APPENDIX E

SITE REPORTS

Full accounts of the JTEC team's visits to three universities and three national projects are contained in Chapters 4 and 5, respectively, and are not repeated here in full. Addresses for these organizations can be found here.

Site: FUJITSU LABORATORIES LTD.
1015 Kamikodanaka
Nakahara-ku
Kawasaki 211, Japan

Date Visited: March 23, 1992

JTEC Attendees:

Feigenbaum
Nii
Schorr
Engelmore
Shelton
Kahaner

Hosts:

Dr. Shigeru Sato

Mr. Tomoharu Mohri

Mr. Jun'ichi Tanahashi

Mr. Hiromu Hayashi

Yasubumi Sakakibara

Fumihiro Maruyama

From Fujitsu Ltd: Susumu Murakami

Jun-ichi Komoda

Hitohide Usami

Fujitsu, the second largest computer company in the world, has been marketing AI tools, especially ESHELL, since 1984. They also have several products, e.g., for machine translation, and hardware for neuro-computing. ESHELL, which used to be the best selling ES tool in Japan, is now trailing behind Hitachi's ES/KERNEL. ESHELL began as a tool on mainframes using UTLISP. The market is shifting to products on UNIX-based workstations written in C. Fujitsu is trying to recapture the market with FORTRAN-based and COBOL-based tools, but its commitment to mainframes may be bucking the trend (not only in the ES area).

There are three groups involved with KBS:

The meeting began with a number of interesting comments, from director of the laboratory, Dr. Sato, whose association with KBS technology goes back to the late 1970s:

• The growth of AI has not been as big as expected, as measured in total sales, but is nonetheless growing. The technology does, however, contribute to the sale of Fujitsu hardware.
• There has been a rapid downsizing (in both the physical and economic senses). Consequently other technologies have gained in priority over AI, and manpower has shifted over to these newer technologies, such as multimedia and distributed networking.
• The principal barrier to growth is technological. They have difficulty in constructing knowledge systems for complex customer applications.
• ICOT is perceived as having had both positive and negative effects:
  • Positive: increased the level of basic research in computer science; developed parallel computing architectures, and increased the number of researchers.
  • Negative: emphasis on PROLOG was good for research, but bad for products; development of PIM machines did not benefit Fujitsu, which is now using 1024 SPARC chips in its AP1000 with a peak capacity of 8.53 GFlops (intended usage is in ultra high-speed scientific calculations).
• Knowledge acquisition is the primary technical barrier, and is the only interesting research topic in ES, but the problem may be too hard.

Our hosts at Fujitsu did a very good job in filling out Part A of the questionnaire. Highlights:

• They have shipped more than 3,000 ES shells, and have sold approximately 20 KBSs.
• They have about 240 systems that they use internally, and know of about 250 systems built by their customers, but could not categorize any of them in terms of operationality. Later in the questionnaire, however, they stated that about 20 percent of the projects started get as far as a completed prototype; and of those, 20 percent get to an operational system status. A best overall guess is that five percent of the reported systems are in routine use. The current success rate in fielding expert systems is better than five percent. Because of the experience base they now have, the success rate is somewhere between 75 and 95 percent. They are better able to select problems which are solvable by this technology.
• The largest percentage of systems developed for internal use are for planning or scheduling. Their most successful system is PSS, a production support system for planning assembly and test of printed circuit boards. It's a relatively small system, built on ESHELL, and runs on a mainframe. The system, which is in daily use, saves about one person per year by speeding up the planning time. A workstation version is under development.
• Fujitsu stresses integration of AI and conventional systems. The company now has ES tools written in COBOL (YPS/KR) and in FORTRAN (FORTRAN/KR), to support integration.
• 60 to 75 percent of the development costs of a system goes into the interface. Better graphical user interfaces (GUIs) are needed. That need has stimulated work on a GUI called GUIDEPOWER.
• Fujitsu worked with NKK on the blast furnace system, which is the largest application the company has been associated with.

Mr. Maruyama presented a novel approach to solving constraint satisfaction and optimization problems which purports to be better than integer programming (Maruyama, Minoda et al. 1991). He works at the Knowledge Processing Laboratory, which has a staff of about 40.

ADDITIONAL COMMENTS ON QUESTIONNAIRE:

Question 8:

Question 15:

Question 16:

Question 19 (main problems seen with existing technology):

1. GUI, as mentioned above.
2. Knowledge changes very fast in the real world (e.g., banking), hence the maintenance cost is too high.
3. There are few development tools, e.g., testing tools.
4. AI does not apply well to CASE.