information processing as a whole and 
- the equal importance of inference and knowledge for knowledge processing. 

The other major result is the importance of parallelism.. Since a lot of software produced 
during the project was first coded in a sequential way, the speed-up by parallelization 
could be experienced in an explicit and dramatic way, i.e. near linear speed-up could be 
experienced in a number of cases. This was by far not happen to happen, so that the 
international research community is grateful to the Japanese researchers to carrying out 
this important experiment and achieve this encouraging result. It is parallelism which 
eventually enabled the project to meet the performance target of 100MLIPS (logical 
inferences per second) for execution of KL1. 

A third major point in my judgment is the ease of logic as a formalism for efficient 
production of reliable software. It is nearly unbelievable how much software was 
produced in about two and a half years written directly or indirectly in KL1. As one could 
see in the demonstrations no problems arose running these large systems. In order to 
appreciate this achievement in a fair way, one has to keep in mind that all this software is 
written for parallel execution. We all know how hard it is to code parallel programs, and 
in fact I know of no project anywhere in the world which has produced parallel software 
at such a large scale. Given the experience with conventional software production (even 
sequential, let alone parallel one) which obviously requires much more time for 
producing software with the same functionality, it is obvious at least to me that one of the 
results of the project is a proof for the claim that software production is enhanced by logic 
by orders of magnitude. 

In addition to these and many other important main results, there are obviously the many 
results of detail, available in many hundreds of published papers and operative systems. 
Whatever the exact number is, we all know from our daily scientic work how many of the 
results of the Japanese colleagues play an important role in our own research which 
would not be the case without the FGCS project. 


4. Evaluation of the projects hypotheses 

One might speculate whether the net results of the project could have been even better, 
would different routes be followed, a topic which I discuss briefly in the present section. 

First of all, betting exclusively on logic has been a real bargain in all respects as the 
discussed results demonstrate. The same is true for dealing with the problem in a 
vertically integral way, from hardware all along through to intelligent functions and 
programs. 

Some people argue that it has been a mistake to test the approach based on parallel logic 
only at such a late stage in the project. On the one hand, there is a point to this argument 
because so far the computing community became hardly interested in the details of the 
approach simply because they could be impressed only by attractive applications. On the 
other hand, how could one manage to demonstrate the taken approach without having 
completed the machines and the basic software? I think this is a shortsighted argument. It 
is one of the major virtues of the Japanese way of carrying out this project that such long-
range goals were undertaken and kept unchanged for such a relatively long period. 

Another issue of possible concern is the specialized nature of the PIM machines, built 
especially to run KL1 efficiently. Would not general purpose parallel machines (like the 
J-machine presented in an invited lecture at the conference) serve the same, if not a better 
purpose? I think this is a good question which cannot be answered at present in a fully 


					- 43 -