探索智能：从机器人到认知科学

需积分: 9 62 浏览量更新于2024-07-17 收藏 9.49MB PDF 举报

"Understanding Intelligence.pdf" 本书《Understanding Intelligence》由Rolf Pfeifer和Christian Scheier合著，旨在探讨人工智能和认知科学领域的基本原理，提供足够的信息以使读者能够开始自己的智能机器人实验和项目。书中包含Alex Riegler绘制的插图和Isabelle Follath创作的漫画，旨在以生动有趣的方式呈现复杂概念。在人工智能领域，本书深入浅出地介绍了智能的本质，涵盖了从基础理论到实际应用的各种主题。作者讨论了如何通过机器模拟人类的认知过程来实现智能，这包括学习、感知、决策制定和问题解决等核心能力。此外，书中还可能涉及神经网络、机器学习、模式识别和自然语言处理等关键技术，这些都是现代人工智能系统的基础。认知科学方面，书中可能分析了人类思维和行为的心理学基础，如知觉、记忆、语言理解和创新思维等。作者可能对比了人类智能与人工智能之间的异同，探讨了如何将这些理解应用于构建更先进的智能系统。专家系统作为人工智能的一个分支，在书中也有所涉及。这部分内容可能讨论了规则引擎、知识表示和推理机制，以及如何利用专家知识来解决特定领域的问题。书中还可能涵盖了一些与人工智能伦理和社会影响相关的话题，如人工智能的责任归属、隐私问题、就业结构变化以及智能机器对人类社会可能产生的长期影响。此外，书中的参考文献和索引为读者提供了进一步研究的资源，帮助他们在人工智能和认知科学的广阔领域内深入探索。通过这本书，读者不仅可以获取理论知识，还能获得实践指导，为自己的研究或项目打下坚实基础。总结起来，《Understanding Intelligence》是一本全面介绍人工智能和认知科学的著作，适合对此感兴趣的学生、研究人员和工程师阅读，它将帮助读者理解和构建智能机器，同时启发他们思考智能的本质及其对社会的深远影响。

As originally planned, the book was to have two parts, one out-

lining the traditional view, the other analyzing what is wrong with

it, with a small section at the end on alternatives. After discussions

with many students, we became convinced that these are not the

main things students are looking for: They don't want to know

the old way of thinking and why it is wrong, they want to know the

new way of thinking and why it is right. This realization entirely

changed the book's character. It still contains a chapter, albeit a

short one, on the traditional approach: The classical way of think-

ing about intelligence is still dominant not only in scienti®c circles,

but also in people's everyday thinking (chapter 2). Most of us are

comfortable with the idea that the human is an information pro-

cessing system and that intelligence is located in the brain. What

other explanation could there be? Chapter 3 presents the reasons

why these kinds of explanations are insuf®cient and have led to

insurmountable problems.

The next part (chapter 4) provides the theoretical groundwork for

understanding the terminology and the various approaches dis-

cussed in the literature. This part may be dif®cult reading for some,

but it is absolutely essential to appreciating the rest of the book. In

it, we discuss the idea of ``complete agents,'' which are the crea-

tures we want to build. You may ®nd that not everything in this

chapter becomes clear at the ®rst reading. The best strategy in such

cases is to read on and return to this chapter again later. As part of

the basics, we have included an introduction to neural networks

(chapter 5), which you may choose to skip if you already have a

background in neural networks. Even in that case, however, we

would recommend that you at least skim the chapter because it

is speci®cally geared toward neural networks for autonomous

agents, a type of network not often covered in detail in the standard

literature.

The next section reviews the major approaches that have been

suggested in the literature, including the Braitenberg vehicles

(chapter 6), the subsumption architecture (chapter 7), and the

evolutionary approach and arti®cial life (chapter 8), as well as

dynamical systems, behavioral economics, and schema-based

approaches (chapter 9).

The section on principles of intelligent systems is one of the core

sections. It summarizes the insights of the ®eld, and the consensus

Ðto the extent that it existsÐin a compact, general way (chapter

10). Chapters 11 through 14 elaborate the principles governing

Preface

design of intelligent systems. They contain a lot of detail, and you

may want to skip them initially. It should present no problem to

the reader to continue with chapter 15 right after chapter 10.

Chapter 15 is a case study of how embodied cognitive science, as

expressed by the design principles, can be applied to human

memory. Memory is clearly one of the most intriguing phenomena

in the study of intelligence, which is why we felt it would be an

excellent candidate for showing the power of embodied cognitive

science for studying what is normally considered a high-level cog-

nitive competence.

