Permanent Link to This Item

Copyright © 2008 IEEE. Reprinted from IEEE Computational Intelligence Magazine, 2008; 3 (1):54-63 This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Adelaide's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to [email protected]. By choosing to view this document, you agree to all provisions of the copyright laws protecting it. Research Zbigniew Michalewicz, University of Adelaide, AUSTRALIA, Frontier and Matthew Michalewicz, SolveIT Software, AUSTRALIA Machine Intelligence, Adaptive Business Intelligence, and Natural Intelligence Machine Intelligence ligence by replicating humans by are silly, the system hardly deserves to ost people recognize Larry directly creating rules to follow or be called “intelligent!” Note also, that Fogel for his work on evolu- creating the neural connections). the term “appropriate action” implies Mtionary programming; today, It is interesting to observe that Larry optimization, as usually the system evolutionary programming is consid- Fogel identified three key elements of should take (or recommend) the best ered one of the early branches of evo- intelligence, namely: course of action. lutionary algorithms, together with ❏ ability to predict, Interestingly, the three components genetic algorithms, evolution strategies, ❏ ability to adapt, and of prediction, adaptation, and optimiza- genetic programming, and many ❏ ability to take appropriate action. tion constitute the core modules of other—sometimes unnamed—popula- Clearly, there is no need to argue adaptive business intelligence systems. tion-based techniques (Baeck et al., that prediction is important—without When we discuss adaptive business intel- 1997). However, it is important to this capability no system (including nat- ligence in the next section of the paper, remember that Larry’s main interest at ural systems) can be called intelligent. the connection with evolutionary pro- that time was in machine intelligence, The concept of gramming will be- and his work on evolutionary program- adaptability is cer- come apparent. ming was just to address some issues of tainly gaining popu- One additional machine intelligence. larity. Adaptability aspect of Larry One of the key observations of has already been Fogel’s research was Larry Fogel was that machine intelli- introduced in every- connected with the gence might be defined as the capabil- thing from automatic concept of so-called ity of a system to adapt its behavior to car transmissions “Valuated State meet desired goals in a range of envi- (which adapt their Space”®. The “Valu- ronments. Consequently, intelligent gear-change patterns ated State Space”® behavior requires prediction. An addi- to a driver’s driving approach provides a tional component of intelligent style), to running convenient structure behavior is adaptation (which is based shoes (which adapt © DIGITALSTOCK for assessing various on prediction, as adaptation to future their cushioning to a runner’s size and decision-making parameters in terms circumstances requires predicting stride), to Internet search engines that people are familiar with. Further, it those circumstances). The final com- (which adapt their search results to a allows individuals to apply subjective ponent of intelligent behavior is the user’s preferences and prior search his- relative importance weights and pro- capability of taking appropriate action tory). These products are very appeal- vides a mechanism for dealing with (Fogel et al., 1966). Consequently, the ing to individual consumers, because, degrees of criticality of parameters. The foundation for evolutionary program- despite their mass production, they are “Valuated State Space”® approach pro- ming research was to generate capable of adapting to the preferences vides a rank ordering of all possible out- machine intelligence by simulating the of each unique user after some period comes and rapid comparison of two or evolutionary process on a class of pre- of time. Finally, the ability to take more potential decisions to determine dictive algorithms (as opposed to the appropriate action is probably the most which is better. approach of generating machine intel- important component of an intelligent For a high-level overview of valuat- system. After all, if a system can predict ed state spaces the reader is referred to Digital Object Identifier 10.1109/MCI.2007.913389 and adapt, but all the decisions made (Michalewicz and Fogel, 2004); however, 54 IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | FEBRUARY 2008 1556-603X/08/$25.00©2008IEEE Authorized licensed use limited to: University of Adelaide Library. Downloaded on December 2, 2009 at 00:38 from IEEE Xplore. Restrictions apply. it is worthwhile to look at this approach as a general problem-solving approach. One of the key observations of Larry Fogel was that machine After all, the major steps in “Valuated intelligence might be defined as the capability of a system State Space”® imply some general to adapt its behavior to meet desired goals in a range problem-solving steps, which require understanding of the problem, rejection of environments. of intuition, building a model of the problem by defining the variables, con- straints, and the objectives. Thus the prediction before we can choose the none of these has explained how to systematic approach proposed for the quickest driving route. At work, we combine these various technologies Valuated State Spaces can be translated need to predict the demand for our into a software system that is capable into a general problem solving method- product before we can decide how of predicting, optimizing, and adapt- ology. When we discuss Puzzle-Based much to produce. And before invest- ing. Adaptive business intelligence Learning in the third section of the ing in a foreign market, we need to addresses this very issue. paper, the connection with “Valuated predict future exchange rates and eco- Clearly, the future of the business State Space”® will be visible. nomic variables. It seems that regard- intelligence industry lies in systems that These links with Larry Fogel’s less of the decision being made or its can make decisions, rather than tools work define the organization of this complexity, we first need to make a that produce detailed reports (Loshin paper. The next part of the paper pre- prediction of what is likely to happen 2003). As most business managers now sents the main concepts behind adap- in the future, and then make the best realize, there is a world of difference tive business intelligence, and the decision based on that prediction. This between having good knowledge and following part discusses the current fundamental process underpins the detailed reports, and making smart state of a new approach to learning, basic premise of adaptive business decisions. Michael Kahn, a technology called Puzzle-Based Learning. A short intelligence. reporter for Reuters in San Francisco, section on Larry’s and authors’ business Simply put, adaptive business intel- makes a valid point in his January 16, experience concludes the paper. ligence is the discipline of combining 2006 story titled, “Business Intelli- prediction, optimiza- gence Software Looks Adaptive Business Intelligence tion, and adaptability to Future”: Since the computer age dawned on into a system capable “But analysts mankind, one of the most important of answering these say applications areas in information technology has two fundamental that actually been that of “decision support.” questions: What is answer questions Today, this area is more important likely to happen in the rather than just than ever. Working in dynamic and future? And, what is the present mounds of ever-changing environments, mod- best decision right now? data is the key dri- ern-day managers are responsible for (Michalewicz et al. ver of a market set an assortment of far reaching deci- 2007). To build such to grow 10 per- sions: Should the company increase or a system, we first cent in 2006 or decrease its workforce? Enter new markets? need to understand about twice the Develop new products? Invest in research the methods and rate of the business and development? The list goes on. But techniques that enable software industry despite the inherent complexity of prediction, optimiza- in general. these issues and the ever-increasing tion, and adaptability ‘Increasingly load of information that business (Dhar and Stein, 1997). At first blush, you are seeing applications being managers must deal with, all these this subject matter is nothing new, as developed that will result in some decisions boil down to two funda- hundreds of books and articles have sort of action,’ said Brendan Barnacle, mental questions: already been written on business intel- an analyst at Pacific Crest Equities. ‘It ❏ What is likely to happen in the ligence (Vitt et al, 2002; Loshin, is a relatively small part now, but it is future? 2003), data mining and prediction clearly where the future is. That is the ❏ What is the best decision right now? methods (Weiss and Indurkhya, 1998; next stage of business intelligence.’” Whether we realize it or not, these Witten and Frank, 2005), forecasting two questions pervade our everyday methods (Makridakis et al., 1988),

Load more