustrate it as a graph. Then, we introduce *A


84/264

!) *E


85/264

PD# #>GBGE !-

PD# is not a replacement, but rather a complementary technology to 2TD#.

While PD# is already widely used across the Web today, it is still a relatively

new technology. The PD# is a meta language, which means it is a language

used to create other languages. %t can provide a basic structure and set of rulesfor developing other mar$up languages. !y using PD#, it is possible to create

a uni6ue mar$up language to model ust about any $ind of information.

Dar$up text, in general, needs to be differentiated from the rest of the docu"

ment text by delimiters. Nust as in 2TD#, the angle brac$ets +`- and the names

they enclose are delimiters called tags. Tags demarcate and label the parts of the

document and add other information that helps de3ne the structure. The PD#

document lets you name the tags anything you want, unli$e 2TD#, which limits

you to prede3ned tag names. Oou can choose element names that ma$e sense in

the context of the document. Tag names are case"sensitive, although either casemay be used as long as the opening and closing tag names are consistent.

The text between the tags is the content of the document, raw information that

may be the body of a message, a title, or a 3eld of data. The mar$up and the

content complement each other, creating an information entity with partitioned

labeled data in a handy pac$age.

%n its simplest form, an PD# document is comprised of one or more named

elements organi=ed into a nested hierarchy. n element consists of an open"

ing tag, some data, and a closing tag. or any given element, the name of the

opening tag must match that of the closing tag. closing tag is identical to anopening tag except that the less"than symbol +`- is immediately followed by a

forward"slash +M-. Keeping this simple view, we can construct the maor portions

of the PD# document to include the following six ingredients: +1- PD# decla"

ration +re6uired-, +2- Aocument Type Ae3nition +or PD#


86/264

!6 *ET message +?74AT-

]

`message

2ello World

`Mmessage

The text between the tags; `message `Mmessage; is the 2ello World text.

%n addition, PD# is both a powerful and essential language for Web


87/264

*A #>GBGE !

model. This is where *A and metadata can provide new machine"processing

capabilities built upon PD# technology.

What is metadata %t is information about other data. !uilding upon PD#,

the W54 developed the *A metadata standard. The goal was to add semanticsde3ned on top of PD#.

While *A is actually built upon a very simple model and it can support

very large"scale information processing. n *A document can delineate precise

relationships between vocabulary items by constructing a grammatical represen"

tation. ssertions in different *A documents can be combined to provide far

more information together than they could separately. s a result, *A provides

a powerful and Cexible 6uery structure.

RDF T1i?


88/264

!! *E


89/264

relates the obect + #-:

The boo$

to the obect + "-:

Go¨ del, scher, Bach: 6n ternal Golden Braid$

Thin$ of a collection of interrelated *A statements represented as a graph

of interconnected nodes. The nodes are connected via various relationships. or

example, let us say each node represents a person. Each person might be related

to another person because they are siblings, parents, spouses, friends, or employ"

ees. Each interconnection is labeled with the relationship name. nother type of relationship is the physical properties of a node, such as the name or ob of a

person +see the riend of a riend application at the end of this chapter-.

The *A is used in this manner to describe these relationships. %t does not

actually include the nodes directly, but it does indirectly since the relationships

point to the nodes. t any time, we could introduce a new node, such as a

newborn child, and all that is needed is for us to add the appropriate relationship

for the two parents.

BASIC ELEMENTS

Dost of the elements of *A concern classes, properties, and instances of classes.

This section presents the language components essential to introducing these

elements.

Synta:

!oth *A and *A


90/264

TABLE -.#/ RDF D90;ent Pa1t8

Aocument 7arts *A Aocument

2eader PD#


91/264

The content of /rdf:Aescription0 elements are called property elements. Aes"

criptions may be de3ned within other descriptions producing nested descriptions.

urther de3nition of %A is found through the /rdf:resource0 attribute and the

/rdf:type0 element introduces structure to the rdf document.While *A is re6uired to be well formed, it does not re6uire PD#"style

validation. The *A parsers do not use Aocument Type Ae3nitions +ATAs- or

PD# et, which provides resource B*%s for words. The predicate is rdf:type, which is in the *A namespace since the /type0 predicate

is built"in to *A. The full name is

http:MMwww.w5.orgM&'''M)8M88"rdf"syntax"ns?type.

lso, the attribute rdf:datatype[xsd;integer indicates the data type

de3nes the range as an integer. The *A uses PD# data types that includes

a wide range of data types. %n addition, *A allows any externally de3ned datatyping scheme.

>ow we will use Table ("& and seriali=e Example ("&: /The boo$ has the title

Go¨del, scher, Bach: 6n ternal Golden Braid ,0as:

`,xml version[&.)

