Professional Documents
Culture Documents
Hal l of F ame Academy of Engineering and American Academy of Arts and Sciences.
Engineering Knowledge
By Alun Preece
A
rguably more than anyone, Ed Feigenbaum was responsi- the technology in intelligent assistants
and software wizards that step users
ble for getting AI out of the lab and into the enterprise. through complex tasks. In the 1990s, there
was a great deal of interest in technical as-
The key insight behind what he termed expert systems was to pects of knowledge sharing, by means of
federated knowledge bases connected to
capture valuable problem-solving knowledge in a knowledge the Internet. Organizations remained
committed to the idea of capturing and
base, such that a machine could perform research and application development. The leveraging their valuable knowledge, and
automated inference on it. Typically, the limits of the technology were quickly iden- expert systems technology became one
knowledge was captured by means of rules tified, but nevertheless numerous useful element of the emerging sociotechnical
or descriptive frames. As a result, signifi- systems were deployed at all levels within knowledge management field. Effective
cant advances were made in techniques enterprisessome of which continue to be heuristic problem-solving techniques re-
for rich knowledge and information rep- used today. main an important tool within many en-
resentation and on efficient methods for In 1994, at the tail-end of the expert terprises, for example, for scheduling and
large-scale inference. Indeed, one of Fei- systems boom, Feigenbaums work was resource-allocation tasks. We can also see
genbaums key interests was the use of heu- recognizedjointly with that of Raj components of knowledge-based systems
ristic methods for solving computationally Reddyby an ACM Turing Award for technology and knowledge engineering
hard problems. Motivated in part to com- pioneering the design and construction of methods in diverse areas such as business
pete with Japans Fifth Generation Com- large-scale artificial intelligence systems, rules systems, database integrity manage-
puter project, Feigenbaum worked to alert demonstrating the practical importance ment, and Semantic Web ontologies. Ideas
companies and organizations to the poten- and potential commercial impact of artifi- from the expert systems field are also evi-
tial benefits of knowledge-based systems. cial intelligence technology. As professor dent in recent high-profile launches such
Feigenbaum coined the term knowledge emeritus at Stanford, Feigenbaum has fo- as IBMs Watson system and the Wolfram
engineering to refer to the incorporation of cused interest, as a Board of Trustees mem- Alpha computational knowledge engine.
knowledge into software systems in order ber of the Computer History Museum, Ed Feigenbaum will always be associated
to perform complex problem-solving tasks. on preserving the history of computer with the 1980s assertion that knowledge
Essentially, knowledge engineering repre- science, and with the Stanford Libraries is power. Thirty years later, a great many
sented the first widespread industrialization on software for building and using digital AI systems developers still agree.
of AI. The AI field was transformed dur- archives. Alun Preece is a professor of intelligent sys-
ing the years of the expert systems boom, Feigenbaums legacy is broad. We can tems in the School of Computer Science and
with significant investment both in basic recognize patterns and characteristics of Informatics at Cardiff University, UK.
J
ohn McCarthy is one of the fathers of our discipline. He or- was under construction, McCarthy
developed the Lisp language based
ganized the famous Dartmouth Conference, where the foun- on symbolic expressions and sym-
bolic evaluation. From the begin-
dations for the whole field were laid and the term AI was coined. ning, he designed it such that reason-
ing systems could make deductions
Not only did he define the goals of AI, he also developed and about programs. It is a language of
great elegancesmall and yet pow-
provided a broad spectrum of key nonmonotonic reasoning in AI and erful and defined with a clear and
tools and methods. logic programming; there are ap- formal semantics. Lisp did not only
His first seminal paper Programs plications in most areas of AI. In become the major AI programming
with Common Sense was published his 1969 paper Some Philosophi- language, it also was the starting
before his famous Lisp paper and is cal Problems from the Standpoint of point for numerous papers on the
still a pleasure to read about logical Artificial Intelligence, together with mathematical foundation of comput-
representation and reasoning from Pat Hayes he included topics from ing. John McCarthy was probably
1959. This was probably the first paper robot action and planning. It even ad- the first to give a formal mathemati-
in which logic as a representation for- dressed the topic of free will. This is cal proof of a compiler for arithmetic
malism was proposed. Other papers certainly the first detailed definition expressions.
