![]() |
|||
![]()
|
![]() |
![]() |
![]() |
5. TRENDS AND OPEN ISSUESAs we have seen in previous sections of this chapter, the field of model-based reasoning (MBR) is very diverse and has similarly diverse applications. This diversity is at the same time a strength and a weakness. The diversity is unfortunately not supported by a set of clear taxonomies and methods for choosing an MBD technique for a certain purpose. The field does not present itself as a unified body of knowledge, apart from a common core where the abstract idea of a model occupies a central position. The MBD field is a collection of loosely connected techniques of variable degree of maturity. In this sense, the diversity is a weakness. Some attempts have been made to present a systematic methodology for subdomains, e.g., (Leitch et al., 1992) for model-based diagnosis, but these attempts have by no means been universally accepted. On the other hand, the diversity is a sign of a field in progress, where new ideas are still being created. The diversity can therefore be seen as a sign of dynamism. According to a more pessimistic view (Dague, 1995), no real intellectual advance has been made in the area of qualitative reasoning since 1984, when de Kleer and Brown (1984) and Forbus (1984) were published, and the same could be said for the more general area of MBR. According to this view, progress has not been in major principles or insights, but rather in smaller improvements in theory, and verification in gradually more ambitious applications. A general problem of MBD systems is high computational complexity and unpredictable computation time. If it is true that the main ideas behind MBD are understood, and the advantages of the MBD approach are equally accepted, what is keeping users from developing such systems en masse? One obvious obstacle is the effort and expertise required to build good models of new domains. It is maybe ironic that a technique that at the outset was supposed to alleviate the knowledge acquisition bottleneck, itself becomes victim of a similar "model acquisition bottleneck." The problem is being attacked along several dimensions:
We have briefly described a number of applications of MBR (and MBD). While a number of impressive applications have been built, their absolute number is still small. The "killer MBR application" that will lead to massive deployment is still missing. It is possible that no such application is imminent, and that MBR will remain a technology reserved for a small number of sophisticated applications. Another view, which is plausible, is that with the emergence of improved modeling support (see above), the cost of applying MBR will be lowered to a point where MBR becomes much more practicable. In the short run, better tools is a key success factor for this scenario to unfold. For example, the PRIDE project is developing a toolkit for easy deployment of MBD systems in the process industry (PRIDE, 1996). In the longer run, it appears that intelligent systems of the future must be model-based in one sense or other, whether we call it AI or not. Modern interconnected society will require highly sophisticated computer systems, for example, in massive industrial plant supervision, coordination of large transportation systems, monitoring of ecosystems on a regional and global scale, autonomous control of deep-space missions, etc. We must expect that systems with the level of intelligence, flexibility, and robustness that will be required for such applications will need internal models of their domains, and the ability to reason about such models (Williams and Nayak, 1996). 6. SUMMARYWe have presented model-based reasoning (MBR) as a field of AI, centered around the concept of representing the external domain in an internal model, and being able to reason about this model in order to solve problems. We have seen that apart from this central idea, the field contains a very diverse set of approaches. No unifying theory or (small) set of techniques can at the moment be presented as the MBR approach. In order to get a detailed understanding of the various approaches, one has to go to the sources. There is a need to provide better guidance to someone interested in applying MBR, maybe on a per-domain or per-task basis. In spite of the lack of a unified theory, MBR has been put to practical use in many impressive real-world applications. The major category of application is model-based diagnosis. The benefits of such applications over more traditional rule-based expert systems, are the ability to reason about novel situations, less dependence upon expert opinion and experience, and higher flexibility and scope for expansion. If such systems have not yet reached their full potential, one reason is the cost and complexity of building models. Once this difficulty has been overcome, through better tools, model reuse, and automatic model building, there is every reason to expect that MBD will be a key technology in intelligent systems of the future.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |