Site: Hitachi Development Laboratory, Ltd.
Kawasaki-shi 215, Japan
Date Visited: March 23, 1992
Dr. Koichiro Ishihara
Dr. Motohisa Funabashi
Dr. Hiroshi Motoda
Dr. Keith Collyer
The JTEC team's hosts from Hitachi did not fill out the application section of the questionnaire in advance, but Mr. Yamanaka filled it out partially after the meeting and sent usa facsimile copy. Hitachi has developed 500 to 600 systems for customers. The company has sold on the order of 4,000 copies of its ES/KERNEL systems, half of that number in the last two years, to approximately 1,000 customers. Approximately 50 systems are in each of the categories of field test, completed prototype, or under development.
Hitachi was the first Japanese computer company to provide ES tools on a workstation, in C. The ES/KERNEL tool is now the largest selling tool in Japan. ES/KERNEL is actually a series of tools, characterized either by the platform on which it runs or the task for which it is specialized. Hitachi's latest tool is ES/KERNEL2. Of Hitachi's task-specific shells, ES/PROMOTE/W-DIAG is a diagnosis ES tool, built on ES/KERNEL and running on a workstation. ES/PROMOTE/PLAN is for planning and scheduling, and was built from scratch. ES/TOOL/W-RI is a rule induction tool, under development. It is a policy within Hitachi to build applications on top of Hitachi tools.
Hitachi is quick to add features to its tools. For example, fuzzy logic is a part of the shell. ATMS and case-based reasoning will be in the ES/KERNEL2 to be released soon. Hitachi's time-to-market for new methodology may well be shorter than any organization in the U.S.
Hitachi also sells, both internally and externally, a guide to building expert systems, called ES/GUIDE. It advocates a spiral model of system development combined with the waterfall model (used primarily for documentation).
In addition to the tools developed by the ES group, the Heavy Industry Division is selling a real-time process control shell (EUREKA-II) and an advanced plant operation system called APOS, which is built on top of EUREKA. These shells are often used by the Heavy Industry Division people to build applications for their customers. They have sold more than 2000 copies of EUREKA.
Hitachi does not have a central group responsible for ES development within or outside Hitachi. Each section is expected to know about and use ES technology as appropriate.
Mr. Okuide reported on the construction planning system that Hitachi developed for Okamura for shield tunneling. The system combines AI, relational databases, CAD and a reporting system.
Initially, the market was in banking and financial diagnostic systems, but this is being overtaken by industrial companies doing scheduling. Our Hitachi hosts also said that the largest application market now is in the insurance industry. Within the domain of planning and scheduling, job shop scheduling is the biggest market, accounting for 40 to 50 percent of the business, followed by process scheduling. This class of applications got started three years ago; a Hitachi engineer worked with a customer to figure out how to do their planning application on ES KERNEL after the customer had failed using a competitor's shell. Initially, the scheduling job took 17 hours, but eventually it was reduced to five minutes of workstation time. The system has saved 7 billion yen. Hitachi's people are familiar enough with scheduling problems so that they are able to divide them into four classes and have techniques to solve them. However, they still cannot do railroad schedules since that requires solving the bypass problem (putting locals on side tracks so that express trains can pass them). Up until now they have done most of the planning applications for customers, but now they are beginning to introduce courses so that customers can learn to do this themselves.
In a private communication, a Hitachi representative indicated that spending $500,000 for an application (as was done at Okamura Construction) was at the high end of what Hitachi's customers were prepared to spend on an application.
Our Hitachi hosts expressed optimism about the future for ESs. However, they find their customers want/require consulting services and, in general, this requirement limits their deployment of ESs.
With respect to advanced KBS research, Hitachi is working on acquiring knowledge directly from text, and on a product for multi-layered cooperative reasoning. It has a product under development (EXCEED3) for qualitative reasoning (!), and one for rule induction, a fuzzy reasoning version of ES/KERNEL. As second generation tools, Hitachi envisions task/domain-specific tools such as the ones mentioned above (ES/PROMOTE/W-DIAG, ES/PROMOTE/PLAN, ES/TOOL/W-RI). The company is working on an advanced ATMS add-on to ES/KERNEL, a case-based machine translation system (at SDL), and a CBR in ES/KERNEL2. Most research funding is internal, with some cooperative work with customers.
Dr. Motoda, who is regarded as one of the foremost researchers in AI in Japan, reported on his work on knowledge reformulation and concept formation using "imagistic" reasoning (also called "frustration-based learning"). There is no apparent line of technology transfer between Motoda's lab and the ES sales/support group.
Hitachi has a small division in Europe (Hitachi Europe Limited, about 15 people), maintaining technical awareness of Western developments, and providing development support for ObjectIQ (the western version of ES/KERNEL2-W). ObjectIQ is about to be released, and will be marketed in Europe by Hewlett-Packard.
Scheduling projects are Hitachi's new field of endeavor, and it is Hitachi's most important customer need. Hitachi's people feel that rule-based systems are not good enough and they need better technology. To this end, Hitachi Europe Limited (HEL) is working with LOGICA on WHISPA. This product uses standard scheduling technology which is categorized as local decision job shop scheduling. At the time of our visit, Hitachi expected a product in mid 1992. Test cases have demonstrated one to two orders of magnitude speed-up in generating schedules. Forty to fifty percent of Hitachi's customers want it. HEL is also working on a research project, called TOSCA, with Austin Tate at the University of Edinburgh. TOSCA uses blackboard technology in its research phase for flexibility in research prototyping. Areas to be incorporated in TOSCA are:
TOSCA will be the next generation tool, and Hitachi will not release it until it gets customer feedback from WHISPA.