followed on this topic, and by 1970, of situation calculus, which is used in In addition to all these contribu-
he developed a modal-logic-based se- so many variants for cognitive robot- tions to programming languages,
mantics of knowledgean issue that ics. This paper also advocated a logic logic, knowledge representation, and
is central in symbolic knowledge rep- of knowledge by discussing various planning, John McCarthy published
resentation and in Web Science today. nonclassical logics. In subsequent pa- many papers discussing and shaping
John McCarthy continued the pro- pers, McCarthy explored the topics the field of AI. It is hardly possible
gram he laid down with that paper, of robot consciousness and introspec- for any researcher in AI not to stand
and in the 1980s, he came up with tion, topics challenging for both ro- on his shoulders!
circumscription as a logical basis botics and knowledge representation. Ulrich Furbach is a professor of artificial
for common-sense reasoning. This In the 1950s, at a time when For- intelligence at the University of Koblenz,
work initiated various approaches to tran was just published and Algol Germany.
A
fter reading Vannevar Bushs seminal Atlantic Monthly ar- In 1968, Engelbart presented NLS in a live
demonstration at the AFIPS Fall Joint Com-
ticle As We May Think in 1945, Douglas Engelbart has puter Conference in San Francisco. Even to-
day, this demonstration is known at the the
spent decades creating better methods of human-computer interac- mother of all demos because it was so far
ahead of its time. Engelbart showed inter-
tion to advance what he described as our collective IQ. Similar active computing, the computer mouse, key
chord, hypertext links, computer-supported
to Bush, Engelbart realized that the post-war manipulation systems (computers) would cooperative work, videoconferencing, com-
world was becoming ever-more complex, and alter how people think, perceive, concep- puter graphics, and word processingall tech-
the problems we as a society would face would tualize, assess, create, and deploy solutions nologies that have become ubiquitous today.
become increasingly complex and urgent at an collectively. To achieve this goal, a team of The following year, SRI became the sec-
alarming rate. He reasoned that radically new people with backgrounds in psychology, ond site connected to the Arpanet. Due to
tools and strategies would be needed for ef- computer programming, computer engi- its work on computer-supported cooperative
fectively addressing these challenges and that neering, and organizational development work and hypertext, the ARC became an
computers could play an important role. His were assembled. In addition to developing online clearing-house for Arpanet resources.
research began with the 1962 report Aug- the tools, the team also experimented and In 1989, with his daughter Christina, Engel-
menting the Human Intellect: A Conceptual used the tools themselves, in the first major bart founded the Bootstrap Institute (now
Framework, funded by the Air Force Office achievements in interactive knowledge work. the Doug Engelbart Institute), which contin-
of Scientific Research. The report described Throughout the 1960s and 1970s, Engel- ues to support the development of enabling
interactive computing, and he conceptualized barts ARC evolved the Online System (NLS) technologies for promoting collaboration.
tools to help intellectual workers make deci- computer system, which facilitated the storage Engelbarts interest in the synergistic re-
sions. Specifically, Engelbart wanted to match and retrieval of electronic documents and the lationship between humans and computers
human capabilities with computer tools. The development of digital libraries. The docu- has played an important role in inspiring a
point where humans and computers meet is ments could be accessed from any connected generation of computer scientists. His 1968
called the graphical user interface. terminal in the lab. A shared electronic space demonstration paved the way for new in-
In 1963, Engelbart set up the Augmenta- was created for software journal entries. GUIs teractive technologies that had not previ-
tion Research Center (ARC) at the Stanford were designed to help users interact with in- ously been imagined.
Research Institute (now SRI International). formation displayed on the screen. Two new
Susan B. Barnes is a professor and the asso-
He was aware that moving from manual (pen, methods of interaction included a five-key, ciate director of the Lab for Social Comput-
pencils, and paper) to automated symbol one-hand chord keyset and the mouse. ing at the Rochester Institute of Technology.
Collective Knowledge
By Steffen Staab
A
t the foundation of AI there is the need to have knowl- a truly democratic momentum that
favored the effective small step under-
edge and to work with this knowledge in order to solve stood and appreciated by the many over
the fortune or wisdom of the few, lead-
our problems or accomplish something intelligent and exciting. ing to an unprecedented global uptake
of his technological innovations. Thus,
When I started working in AI in the early 1990s, one of the most the World Wide Web Consortium
(W3C), which he founded, has achieved
pressing issues was the bottleneck of available in text and media sources, but its immense influence not by any regu-
procuring this knowledge from experts we all have structured data and knowl- latory decision power, but simply by the
or from a limited set of documents. edge en masse, though often we are not trust that so many people put into it be-
Sir Tim Berners-Lee changed this situa- even aware of it. By putting data on the cause of the way it was conceived.
