Computational Intelligence（Concepts to Implementations）Part 1.pdf_P36-P40

Historical Views of Computational Intelligence 29 Darwinian evolution.In this new perception of evolution,it appears that natural selection and self-organization work hand-in-hand.That is, evolution natural selection self-organization It is the authors'opinion that the neo-Darwinian view of evolution tends to con- strain evolutionary computation to a supporting role in computational intelligence, while the incorporation of self-organization facilitates the viewpoint that evolution- ary computation is computational intelligence's foundation. Self-organization remains an active area of inquiry.See,for example,the works of Stuart Kauffman (1993,1995). It should be evident to you by now that adaptation and self-organization are intertwined,an idea that we return to at various points in this book.It should also be evident that we consider adaptation and self-organization to play important roles in computational intelligence.With our discussions of adaptation and self- organization complete,it is time to look at computational intelligence,starting with early work in the field. Historical Views of Computational Intelligence As is the case with adaptation and self-organization,there is no universally accepted definition of computational intelligence.In this section,we present views of com- putational intelligence by other researchers.As you will see,these views are not the same.In the next section,we present our view of computational intelligence.It is somewhat different from the views presented in this section. In an editorial in IEEE Transactions on Neural Networks,then editor-in-chief Robert Marks wrote,"Neural networks,genetic algorithms,fuzzy systems,evolu- tionary programming,and artificial life are the building blocks of CI."He further stated,"Although seeking similar goals,CI has emerged as a sovereign field whose research community is virtually distinct from AI(Marks 1993). David Fogel said in 1995 that CI generally describes "methods of computation that can be used to adapt solutions to new problems and do not rely on explicit human knowledge.” Walter Karplus of the University of California at Los Angeles,who was then pres- ident of the IEEE Neural Networks Council(NNC),offered the following comment at the June 2,1996,meeting of the ADCOM of the NNC:"CI substitutes inten- sive computation for insight into how the system works.NNs,FSs,and EC were all shunned by classical system and control theorists.CI umbrellas and unifies these and other revolutionary methods." Bezdek (1998),who has probably thought about computational intelligence more than most other researchers,asserts that computational intelligence is a proper subset of artificial intelligence but that artificial intelligence is not a subset of

Historical Views of Computational Intelligence Darwinian evolution. In this new perception of evolution, it appears that natural selection and self-organization work hand-in-hand. That is, evolution = natural selection + self-organization It is the authors' opinion that the neo-Darwinian view of evolution tends to constrain evolutionary computation to a supporting role in computational intelligence, while the incorporation of self-organization facilitates the viewpoint that evolutionary computation is computational intelligence's foundation. Self-organization remains an active area of inquiry. See, for example, the works of Stuart Kauffman (1993, 1995). It should be evident to you by now that adaptation and self-organization are intertwined, an idea that we return to at various points in this book. It should also be evident that we consider adaptation and self-organization to play important roles in computational intelligence. With our discussions of adaptation and selforganization complete, it is time to look at computational intelligence, starting with early work in the field. Historical Views of Computational Intelligence As is the case with adaptation and self-organization, there is no universally accepted definition of computational intelligence. In this section, we present views of computational intelligence by other researchers. As you will see, these views are not the same. In the next section, we present our view of computational intelligence. It is somewhat different from the views presented in this section. In an editorial in IEEE Transactions on Neural Networks, then editor-in-chief Robert Marks wrote, "Neural networks, genetic algorithms, fuzzy systems, evolutionary programming, and artificial life are the building blocks of CI." He further stated, "Although seeking similar goals, CI has emerged as a sovereign field whose research community is virtually distinct from AI" (Marks 1993). David Fogel said in 1995 that CI generally describes "methods of computation that can be used to adapt solutions to new problems and do not rely on explicit human knowledge." Walter Karplus of the University of California at Los Angeles, who was then president of the IEEE Neural Networks Council (NNC), offered the following comment at the June 2, 1996, meeting of the ADCOM of the NNC: "CI substitutes intensive computation for insight into how the system works. NNs, FSs, and EC were all shunned by classical system and control theorists. CI umbrellas and unifies these and other revolutionary methods." Bezdek (1998), who has probably thought about computational intelligence more than most other researchers, asserts that computational intelligence is a proper subset of artificial intelligence but that artificial intelligence is not a subset of

