[Top] [All Lists]

[ontolog-forum] History of AI and Commercial Data Processing

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Tue, 23 Jun 2009 08:49:21 -0400
Message-id: <4A40CF51.408@xxxxxxxxxxx>
 From time to time, I let loose a little rant about the Semantic Web.
It's not that I'm opposed to the SemWeb.  On the contrary, the goals
are wonderful.  But I'm frustrated by seeing another repetition of
a phenomenon that has plagued the field of artificial intelligence
since its inception over half a century ago.    (01)

AI is dominated by brilliant people who are totally out of touch
with anything and everything that goes on in the field of commercial
data processing.  There is no question that many of their ideas
could, if properly implemented, revolutionize commercial systems.    (02)

But that little qualifier "if properly implemented" is the major
hurdle.  Most of the AI researchers over the past half century have
had little or no conception of how to implement anything that the
commercial market could use.  Conversely, few of the people in
the commercial field have any idea of how to use anything that
the AI researchers have produced.    (03)

Since I worked at IBM for 30 years, I gained an appreciation for
both sides of the divide, and I tried to do my part in helping to
bring them together.  But I was painfully aware of the great many
developments that widened it.    (04)

In the first decade of AI (mid 1950s to mid '60s), most of the
cutting edge research was done on IBM hardware.  Art Samuels,
for example, was a pioneer in both game-playing programs and
machine learning.  At 3 o'clock in the morning, he would run
multiple versions of his checker-playing programs on IBM 704
computers on the assembly line in Poughkeepsie.  Nat Rochester,
who had been the head engineer for the IBM 701, also did early
research on neural nets.  Nat was one of the four organizers of
the founding meeting for AI in 1956 (along with John McCarthy,
Marvin Minsky, and Claude Shannon).    (05)

The first split with commercial DP was also one of the most
brilliant.  John McCarthy designed the LISP language, which he
tailored to run on the IBM 704.  Although it ran on IBM hardware,
it was so remote from commercial languages like FORTRAN and COBOL,
that the LISP culture was disjoint from the commercial directions
and applications.  However, the fact that LISP ran on the same
hardware meant that some sharing was possible.    (06)

The total split occurred in 1964, when IBM announced the System/360
line of compatible computers.  Since the AI groups at MIT and Stanford
needed to buy new hardware, they agreed to buy compatible machines so
that they could share programs.  Around that time, the new DEC PDP-6
cost about the same as an IBM 360 Model 50, but it was as fast as
the Model 65.  Both MIT and Stanford bought PDP-6s, and most other
AI centers followed suit.  Of course, the MIT and Stanford systems
became incompatible with the first lines of code written for each,
but at least they belonged to the same culture.    (07)

The next opportunity to bring AI into the mainstream of commercial
systems came in the 1980s.  The Japanese Fifth Generation Project
spurred interest in AI around the world, and many commercial companies,
including IBM, thought that the time was ripe to bring AI into the
mainstream.  But most of the AI community was still running on DEC
equipment, which was not in the mainstream of commercial software.
The rest of the AI community had migrated to LISP machines, which
were so far removed from the commercial community that any kind
of collaboration was unimaginable.    (08)

But 1981 also brought the IBM PC, and many entrepreneurs developed
expert systems (and versions of LISP) to run on the PC.  At that
time, IBM France had implemented an outstanding version of Prolog
that ran on both the PC and the mainframes.  On the mainframes,
IBM Prolog reached the performance goals of the Japanese Fifth
Generation:  One MEGALIPS  (Millions of Logical Inferences per
Second), and it it was well integrated with relational databases.
It provided an excellent platform for building deductive systems
as adjuncts to commercial databases.    (09)

I tried to persuade IBM developers who were interested in AI
to build their software on top of Prolog, which was IBM's only
world-class AI product.  But instead, they decided to build
half-vast imitations of EMYCIN from Stanford and OPS5 from
Carnegie Mellon.  Those products failed miserably in the
marketplace, and they richly deserved to do so.    (010)

Some of the AI companies founded in the 1980s still survive,
and they continue to produce useful AI software.  The biggest
is Symantec, which had the goal of developing software for
natural language processing.  After a couple of years, the
VCs put their own people in charge, who made money by selling
the utilities that were intended to support NLP.  The most
famous AI company from the 1980s is Cyc, which survived on
research grants rather than profitable products.    (011)

In the late 1980s, the AI hype machine dried up, and the hype
shifted to Object Oriented Programming Systems (OOPS), another
technology from the 1960s -- Simula 67, which also inspired
Smalltalk and many other languages.  The AI gang responded
with the Common Lisp Object System (CLOS), which was ignored
by the commercial community because it looked like LISP.    (012)

In the 1990s, however, Sun produced a version of CLOS with
a notation that looked like C, and named it Java.  Fortunately,
Sun had enough people who understood both commercial DP and
AI software that they were able to make Java successful.    (013)

When I look at the Semantic Web, the original goals as outlined
by Tim B-L, and the kinds of software they have implemented,
it's "deja vu all over again".    (014)

Following is an excerpt from a previous note I sent to this forum.    (015)

__________________________________________________________________    (016)

My major complaint about the Semantic Web is that they ignored all
the development techniques that worked successfully for years, and
they failed to provide a migration path.    (017)

Following are some of the most egregious blunders:    (018)

  1. Ignoring the fact that every major web site is built on top
     of a relational database.  The major sites use big commercial
     databases.  Smaller sites are based on LAMP -- Linux, Apache,
     MySQL, and Perl, Python, or PHP.    (019)

  2. Building RDF on top of triples, instead of the SQL n-tuples.    (020)

  3. Failing to integrate their notations with UML diagrams, which
     include type hierarchies and various notations for constraints.    (021)

If the Semantic Web had addressed these three issues from the beginning,
it would have been integrated into the mainstream of data processing in
about 3 or 4 years.  Today, we would have seen some truly spectacular
applications.    (022)

The SemWeb still has a chance, but it has to be integrated with the
mainstream of data processing before it can become the mainstream.    (023)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (024)

<Prev in Thread] Current Thread [Next in Thread>