`rdf:*A

xmlns:rdf[http:MMwww.w5.orgM&'''M)8M88"rdf"syntax"ns?

xmlns:dc[http:MMpurl.orgMdcMelementsM&.&M

`rdf:Aescription rdf:about[/http:MMwww.ama=on.comMboo$s0`dc:titleGo¨ del, scher, Bach: 6n ternal Golden Braid ̀ Mdc:title

`Mrdf:Aescription

`Mrdf:*A

3ote: dc stands for Aublin 4ore: a well"established *A vocabulary for publi"

cations +see http:MMdublincore.o r gM-.

http://www.web-iq.com/people/Johnhttp://xmlns.com/wordnet/1.6/http://xmlns.com/wordnet/1.6/http://www.w3.org/1999/02/22-rdf-syntax-ns#typehttp://www.w3.org/1999/02/22-rdf-syntax-ns#http://purl.org/dc/elements/1.1/http://www.amazon.com/bookshttp://dublincore.org/http://www.web-iq.com/people/Johnhttp://xmlns.com/wordnet/1.6/http://xmlns.com/wordnet/1.6/http://www.w3.org/1999/02/22-rdf-syntax-ns#typehttp://www.w3.org/1999/02/22-rdf-syntax-ns#http://purl.org/dc/elements/1.1/http://www.amazon.com/bookshttp://dublincore.org/


92/264

"% *E0


93/264

"&!T<

Reiati9n

The *A allows us to ma$e statements about statements using a rei3cation mech"

anism. This is particularly useful to describe belief or trust in other statements.

Example ("8 discusses the interpretation of multiple statements in relationshipto *A statements.

E*AMPLE -.%/ %nterpreting Dultiple


94/264

Aouglas *.2ofstadter

Gbel, Escher,!ach: n EternalGolden !raid

") *E


95/264

"-*A


96/264

"6 *EA- of two sub4lassFf statements is a subset of the intersection of the classes:

`rdfs:4lass rdf:%A[A


97/264

"*A ame 4omment

rdf:type *elated a resource to its class

rdfs:sub4lassFf %ndicates membership of a class

rdfs:sub7ropertyFf %ndicates speciali=ation of properties

rdfs:domain domain class for a property type

rdfs:range range class for a property type

rdfs:label 7rovides a human"readable version of a resource name.

rdfs:comment Bse this for descriptions.

rdfs:member member of a container.

rdf:first The 3rst item in an *A list. lso often called the head.rdf:rest The rest of an *A list after the 3rst item, called the tail.

rdfs:seelso resource that provides information about the subect

resource

rdfs:isAefined!y %ndicates the namespace of a resource.

rdf:value %denti3es the principal value +usually a string- of a property

when the property value is a structured resource.

rdf:subect The subect of an *A statement.

rdf:predicate The predicate of an *A statement.

rdf:obect The obect of an *A statement.

&. 6uadrilaterals+P- polygons+P-

8. polygons+P- shapes+P-

5. 6uadrilaterals +s6uares-

nd now from this $nowledge the following conclusions can be deduced:

&. polygons +s6uares-

8. shapes +s6uares-

5. 6uadrilateral+P- shapes+P-

The hierarchy relationship of classes is shown in igure ("9 for the simple related

classes +ontology- of shapes.

4onsider the range restriction that organi=ation charts can only include 6uadri"

laterals and suppose that we want to use s6uares in an organi=ation chart appli"

cation. Fur restriction actually prohibits s6uares from being used. The reason is

because there is no statement specifying that s6uares are also a member of the

polygon class. What we need is for a s6uare to inherit the ability to use the classof polygons. This is accomplished through *Aow the meaning can be used to process software. s a result, *A


98/264

"! *Euclear 7hysics is taught by 7rofessor


99/264

""*A uclear 7hysics 7rof.


100/264

% < contains certain triples

T2E> add to < certain additional triples

+where < is an arbitrary set of triples-.The entire set of inference rules can be found in the *A speci3cation.

simple *A inference engine example is presented in 4hapter J.

RDF and RDFS Li;itati9n8

The *A uses only binary properties. This restriction is important because we

often use predicates with two or more arguments. ortunately, *A allows such

predicates to be simulated.

nother *A limitation results from properties as special $inds of resources.The properties themselves can be used as the obect in an obect"attribute"value

statement. This Cexibility can lead to modelers becoming confused.

lso, the rei3cation mechanism is very powerful, but may be misplaced in

the *A language, since ma$ing statements about statements is complex.The *A promotes the use of standardi=ed vocabularies, standardi=ed types

+classes- and standardi=ed properties. While *A PD#"based syntax is well

suited for machine processing, it is not user"friendly.

To summari=e, *A is not an optimal modeling language. 2owever, it is an

accepted stan

sw-1 (thinking on the web)

Documents