Content uploaded by Daniel Dajun Zeng
Author content
All content in this area was uploaded by Daniel Dajun Zeng on Oct 18, 2014
Content may be subject to copyright.
Editor: Alun Preece
University of Aberdeen
apreece@csd.abdn.ac.uk
Global IS
areas Chinese AI researchers are studying and predict related
future activities.
Automatic geometrical theorem proving:
Beyond mathematical mechanization
In 1997, Wen-Tsun Wu, a mathematics professor and
an academician from the Chinese Academy of Sciences
(CAS), won the Herbrand Award for Distinguished Contri-
butions to Automated Reasoning, the highest recognition
for researchers in that field. Other recipients of this award
include pioneers such as Larry Wos, Woody Bledsoe, and
Alan Robinson. Wu invented Wu’s method in the 1970s,
initiating the use of algebraic geometrical methods to obtain
analytical solutions to complex mathematical problems.1
As Deepak Kapur, editor in chief of the Journal of Auto-
mated Reasoning, pointed out, “Wu’s work re-invigorated
the field of Automated Geometrical Theorem Proving.”2
The cornerstone of Wu’s method is Wu-Ritt’s zero de-
composition algorithm, which Wu and others have used
to develop a theoretical framework that scientists and en-
gineers can apply to solve a range of problems, including
algebraic-equation systems and differential-equation sys-
tems. Researchers have also applied Wu’s method and
related frameworks to areas such as intelligent CAD/
CAM, computer vision, and robot design. In 1990, Wu
founded the Center of Mathematical Mechanization Re-
search at CAS to pursue this wide range of research direc-
tions and application innovations.
We predict that in the next 50 years, researchers will ex-
tend mathematical mechanization based on Wu’s method,
applying it in a much broader set of mathematical studies.
In addition, Wu’s method will likely play an important role
in an increasing number of application areas. In particular,
we expect wide use of the intelligent CAD/CAM platform
based on Wu’s framework, and we predict that its commer-
cialization will bring about a new generation of CAD/CAM
technologies.
China is also investing intensively in robot research, so
we wouldn’t be surprised to see Wu’s method applied to
industrial, underwater, servant, or even lunar robots.
Intelligence science: Toward a molecular-
level understanding
Intelligence science intersects with a range of fields,
including brain science, cognitive science, neural science,
psychology, molecular biology, biological physics, mathe-
matical and physical sciences, computer science, and infor-
mation science. The highly complex nature of human brain
functions requires a cross-disciplinary approach to explore
and understand the nature of intelligence and cognition at
various levels, from molecules and cells to the entire brain
and its functions. Clearly, research methodologies and re-
sults from intelligence science and AI can mutually benefit
each other.
The Chinese National Basic Research Program (also
called the 973 Program) has funded several major projects
investigating brain-related research and intelligence science
(see www.973.gov.cn/English/ReadItem.aspx?itemid=147;
also see itemid=247, 159, and 240). Another recently com-
pleted project, headed by CAS academician Xiongli Yang,
is “Fundamental Research on Brain Functions and Major
Brain Diseases.” This project studied cell and molecular
mechanisms of basic neural activities and the mechanism
underlying the formation of several important brain diseases.
Another project, “Fundamental Research on Brain Develop-
ment and Adaptation,” headed by CAS academician Aike
Guo, focused on neural nutrition factors regulating neural
cell division, survival, migration, and growth, as well as the
development and adaptation of perception, memory, and
visual cognition. A third project, “Advanced Interdiscipli-
nary Research on Intelligence Science,” funded by CAS, led
to interesting findings in selective attention and perception
and resulted in new advances in brain-imaging technology.
In yet another project, headed by CAS academician Chaoyi
Li, researchers studied cats’ brains and revealed a new low-
level cognitive structure that consists of many small spher-
AI Research in China:
50 Years down the Road
Ruqian Lu, Chinese Academy of Sciences
Daniel Zeng and Fei-Yue Wang, Chinese Academy of Sciences and University of Arizona
MAY/JUNE 2006 1541-1672/06/$20.00 © 2006 IEEE 91
Published by the IEEE Computer Society
Year 2006 marks the 50th anniversary of the birth of
modern artificial intelligence research. Chinese re-
searchers have been conducting AI research for more than
four decades. Here, we sample some of the most promising
oids for processing complex images. This
structure differs from all known brains’cog-
nitive structures and has proven important in
image processing.3
The recently published call for proposals
from the Intelligent Information Processing
unit of the 973 Program substantially empha-
sizes cognitive science and brain science.
The goal is to further explore the nature
of human brain functions and intelligence,
develop the computational theories of cog-
nitive science and intelligent systems, and
investigate new theoretical and technological
foundations enabling next-generation intelli-
gence system design and development.
In the next 50 years, we expect China to
make major advances in intelligence science
research in the study of various types of brain
activities such as consciousness, attention,
learning, memory, language, thinking, rea-
soning, and even emotion. Some particularly
promising research directions include
• how brains integrate and coordinate
nerve cell cluster activities,
• how nerve cell clusters perceive, express,
transmit, and reconstruct visual signals
and awareness,
• how we can use experimental methods
such as nuclear magnetic resonance to
observe nerve cell cluster activities, and
• how we can develop and evaluate mathe-
matical and computational methods to
model and simulate nerve cell cluster
activities.