30 Chapter Two-Computational Intelligence the much more complex biological intelligence.Rather,he believes that biological intelligence is used to guide artificial intelligence (and thus computational intel- ligence)models of it.He also views computational pattern recognition as one of many subsets of computational intelligence.In Bezdek's scheme,biological intelli- gence is organic(carbon-based),while computational intelligence(and its subsets) and artificial intelligence are examples of machine intelligence and are thus silicon- based.He believes that some computational models lack biological equivalents. Now that we've briefly toured the historical views of computational intelligence, let's see how the concepts we discussed previously,adaptation and self-organization, fit into it. Computational Intelligence as Adaptation and Self-organization This section discusses the authors'view of computational intelligence,in which adaptation and self-organization play key roles.The authors have a different view with respect to several aspects of computational intelligence presented above. We assert that intelligence is manifested both in carbon-based and silicon-based systems,and sometimes in hybrids of the two.In fact,intelligence need not be lim- ited to systems based on carbon and silicon:Other substances are the active subjects of inquiry in fields such as molecular computing.It does not matter what kind of system produces the intelligence for it to exist. It follows that the statement that some computational models do not have bio- logical equivalents is irrelevant to this discussion.(It could be argued that compu- tational models implemented by humans have biological analogies since humans conceived of,designed,developed,and tested them.The validity of this statement, however,is also irrelevant.)What is relevant is that no distinction should be made between biological and nonbiological intelligence.Thus,we assert that statements arguing biological equivalency,one way or the other,are not relevant to the discus- sion of intelligence or computational intelligence. In this book,computational intelligence is defined as a methodology involving computing that provides a system with an ability to learn and/or to deal with new situations,such that the system is perceived to possess one or more attributes of reason,such as generalization,discovery,association,and abstraction.The output of a computationally intelligent system often includes predictions and/or decisions. Put another way,CI comprises practical adaptation and self-organization concepts, paradigms,algorithms,and implementations that enable or facilitate appropriate actions (intelligent behavior)in complex and changing environments. Computational intelligence systems in silicon often comprise hybrids of para- digms such as artificial neural networks,fuzzy systems,and evolutionary compu- tation systems,augmented with knowledge elements.Silicon-based computational

Chapter Two--Computational Intelligence the much more complex biological intelligence. Rather, he believes that biological intelligence is used to guide artificial intelligence (and thus computational intelligence) models of it. He also views computational pattern recognition as one of many subsets of computational intelligence. In Bezdek's scheme, biological intelligence is organic (carbon-based), while computational intelligence (and its subsets) and artificial intelligence are examples of machine intelligence and are thus siliconbased. He believes that some computational models lack biological equivalents. Now that we've briefly toured the historical views of computational intelligence, let's see how the concepts we discussed previously, adaptation and self-organization, fit into it. Computational Intelligence as Adaptation and Self-organization This section discusses the authors' view of computational intelligence, in which adaptation and self-organization play key roles. The authors have a different view with respect to several aspects of computational intelligence presented above. We assert that intelligence is manifested both in carbon-based and silicon-based systems, and sometimes in hybrids of the two. In fact, intelligence need not be limited to systems based on carbon and silicon: Other substances are the active subjects of inquiry in fields such as molecular computing. It does not matter what kind of system produces the intelligence for it to exist. It follows that the statement that some computational models do not have biological equivalents is irrelevant to this discussion. (It could be argued that computational models implemented by humans have biological analogies since humans conceived of, designed, developed, and tested them. The validity of this statement, however, is also irrelevant.) What is relevant is that no distinction should be made between biological and nonbiological intelligence. Thus, we assert that statements arguing biological equivalency, one way or the other, are not relevant to the discussion of intelligence or computational intelligence. In this book, computational intelligence is defined as a methodology involving computing that provides a system with an ability to learn and/or to deal with new situations, such that the system is perceived to possess one or more attributes of reason, such as generalization, discovery, association, and abstraction. The output of a computationally intelligent system often includes predictions and/or decisions. Put another way, CI comprises practical adaptation and self-organization concepts, paradigms, algorithms, and implementations that enable or facilitate appropriate actions (intelligent behavior) in complex and changing environments. Computational intelligence systems in silicon often comprise hybrids of paradigms such as artificial neural networks, fuzzy systems, and evolutionary computation systems, augmented with knowledge elements. Silicon-based computational