The next section is about building and evaluating autonomous

agents, both real and simulated. Chapter 16 provides insights into

the intertwined and subtle considerations involved in the design

process. It turns out that there are no precise recipes for success-

ful design. Actually designing interesting agents, as robots or as

simulated creatures, requires a lot of creativity. Chapter 17, on

evaluation, is very important; embodied cognitive science is an

enormously dynamic ®eld, but there is a de®nite lack of rigor and

scienti®c method. So do read this chapter carefully and try to apply

it to your own experiments.

The last part of the book concerns future and integration. Chapter

18 outlines where the ®eld may go in the coming years. This

includes scienti®c developments, as well as technological devel-

opments and applications in industry and society. It has an inten-

tionally speculative character and we hope that this will add to the

®eld's attractiveness. Proposals offered range from such down-to-

earth suggestions as building more complex robots to amphibious

robots inhabiting the sewage systems of our large cities. Chapter

19, entitled ``Intelligence revisited,'' brings together everything we

have said throughout the book in a theoretical framework. Such a

framework can be viewed as a ®rst step toward a theory of intelli-

gence. We conclude by outlining some of the more important

implications of the insights gained on society at large.

Scope

This book presents all the background knowledge required for

understanding the fundamental principles underlying intelligence.

It also covers the reasons why the classical approach has failed.

Moreover, it provides enough detailed materials on all aspects of

physical and simulated agents (robot design, neural networks,

Preface

xvi

control architectures, learning algorithms, programming) so that

students can start doing experiments and projects on their own.

The entire setup is highly interdisciplinary. Connections between

disciplines are woven into the entire text. Thus, the book provides

a rich source of associations that will be valuable in particular to

the motivated and talented students with interests reaching beyond

one discipline. A glossary at the end should also serve as a useful

reference.

This book does not cover algorithms of classical arti®cial intelli-

gence. It is also not a systematic technical introduction into robot

vision or other classical topics of robotics such as motion planning.

This is due to its complete-agent perspective: its focus is on the

agent as a whole, on its behavior, rather than on individual com-

petences. This perspective has been the main organizing principle.

It is also the reason some topics that are treated in a single chapter

in other textbooks are distributed over several in this one. Learn-

ing, for example, is essential to neural networks, categorization,

sensory-motor control, and value systems, all of which are princi-

ples concerning complete agents and include aspects other than

learning.

This book is not about robot building; it is not about designing

circuit boards and controllers for robots. Though design issues

have a prominent place, it is not an engineering text. The book

covers a lot of ground from a conceptual perspective. This is

unavoidable. It re¯ects the nature of intelligence, which is multi-

facated and messy. Also, because it is the ®rst comprehensive book

about this ®eld, it includes materials you might disagree with.

Although we have tried our best to bring everything together and

merge it into a coherent framework, many issues have not yet been

settled in the ®eld of embodied cognitive science.

History of This Project

The whole enterprise that has come to fruition with this book

started roughly 10 years ago when we began to run into funda-

mental problems with arti®cial intelligence. In the mid-1980s we

had already been working with expert systems for a number of

years. Over time we realized, as did many others, that the technol-

ogy did not ful®ll its promises. Accomplishing what we proposed

turned out to be much harder than expected: Only a very few of the

projects we undertook ended up with systems that could be used in

Preface

xvii

everyday routine practice. The problems were not simply of prac-

tical nature, they were somehow insurmountable. At about that

time, Terry Winograd and Fernando Flores's seminal book Under-

standing Computers and Cognition was published. Although we

initially did not understand all the issues involved, we realized

immediately that there was something important about this book.

It tied in smoothly with long-standing criticisms of arti®cial intel-

ligence by Hubert Dreyfus, and with at least some of the points

Gerald Edelman and George Reeke had been making for some time.

The main point of criticism for all these critics was arti®cial intel-

ligence's exclusive information processing perspective. All argued

that viewing human intelligence as information processing is mis-

leading and does not provide the best type of explanation for it. We

took the Winograd and Flores book seriously and ran a seminar

about it. In the meantime, William J. Clancey, an expert systems

specialist at Stanford, had also started thinking about the founda-

tions of classical arti®cial intelligence. His thinking has been very

in¯uential for our endeavors.