Other relevant research areas include mod-
eling the creative thinking process, study-
ing the mapping of thoughts to language,
and studying the coevolution of reasoning
and language skills.
Large-scale knowledge
processing: An open approach
In “Some Challenges and Grand Chal-
lenges for Computational Intelligence,”4
Edward Feigenbaum posed three major
challenges to future computer science
development:
• developing computers that can pass the
Feigenbaum test, a restricted version of
the Turing test in a given subject domain,
• developing computers that can read docu-
ments and automatically construct large-
scale knowledge bases to significantly
reduce the complexity of knowledge-
engineering efforts, and
• developing computers that can compre-
hend Web contents and automatically
construct related knowledge bases.
Researchers in China are tackling the last
two challenges. Although both are essen-
tially a massive knowledge-engineering
challenge, the difference between the two
is that the third challenge involves an open
environment. Openness typically refers to
• a lack of standardization in knowledge
representation and semantic understanding,
• the dynamic nature (that is, emergence and
disappearance) of knowledge sources, and
• contradictions, ambiguities, noise, incom-
pleteness, and nonmonotonicity in knowl-
edge.
Answering these challenges calls for
new methodological and technological
innovations. One ongoing project in China
addresses this issue. In 1995, Cungen Cao
proposed building a National Knowledge
Infrastructure; in 2000, the Chinese NKI
project officially started.5The CNKI’s
early effort, inspired by the Cyc project,6
had focused on developing a nationwide
network-based knowledge-service system,
serving scientific education, research, and
social services needs. This system has three
components:
• an internal platform providing the core
usable services and software libraries,
• external applications developed by part-
nering organizations in various applica-
tion domains, and
• a module interfacing with the outside
world (including the Web and Semantic
Web).
The CNKI’s knowledge base has approxi-
mately 3 million knowledge records; when
completed, it will contain approximately
100 million records. CNKI aims to provide
knowledge to anyone who needs it at any
time, anywhere, and to support community
knowledge, communications, and coordi-
nation needs.
The Internet has become an indispens-
able source of knowledge for people around
the globe. According to recent surveys,7
China has over 100 million Internet users,
and the number is still growing rapidly.
Closely related to Feigenbaum’s third chal-
lenge, despite the Internet and Web’s exten-
sive use and development in the past decade,
many critical and pressing challenges remain
in Web-based information management.
First, the effectiveness of Web-based in-
formation retrieval measured by recall and
precision is still low. Second, knowledge
management on the Web is still based on
semistructured Web pages as opposed to
semantics-driven knowledge items. Third,
current Web technology supports access to
raw information but not to processed, refined,
customized knowledge. For instance, after
you enter one or more keywords into a search
engine such as Google, the search presents
you with hundreds of thousands of poten-
tially relevant Web pages. Clearly, we need
more advanced knowledge-driven online
search and browsing technology to address
these challenges.
Such new technology should demonstrate
both high recall and precision and support
online knowledge processing and mining.
It should have adequate natural language
processing capabilities to process and inte-
grate potentially conflicting, uncertain, and
ambiguous knowledge. In addition, it should
be able to organize, edit, refine, and mine
data and knowledge and facilitate the transi-
tion from information to knowledge. Fudan
University computer scientists are developing
elements of this technology. They plan to
provide a generic online knowledge acquisi-
tion and management tool to users as part of
their Web environment.
Computer-facilitated art and
animation: From research to
industry
The Chinese animation industry is experi-
encing significant growth. Approximately
3,600 television channels nationwide broad-
cast cartoon programs, and another 50 chan-
nels dedicated specifically to cartoons are
expected in the near future (see www.
Chinanim.com/dh1 (in Chinese)). According
to a recent estimate from www.Chinanim.
com, the total air time of cartoon programs
exceeds one million minutes annually in
China. Such a huge demand is creating a
serious shortage of professional animators
(the current gap is approximately 150,000).
Nationally, more than 200 colleges and uni-
versities offer majors related to cartoon art
and production.
Computers are playing an increasingly
important role in the animation industry. In
particular, computer-facilitated animation
is becoming a necessity. The CAS Acad-
emy of Mathematics and Systems Science
92 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
MAY/JUNE 2006 www.computer.org/intelligent 93
has been conducting research in this area
since 1989 and is cooperating with the Bei-
jing University of Technology to commer-
cialize its research results.8
In the next 50 years, computer-facilitated
art and animation will find more extensive
and wider applications in the real world. For
example, a computer will be able to automat-
ically translate 80 percent of a natural lan-
guage script into animation. It will be able to
1. analyze the natural language script,
2. understand the story’s plot,
3. decompose the script into a detailed
scene-by-scene script, and
4. plan and execute animation produc-
tion according to the movie director’s
interpretation, including scenarios,
roles, actions, cameras, color, and
light planning.
Obviously, each of these steps confronts
many AI challenges, and many relate to
image reasoning and common-sense rea-
soning. Research in these areas could pos-
sibly benefit other applications as well.
