CHAPTER 5

SUMMARY AND CONCLUSIONS

Both the CIT·Xpert and TFRUIT·Xpert systems were generally well received by IFAS research and extension experts. Growers and county extension agents were also receptive to the diagnostic expert systems. However, this was not without exception as several county agents and even a few experts verbally expressed, they were "threatened by the application of advanced computer programs because in their opinion such systems had the potential to replace them in the work place". It was also observed that others felt that the systems practical applicability was too narrow and limited, and that the systems were to slow and not "user friendly". Yet, most people who reviewed and tested the expert systems were encouraged to find the application of artificial intelligence being applied in horticulture. Generally, all of the people who operated the expert systems believed that they had significant management and educational value. Many supported the concept that such systems could serve as a viable alternative diagnostic tool in instances when experts are not readily available as well as compliment expert diagnosis. Some suggested their greatest value may in fact be as an educational tool for students and possibly the master gardener training program rather than simply a diagnostic tool for growers or consultants. Additionally, many people favored the expert system use as an extension tool to increase the availability of expert diagnoses through county extension offices and personnel.

Expert system development is a time consuming and lengthy process. Knowledge acquisition has historically been identified as the bottleneck in expert system development because of the difficulty mimicking human reasoning processes during the of knowledge acquisition phase. The development of CIT·Xpert and TFRUIT·Xpert systems were primarily derived from descriptions of symptoms extracted from expert knowledge documents authored by two participating experts as well as personal discourse. Each expert knowledge document accurately described specific symptoms. The completion of the expert systems would not have been as timely without the personal initiative of the experts to review existing literature, and their understanding of the knowledge acquisition process. Without the experts willingness to devote a considerable amount of time and effort to identify symptoms in a coherent and consistent language, the knowledge acquisition phase would have been a much more laborious and cumbersome process.

Prior to the development of the TFRUIT·Xpert system, we identified areas of crop nutrition research that were lacking which led to the identification of new knowledge in the area of crop nutrition. We successfully identified cultivar-specific symptoms of nutrient deficiencies and documented nutrient ranges under greenhouse conditions in sand culture.

The matrix format provided a convenient way to organize the knowledge-base. New symptoms can be easily added, and old ones can be modified allowing the systems to be easily altered or updated. At the current prototype stage, the design is satisfactory but several further improvements can be made. It has been suggested that additional images need to be collected since the image database for both expert systems is incomplete. There was also considerable interest in further developing the systems so that additional specialist may have input into the knowledge-base. Therefore, a further review by additional outside experts may assist in identifying any additional problems with either system.

Results of the testing conducted on the expert systems indicated that the diagnoses of the expert systems generally agreed with the diagnoses of the participants in most of the cases tested. In some cases when the expert systems did not agree with the outside experts, there was not a consensus among the experts. The suggestions provided by the participants indicated that some changes should be made to the user interface, the selection of representative slide images and symptom text descriptions within symptom classes. Real-world cases were found to be very useful for testing of the expert systems. Unfortunately, real-world cases are limited by the seasonality of diseases, disorders and pest damage within the domain and are susceptible to spoilage or contamination during lengthy testing procedures.

Evaluation of the diagnostic expert systems also showed that nonexpert end users are able to make improved diagnoses using the diagnostic expert systems. However, the output of the diagnostic systems as with all computer-based expert systems, depends upon the ability to recognize or characterize symptoms. Since diagnosis of diseases, disorders and pest damage relies heavily on symptom identification, we should question the ability of nonexperts to correctly distinguish the symptoms of disease, disorders and pest damage from similar symptoms caused by other stresses. Our testing indicates that nonexperts who did not use the expert systems were able to identify symptoms of diseases, disorders and pest problems but not as well as when they utilized the diagnostic expert systems. In some cases nonexperts that used the diagnostic expert systems were able to correctly identify symptoms of diseases, disorders and pest problems as well as the experts who did not use the systems.

The results of our testing indicates that we successfully developed two prototype diagnostic expert systems useful for diagnosing diseases, disorders and pest damage of citrus and six selected tropical fruit crops. Full screen color images enhance the diagnostic capabilities of the systems and hyperlinks to publications can aid users in finding information they require. The testing results indicate that most experts agreed with the reasoning methods, they agreed with the expert systems diagnoses, and that with further modifications and testing nonexpert end users could be able to utilize the diagnostic expert systems to perform near the level of human experts. The diagnostic expert systems developed adequately demonstrated the viability of expert system technology.

A great deal of time and energy has been invested by many people in the development of CIT·Xpert and TFRUIT·Xpert. The development of these systems was an ambitious effort that had not been attempted before, not only in terms of the amount of information collected but also in the way the developers relied on the collaboration and pooling of resources from numerous individuals. The release of the first version software prototypes aims to bring the citrus and tropical fruit crop industries up to date with current computer technology. As more knowledge is gained and a more complete understanding of the nature and requirements of endusers, the user interface and matrices can be re-examined and improvements can be made. Realistically, rather than these systems remaining static, they should slowly evolve. One of the advantages of computer software is that it is much more cost effective to archive and upgrade than conventional publications.

It is hopeful that this thesis project will stimulate the investigation into similar computer-based projects and therefore this project can serve as a benchmark to begin both an effort to develop additional computer-based diagnostic expert systems with applications in horticulture and to further investigate the feasibility of including additional multimedia features such as video and sound. Already, as a result of the development of CIT·Xpert and TFRUIT·Xpert some researchers as well as producers have suggested that similar diagnostic expert systems should be developed for other major horticultural commodities such as strawberries, temperate fruit crops as well as vegetable crops. The development of CIT·Xpert and TFRUIT·Xpert programs have shown computer-based diagnostic expert systems can successfully unify existing knowledge and new information by utilizing current computer technology to provide an effective alternative methodology for assiting in diagnosing production problems commonly faced by fruit crop producers.