FGCS Assessment
Gilles Kahn
INRIA Sophia Antipolis, FRANCE
not among the top
leaders- a slightly sectarian attitude: they have seen the Truth revealed, nothing
else deserves paying attention to.
The next remark has to do with something that has unfolded during
the project in the commercial world: artificial intelligence, as a business, has not
exploded as expected. Progressively - to the dismay of the many gold seekers in
Silicon Valley, but not to the surprise of true scientists - it has turned out that
the limiting factor in the development of AI is not hardware, not even software,
it is the capacity of human beings to model satisfactorily a larger and larger
number of problems that were not previously solved on the computer. So AI, and
expert systems have developed and matured, but not at the rythm of electronic
circuit technology. As a result, many companies have dropped out of the field
entirely and a company like Thinking Machines Corporation has fundamentally
redirected its marketing efforts away from AI. So it is certainly the case that
some AI applications are compute-bound -and you have worked on them-, but
the pressure to solve these problems is not drastically different of that of solving
a number of other scientific problems.
In view of the remarks above, my assessment is as follows. The FGCS project has
accumulated considerable experience on MIMD computing, in terms of programming
and architecture as well. This experience is probably unique. All methods that would
be applicable only for a small number of processors have been rejected off hand. This
is a very sound approach for basic research. In terms of software, you have designed
and implemented bold and elegant ideas. I believe that many of these ideas will be
useful, and used, outside the ideological context of FGCS_ The basic research that you
have performed has been published in the open literature, it is deep and durable, it
has earned you the esteem of many scientists around the world. Globally, my opinion
is very positive.
Now, I will turn to the aspect of your work that is closest to my personal area of
research.
Language Issues
A priori, all of the work of FGCS revolves around one language, KL1. KL1 is an
original construction. Aspects of KL1 are described in many papers. The implemen-
tation of KL1 must be fairly similar on all of your hardware platforms, otherwise you
would have a difficulty porting PIMOS, and applications. It would also be difficult
to train new users. Yet, I see no single report which is "the definition of KL1", that
I could read at leisure to form a precise opinion of the language. I think one of you
wrote that KL1 = FGHC + meta - control + convenient-things. That leaves a bit too
much room for my imagination. In fact, I am not totally certain, given my previous
- 69 -