Knowledge as a commodity:
From software to knowware
Although knowledge-based software
engineering has become a widely accepted
R&D area, knowledge is still largely embed-
ded in software products. Recent research
suggests that isolating knowledge from soft-
ware and treating the two as independent
entities for both research and system devel-
opment will bring about many advantages.9
This might also address many of the diffi-
culties that current software and intelligent
system development efforts face.
For example, following this isolation prin-
ciple, complicated software might be devel-
oped jointly by two teams: the software engi-
neering team and the knowledge engineering
team. Software engineers don’t need to
learn new domain knowledge when they
switch from one application domain to
another, and domain experts can play a more
active role in software development by con-
tributing their knowledge directly to system
development. Following this emerging
framework, knowware is an independent,
standards-compliant, and executable knowl-
edge module. An early example of knowware
was the Promis object, where domain models
(preliminary forms of knowware) are dealt
with independently from the information
management systems.10
In the next 50 years, we expect the IT
industry to comprise three parts: hardware,
software, and knowware. Knowware devel-
opment will exploit AI and software engi-
neering research. Similar to the establishment
of software engineering as a field, we expect
knowware engineering to emerge as a study
of scientific principles, engineering guidance,
and best practices for managing knowware
development. In effect, knowware engineer-
ing will be the large-scale-production form of
knowledge engineering. It will have its own
model of knowware development life cycles,
practical guidelines on industry-strength
knowledge acquisition, and suggestions for
efficiently maintaining and updating knowl-
edge and for resolving conflicts, ambiguities,
noise, redundancies, and incompleteness.
We don’t have space here to cover all
the important AI topics that Chinese re-
searchers are actively studying (including
machine learning, pattern recognition, and
emerging social computing and informatics).
For a comprehensive picture of Chinese AI
research, please refer to related Chinese jour-
nals, such as the Journal of Pattern Recogni-
tion and Artificial Intelligence (in Chinese),
the Journal of Computer Science and Tech-
nology (in English), the Journal of Software
(in Chinese), and the Journal of Computer (in
Chinese).
References
1. W. Wu, “On the Decision Problem and the
Mechanization of Theorem-Proving in Ele-
mentary Geometry,” Scientia Sinica, vol.21 ,
1978, pp. 159–172. (Republished as “Some
Recent Advances in Mechanical Theorem
Proving of Geometries,” Automated Theorem
Proving: After 25 Years, W.W. Bledsoe and
D.W. Loveland, eds., 1984, pp. 235–242.)
2. D. Kapur, “Geometry Theorem Proving
Using Hilbert’s Nullstellensatz,” Proc. Symp.
Symbolic and Algebraic Manipulation (SYM-
SAC 86), ACM Press, 1986, pp. 202–208.
3. W. Xu, Z.-M. Sheng, and C.-Y. Li, “Spatial
Phase Sensitivity of V1 Neurons in Alert
Monkey,” Cerebral Cortex, vol. 15, no. 11,
2005, pp. 1697–1702.
4. E.A. Feigenbaum, “Some Challenges and
Grand Challenges for Computational Intelli-
gence,” J. ACM, vol. 50, no. 1, 2003, pp.
32–40.
5. C. Cao et al., “Progress of National Knowl-
edge Infrastructure,” J. of Computer Science
and Technology, vol. 17, 2002, pp. 523–534.
6. D.B. Lenat and P.V. Guha, “Building Large
Knowledge Based Systems,” Representation
and Inference in the CYC Project, Addison
Wesley, 1990.
7. “The 17th CNNIC Chinese Internet State-of-
Art Report,” 2006, www.cnnic.net.cn/html/
Dir/2006/01/17/3508.htm (in Chinese).
8. R. Lu and S. Zhang, Automatic Generation of
Computer Animation, LNAI 2160, Springer,
2002.
9. R. Lu, “From Hardware to Software to
Knowware: IT’s Third Liberation?” IEEE
Intelligent Systems, Mar./Apr. 2005, pp. 82–85.
10. R. Lu and Z. Jin, Domain Modeling Based
Software Engineering: A Formal Approach,
Kluwer Academic Publishers, 2000.
Ruqian Lu is an
academician in the
Chinese Academy of
Sciences and a pro-
fessor of computer
science at the Acad-
emy of Mathematics
and Systems Sci-
ences, the Institute
of Computing Technology, and Fudan Uni-
versity. Contact him at rqlu@math.ac.cn.
Daniel Zeng is an
associate professor
in the University of
Arizona’s Depart-
ment of Manage-
ment Information
Systems and the
director of the
Intelligent Systems
and Decisions Laboratory. He’s also affili-
ated with the Institute of Automation at
the Chinese Academy of Sciences. Con-
tact him at zeng@eller.arizona.edu.
Fei-Yue Wang is a
professor in the
University of Ari-
zona’s Systems &
Industrial Engi-
neering Depart-
ment and the direc-
tor of the Program
in Advanced
Research of Complex Systems. He’s also
director of the Key Laboratory of Com-
plex Systems and Intelligence Science at
the Chinese Academy of Sciences. Con-
tact him at feiyue@sie.arizona.edu.