3 Chapter Two-Computational Intelligence output generation.These represent processes that include actions related to safety and survival.For example,the arrow from sensing to output generation could represent a person's reflex actions when touching a hot stove.The arrow from preprocessing and algorithms to output generation could represent reactions of someone who happens upon a rattlesnake while hiking.Each of the arrows passes through the outer shell of the world model (embedded knowledge). In addition to reactions,outputs of the preprocessing and algorithms node include processed data and clustering,which may be used as inputs for the adap- tation and self-organization node.Products of adaptation and self-organization include reason,as described previously,as well as prediction and decision.Note that it is quite possible to reason,predict something,or decide to do something without actually taking action.Only when the reason,prediction,or decision is implemented,resulting in an action on or communication with the environment, is intelligent behavior said to have occurred. Complexity is often described as an attribute of intelligence (see,for example, Fogel 1995 and Bezdek 1994);for a discussion of complex adaptive systems that is applicable to intelligent systems,see Holland(1992).In Figure 2.4,complexity may generally be considered to increase as we move from sensing through prepro- cessing and algorithms,and through adaptation and self-organization to output generation.A note of caution is appropriate here.Without a complete definition and characterization of complexity,and subsequent application to intelligent sys- tems,which is beyond the scope of this book,it may be premature to characterize systems that effect intelligent behavior as more complex than,say,sensing systems such as human sight. Stochasticity,or randomness,is also sometimes listed as an attribute of intelli- gent systems.It is somewhat uncertain whether the attribute should be represented as randomness,pseudorandomness,or chaos.(Note that computer systems cannot generate randomness,just pseudorandomness.)However it is represented,it seems to permeate many aspects of carbon-based intelligent systems,from basic biology to behavioral intelligence,as well as most silicon-based intelligent processes and systems. In the representation in Figure 2.4,nodes at the tails of arrows need not be subsets of those at the heads,and any node can provide input to the output generation node.For example,sensing is not necessarily a subset of preprocessing and algorithms.Furthermore,sensing can provide an input to output generation via reflex. The world model at the top center of the diagram(which includes data and knowledge)and the arrows going to and from it require additional explanation. For each of the four nodes (sensing,preprocessing and algorithms,adaptation and self-organization,and output generation)arrows run both to and from the world model,signifying a flow of "information"in both directions

Chapter TwomComputational Intelligence output generation. These represent processes that include actions related to safety and survival. For example, the arrow from sensing to output generation could represent a person's reflex actions when touching a hot stove. The arrow from preprocessing and algorithms to output generation could represent reactions of someone who happens upon a rattlesnake while hiking. Each of the arrows passes through the outer shell of the world model (embedded knowledge). In addition to reactions, outputs of the preprocessing and algorithms node include processed data and clustering, which may be used as inputs for the adaptation and self-organization node. Products of adaptation and self-organization include reason, as described previously, as well as prediction and decision. Note that it is quite possible to reason, predict something, or decide to do something without actually taking action. Only when the reason, prediction, or decision is implemented, resulting in an action on or communication with the environment, is intelligent behavior said to have occurred. Complexity is often described as an attribute of intelligence (see, for example, Fogel 1995 and Bezdek 1994); for a discussion of complex adaptive systems that is applicable to intelligent systems, see Holland (1992). In Figure 2.4, complexity may generally be considered to increase as we move from sensing through preprocessing and algorithms, and through adaptation and self-organization to output generation. A note of caution is appropriate here. Without a complete definition and characterization of complexity, and subsequent application to intelligent systems, which is beyond the scope of this book, it may be premature to characterize systems that effect intelligent behavior as more complex than, say, sensing systems such as human sight. Stochasticity, or randomness, is also sometimes listed as an attribute of intelligent systems. It is somewhat uncertain whether the attribute should be represented as randomness, pseudorandomness, or chaos. (Note that computer systems cannot generate randomness, just pseudorandomness.) However it is represented, it seems to permeate many aspects of carbon-based intelligent systems, from basic biology to behavioral intelligence, as well as most silicon-based intelligent processes and systems. In the representation in Figure 2.4, nodes at the tails of arrows need not be subsets of those at the heads, and any node can provide input to the output generation node. For example, sensing is not necessarily a subset of preprocessing and algorithms. Furthermore, sensing can provide an input to output generation via reflex. The world model at the top center of the diagram (which includes data and knowledge) and the arrows going to and from it require additional explanation. For each of the four nodes (sensing, preprocessing and algorithms, adaptation and self-organization, and output generation) arrows run both to and from the world model, signifying a flow of "information" in both directions