tion completely. Semantic Web, he conceived the drastic The Web is now an unprecedented
First, he recognized that people are next step that is necessary to both find social machine that is producing both
willing to provide knowledge for free implicit knowledge and gain insight into implicit and explicit collective knowl-
when they are given the right incentives the knowledge collected by the many. edge. Immense progress in AI, like the
and the right tooling. He developed for- The World Wide Web and the Se- recent success of IBMs Watson on the
mats (HTML), protocols (HTTP), and mantic Web are two, of his many, tech- game show Jeopardy (http://www.ibm.
services (Web server) that led to the nical contributions that Berners-Lee com/innovation/us/watson), has been
World Wide Web, a base of collected has given to the AI community that achieved by building on it.
knowledge that is bigger than I could have changed its playing field forever. To the luck of all of us, Berners-Lee is
have ever dreamed of. By its very princi- It is important to note that he did not not resting; he is working on Web Science
ples, it not only constitutes a completely just conceive the ideas, but he was a to better understand the socio-technological
unstructured aggregation of text and me- promoter, developer, and nurse to his machine that has built and is building the
dia, but through the participation of and brain children. In fact, if it would have Web, further changing what it means to
recommendation by all people with ac- been for the technological develop- produce and harness intelligence.
cess to computing networks, it became a ment alone, we might not see either a
truly collective source of knowledge long Web or Semantic Web today. Berners- Steffen Staab is a professor for databases
and information systems in the Department
before the term Web 2.0 was coined. Lee was also a creator of the social of Informatics and director of the Institute
Second, he recognized that all of us machinery that backed his technical for Web Science and Technologies (WeST)
not only have knowledge implicitly innovations. He built it up the with at the University Koblenz-Landau.
A
s an indispensable constituent of AI, fuzzy logic is a super- corresponding rules for consistent math-
ematical operations (fuzzy arithmetic).
set of conventional (Boolean) logic that has been extended In addition, Zadeh is credited, along with
John R. Ragazzini, in 1952, with having pi-
to handle the concept of partial truth, where the truth value can oneered the development of the z-transform
method in discrete time signal processing
range between completely true and completely false. As the creator and analysis. These methods are now stan-
dard in digital signal processing, digital
of a new field of mathematicsfuzzy set applications for everything from consumer control, and other discrete-time systems
theory and fuzzy logicLotfi Zadehs in- products, industrial systems, and opera- used in industry and research.
tellectual contributions are myriad. He is tions research to medicine, geology, and Lotfi Zadeh belongs to a world where
also known for his research in system the- physics. there are no boundaries limited to time
ory, information processing, AI, expert Because of the importance of the relax- or place. He really is best characterized
systems, natural language understanding, ation of Aristotelian logic, which opens as an internationalist. He is a fellow of
and the theory of evidence. His current re- up applicability of rational methods to the IEEE, American Academy of Arts and
search is focused on fuzzy logic, comput- majority of practical situations without di- Sciences, ACM, AAAI, and International
ing with words, and soft computing, which chotomous truth values, Zadeh is one of Fuzzy Systems Association (IFSA). He
is a coalition of fuzzy logic, neurocomput- the most referenced authors in the applied has published extensively on a wide vari-
ing, evolutionary computing, probabilistic mathematics and computer science fields. ety of subjects relating to the conception,
computing, and parts of machine learning. In the theory of fuzzy sets, he proposed us- design, and analysis of information and
In 1965, Zadeh conceived of the idea ing a membership function (with a range intelligent systems and serves on the ed-
that developed into what is now known covering the interval [0, 1]) operating on itorial boards of more than 60 journals.
as fuzzy logic, a model for human reason- the domain of all possible values. He pro-
ing in which everything, including truth, is posed new operations for the calculus of Derong Liu is a professor in the Institute of Au-
tomation, Chinese Academy of Sciences, and a
a matter of degree. The theory challenges logic and showed that fuzzy logic was a
professor in the Department of Electrical and
classical logics belief in absolute true or generalization of classical and Boolean Computer Engineering at the University of Illi-
false. Although initially met with disdain, logic. He also proposed fuzzy numbers as nois at Chicago. He is also the editor in chief
fuzzy logic is widely accepted today, with a special case of fuzzy sets, as well as the of the IEEE Transactions on Neural Networks.