While on sabbatical leave in 1990±91 at the Free University of

Brussels, in Luc Steels' Arti®cial Intelligence Laboratory, I (Rolf

Pfeifer) had plenty of time to think about the fundamental problems

in arti®cial intelligence. This was when I completely changed my

whole research program. I had become convinced that we needed

an entirely new approach to the study of intelligence, if we wanted

to make progress in really understanding it. Since then we have, in

our research, been dealing only with embodied cognitive science.

In 1991, a workshop on ``emergence'' was held in the beautiful

monastery of Corsendonk in Belgium, attended by most of the

players in the ®eld: Rodney Brooks of MIT; Bill Clancey of the

Institute for Research on Learning; John Hallam of the University of

Edinburgh; Stevan Harnad of University of Southampton; Leslie

Kaelbling of Brown University; Chris Langton of Los Alamos and

the Santa Fe Institute; Maja Mataric, now of the University of

Sourthern California, Los Angeles; David McFarland of Oxford

University; Tim Smithers, now of University of Navara, San Sebas-

tian; Luc Steels of the Free University of Brussels; Chuck Taylor of

UCLA; and Francisco Varela of the Ecole Polytechnique in Paris.

This meeting can be seen somehow as the founding meeting of the

still very young ®eld of embodied cognitive science. Roughly three

years later, Lin Chen of the Beijing Open Laboratory for Cognitive

Science invited Rolf Pfeifer to deliver a series of lectures on the

Preface

xviii

new approach to cognitive science. He was the ®rst to suggest that

these lectures be turned into a book. We would like to thank Lin for

this suggestion. Since the Corsendonk meeting, much progress has

been made. The ®eld has matured into a scienti®c discipline. It is

time for a comprehensive textbook.

Acknowledgments

Many people have helped make this book possible, starting with

Luc Steels, who invited Rolf Pfeifer to ®ll the Swift AI Chair at the

Free University of Brussels, and the members of his group, in par-

ticular Jo Decuyper, Bernard Manderick, Philip Rademakers, and

Johan Vanwelkenhuyzen, who were very patient in discussing over

and over again the same sorts of issues. Rolf Pfeifer would also like

to thank Rodney Brooks for inviting him to spend his recent sab-

batical at the ``Zoo'' at the MIT Arti®cial Intelligence Laboratory.

The ``Zoo,'' a creation of Rodney Brooks, is a colorful collection of

interesting people who don't ®t into any of the established cate-

gories. The stay at MIT has greatly contributed to this book. In par-

ticular, the many extensive discussions with Yasuo Kuniyoshi of

the Electrotechnical Laboratory in Tsukuba, Japan, a member of the

``Zoo'' at the time, have been highly pro®table. Then we would like

to thank the members of the AI Lab of the University of Zurich:

Andreas Aebi, Nick Almassy, Raja Dravid, Peter Eggenberger,

Hansruedi Fru

h, Bernd Goetz, Charlotte Hemelrijk, Michelle Hoyle,

Hiroshi Kobayashi, Dimitri Lambrinos, Lorenz Leumann, Lukas

Lichtensteiger, Christopher Lueg, Marinus Maris, Daniel Meier,

Ralf Mo

ller, Martin Mu

ller, Alex Riegler, Thomas Rothen¯uh,

Simona Rusnak, Ralf Salomon, Rene

te Boekhorst, Paul Verschure,

Erik Vinkhuyzen, and Thomas Wehrle. Without their contribu-

tions, these ideas could not have evolved the way they did. The

current members of the lab all invested enormous amounts of work

and creativity to make it work.

A big thanks goes to Norman Cook of Kansai University and to

Paolo Gaudiano of Boston University, and to three anonymous

reviewers for their critical reading and their extremely valuable

and constructive remarks. Their suggestions have helped improve

the text tremendously. Many students from our classes have also

supplied us with valuable suggestions. In particular, we thank

Marcel Altherr, Natalie Glaus, and Alex Schro

der. Isabelle Follath

deserves all the credit for the cartoons, without which the book

Preface

xix

剩余717页未读，继续阅读

clf2017

粉丝: 0
资源: 8

探索智能：从机器人到认知科学

On Intelligence.pdf

Big Data and Computational Intelligence in Networking-CRC(2018).pdf

Applied Artificial Intelligence_Where AI Can Be Used in Business 2018.pdf

polikar2012.pdf

秋风秋雨.pdf

Handbook.of.Practical.Logic.and.Automated.Reasoning,.Harrison,.CUP,.2009.pdf

Machine Learning Techniques in ADAS.pdf

机器人用英语怎么写.pdf

2019考研英语核心词汇.pdf

人工智能的顶级会议.pdf

最新资源