Computational Intelligence as Adaptation and Self-organization 33 World Model (embedded knowledge) Survival Culture Data Goals Available Resources Values Adaptation Strategies Figure 2.5 An expanded view of the world model. The sizes of the arrowheads are meant to very roughly reflect the relative quantities of the flows.For example,the flow from sensing to the world model is much greater than the flow to sensing from the world model.And,as we move from the sensing node through preprocessing and algorithms,and then through adaptation and self-organization to output generation,a greater proportion of the flow comes from the world model to the node. Figure 2.5 is an expanded view of the world model,within which some of the categories of "information"are stored.Note that the world model is dynamic,con- stantly being revised and updated.In Figure 2.5,the knowledge complexity generally increases moving from left to right (keeping in mind the previous note of caution about complexity).Only a few components of the model are given. The diagrams in Figures 2.4 and 2.5 are simplistic,but they are meant to convey the authors'belief that there should be no distinction between carbon-and silicon- based intelligence.A system simply possesses one or more of the attributes shown in the figures,and the actions on and communications to the environment are intelli- gent to some degree,depending on the system attributes. So,where's the computational intelligence?In accordance with our earlier def- initions,it resides primarily in the adaptation and self-organization node.We also believe that elements of computational intelligence can be found in the preprocess- ing and algorithm node and in the output generation node.As represented,com- putational intelligence is buried deeply in the core of the system,be it biological or machine,perhaps the furthest from the interface with the environment.It is an area in which developments are occurring that will lead to exciting new analytical tools. At the risk of oversimplifying the concept of computational intelligence as illus- trated in Figure 2.4,we extract the portion of the figure most closely associated with computational intelligence and depict it with Figure 2.6.This prompts another definition,as follows:Computational intelligence comprises adaptation and self- organization using processed data and embedded knowledge as input and produc- ing predictions,decisions,generalizations,and reason as output.The embedded knowledge resides within the system,while the processed data originates outside the system

Computational Intelligence as Adaptation and Self-organization Figure 2.5 An expanded view of the world model. The sizes of the arrowheads are meant to very roughly reflect the relative quantities of the flows. For example, the flow from sensing to the world model is much greater than the flow to sensing from the world model. And, as we move from the sensing node through preprocessing and algorithms, and then through adaptation and self-organization to output generation, a greater proportion of the flow comes from the world model to the node. Figure 2.5 is an expanded view of the world model, within which some of the categories of "information" are stored. Note that the world model is dynamic, constantly being revised and updated. In Figure 2.5, the knowledge complexity generally increases moving from left to right (keeping in mind the previous note of caution about complexity). Only a few components of the model are given. The diagrams in Figures 2.4 and 2.5 are simplistic, but they are meant to convey the authors' belief that there should be no distinction between carbon- and siliconbased intelligence. A system simply possesses one or more of the attributes shown in the figures, and the actions on and communications to the environment are intelligent to some degree, depending on the system attributes. So, where's the computational intelligence? In accordance with our earlier definitions, it resides primarily in the adaptation and self-organization node. We also believe that elements of computational intelligence can be found in the preprocessing and algorithm node and in the output generation node. As represented, computational intelligence is buried deeply in the core of the system, be it biological or machine, perhaps the furthest from the interface with the environment. It is an area in which developments are occurring that will lead to exciting new analytical tools. At the risk of oversimplifying the concept of computational intelligence as illustrated in Figure 2.4, we extract the portion of the figure most closely associated with computational intelligence and depict it with Figure 2.6. This prompts another definition, as follows: Computational intelligence comprises adaptation and selforganization using processed data and embedded knowledge as input and producing predictions, decisions, generalizations, and reason as output. The embedded knowledge resides within the system, while the processed data originates outside the system