Computational Linguistics
and Cognitive Science
By Ee-Peng Lim
N
oam Chomsky is a distinguished linguistics and cog- have an innate ability to construct
language grammar and generate sen-
nitive scientist. He developed grammar rules for natu- tence phrases. His view was consid-
ered highly controversial at the time
ral languages in the 1950s, laying the foundation work for lan- because it changed the way language
was studied and shaped the sub-
guage studies. The work also led to a large body of interesting sequent development of cognitive
science.
research on natural language process- expressive power. The Chomsky hier- Chomsky has authored more than
ing and subsequently computational archy has been used in linguistics to 70 books and 1,000 articles. At MIT,
linguistics. Chomsky proposed the study languages as well as by com- institute professorship is awarded to
well-known transformational gram- puter science researchers to study the most distinguished faculty. He
mar and authored the highly in- compiler and automata theory. became an endowed chair professor
fluential and cited book Syntactic Chomsky is the pioneer linking at MIT at the age of 37 and an insti-
Structures, which brought the com- linguistics with cognitive psychol- tute professor at 47.
puter science and linguistics fields ogy. His review of B.F. Skinners
closer to each other. He also created book Verbal Behavior pointed out
Ee-Peng Lim is a professor in the School
the well-known Chomsky hierarchy that language cannot be acquired by of Information Systems and director of the
of formal languages that classifies a stimulus response process. He in- Knowledge Discovery Systems Lab at Singa-
formal languages by their different stead proposed the view that humans pore Management University.
Raj Reddy
Raj Reddy is the Mozah Bint Nasser University Professor of Com-
puter Science and Robotics in the School of Computer Science at
Carnegie Mellon University. After receiving a BE from the Guindy
Engineering College of the University of Madras, India, an MTech
from the University of New South Wales, Australia, and a PhD in
computer science from Stanford University, Reddy served on the fac-
ulties of Stanford University and Carnegie Mellon University for
more than 40 years. He was the founding director of the Robotics
Institute and the dean of the School of Computer Science at CMU.
His AI research has concentrated on perceptual and motor aspects of
intelligence such as speech, language, vision, and robotics. In speech,
he and his colleagues developed large vocabulary connected speech
recognition systems (including Hearsay and Harpy) as part of the
DARPA speech understanding research program. Reddy served as
the chief scientist of the Centre Mondial Informatique et Ressource
Humaines and was awarded the Legion dHonneur by French Presi-
dent Mitterrand in 1984. He has received the ACM Turing Award,
Okawa Prize, Honda Prize, and Vannevar Bush Award. He is also
a member of the National Academy of Engineering and American
Academy of Arts and Sciences.
R
aj Reddy is one of the early and most influential pioneers true pioneer in exploring the role of
technology in service of society. He
in AI and computer science. He has been instrumental in not only created the first universal
digital library, a free, online digital
broadening the practical real-world impact of AI technology library that includes millions of
books, monuments, and archives that
through a relentless pursuit of designing and constructing various can be accessed anytime, anywhere,
but also continues advocating the
practical, large-scale AI systems. His and the graduates from CMU have need to provide anytime, anywhere
name is associated with many found- gone on to become leaders in the access to creative works of the hu-
ing and critical breakthroughs in the field. He was the founding director of man race. To help bridge the digital
perceptual and motor aspects of intel- the Robotics Institute, the first such divide, he was instrumental in estab-
ligence such as speech, language, vi- institute or department in the US. lishing the Rajiv Gandhi University
sion, and robotics. In speech, he and Reddys contribution in AI, how- of Knowledge Technologies in India,
his colleagues developed large vocab- ever, is far broader than speech and to cater to the educational needs of
ulary continuous speech recognition beyond improving a machines intel- the thousands of low income, gifted
systems, such as Hearsay, Dragon, ligence. His distinct visionary work rural youth.
Harpy, and Sphinx. Since then, the of using technologies for society and
CMU team has been responsible for humanity set him apart from other Harry Shum is corporate vice president at
many advances in speech research outstanding technologists. He is a Microsoft.
Probability, Causality,
and Intelligence
By Eric Horvitz
M
ethods for machine learning, reasoning, and decision Building on his studies of inference
and learning from data, in the 1990s,
making under uncertainty with probabilistic graphical Pearl turned his attention to the foun-
dations of science, exploring standing
models lay at the heart of a 25-year rolling revolution in AIand challenges with understanding causality
and causal inference. His work at the
Judea Pearl has been a scholar and visionary at the forefront of intersection of computer science, statis-
tics, and the philosophy of science shed
this revolution. Pearl has pursued prin- are specified (observations such as pa- new light on causality and statistical in-
ciples of intelligent reasoning by elu- tient symptoms) and probabilistic up- ference. In a series of papers, Pearl in-
cidating foundational representations dates are desired for sets of unobserved troduced clarity on topics that had long
and inferential machinery for reasoning or hidden variables of interest (hypoth- been rife with challenge and confusion.
under uncertainty. His research played eses such as the potential diseases af- His 2000 book Causality: Models,
a critical role in catalyzing a paradigm flicting a patient). Beyond quantitative Reasoning, and Inference has grown to
shift in computer science, resulting in an methods, Pearls work included efforts become yet another classic treatise.
effective and fruitful coupling of Bayes- to understand qualitative patterns of Pearl has been a veritable pioneer in
ian statistics and computer science. inference such as intercausal reason- pursuing a scientific understanding of the
In the mid-1980s, Pearl and his stu- ing, capturing how beliefs may change mechanisms underlying thought and intel-
dents were a fount of results on prob- in the face of evidence favoring alterna- ligent behavior and their embodiment in
abilistic graphical models. Pearl and tive hypotheses. computational representations and pro-
colleagues specified with clarity impor- Pearl provided an influential synthe- cedures. He has provided the computer
tant proofs and procedures on the rep- sis of important results and methods in science and statistics communities with a
resentation and manipulation of proba- his 1988 book Probabilistic Reason- precious cache of results, algorithms, and
bilistic independence within Bayesian ing in Intelligent Systems: Networks insights, and has mentored several genera-
networks. Pearl also developed useful of Plausible Inference. The book was tions of students along the way.
algorithms for performing probabilistic a revolutionary compendium of ideas
Eric Horvitz is a distinguished scientist at
inference within these graphical repre- and vision when it was published and it Microsoft Research, past president of
sentations. Such inference includes diag- remains a classic treatise on representa- AAAI, and a fellow of the AAAI and of the
nostic reasoning, where certain variables tion and reasoning under uncertainty. American Academy of Arts and Sciences.
M
uch of early AI research focused on heuristic search- physical environment and reasoning
about its own actions.
based problem solving and AI planning in constrained To attest to the influence of Nilsson
work in AI research and education,
environments such as microworlds. Nils Nilsson was a major con- most first AI courses include A* and
Strips at the beginning of the sylla-
tributor to these studies. He coinvented (with colleagues Peter Hart bus. Nilsson is also known for his re-
search in machine learning, pattern
and Bertram Raphael) one of the became the de facto representational recognition, and computational in-
best-known AI algorithms, A*, formalism for AI planning research telligence. His long, distinguished re-
which delivers high performance in and laid the foundation for new ways search career has been spent between
graph traversal by using heuristics. of applying theorem-proving tech- SRI Internationals Artificial Intelli-
A* possesses many desirable prop- niques to problem solving. Strips has gence Center and Stanford University.
erties. It will always find a solution been extended in several important In addition to conducting cutting-
if one exists. When given an admis- dimensions (such as to enable learn- edge AI research, Nilsson has pub-
sible heuristic function, one that will ing, incorporate conditionals, and lished a number of influential AI
never overestimate the cost of reach- handle probabilistic state information textbooks and a definitive history
ing from the current state to the goal and actions) to support complex prob- of the field. He has made important
state, A* is optimal and efficient. A* lem solving in a variety of application contributions to the AI research com-
and its more space-efficient varia- domains. Furthermore, it makes it munity as the AAAIs fourth presi-
tions have enjoyed success in a range possible to formally analyze a range dent and as a cofounder of Morgan
of applications going beyond AI. of AI problems from the point of view Kaufmann Publishers, which has
Jointly with his colleague at the of computational complexity. One of published many important AI works.
Stanford Research Institute, Richard the early applications of Strips was in
Daniel Zeng is a professor and Eller Fellow
Fikes, Nilsson also developed Strips, a Shakey, a mobile robot developed by
in the Eller College of Management at the
formal language to describe AI plan- Nilsson and his colleagues at the Stan- University of Arizona and a research pro-
ning problems, and related automated ford Research Institute. Shakey was fessor at the Institute of Automation in the
planning techniques. Strips quickly capable of interacting with a simple Chinese Academy of Sciences.
computingnow.computer.org