Foreword
When I began my research career as a graduate student at Rensselaer Polytechnic
Institute in 1989, the phrase “knowledge graph” was not in use. The use of graphs,
however, as a notation for “knowledge representation” (KR) was quite common.
CLASSIC, the first real implemented description logic, was just being introduced
from Bell Labs, and although it had a linear syntax, the community was still in the
habit of drawing graphs that depicted the knowledge that was being represented.
This habit traced its history at least as far as M. Ross Quilian ’s work on Semantic
Networks, and subsequent researchers imagined knowledge to be intrinsic in the
design of Artificial Intelligence (AI) systems, universally sketching the role of
knowledge in a graphical form. By the late 1980s the community had more or less
taken up the call for formalisation proposed by Bill Woods and later his student,
Ron Brachman; graph formalisms were perhaps the central focus of AI at the time,
and stayed that way for another decade.
Despite this attention and focus, by the time I moved from academia to industrial
research at IBM’s Watson Research Centre in 2002, the knowledge representation
community had never really solved any problems other than our own. Knowledge
representation and reasoning evolved, or perhaps devolved, into a form of mathe-
matics, in which researchers posed difficult-to-solve puzzles that arose more from
syntactic properties of various formalisms than consideration of anyone else’s
actual use cases. Even though we tended to use the words, “semantic” and
“knowledge”, there was nothing particularly semantic about any of it, and indeed
the co-opting by the KR community of terms like semantics, ontology, episte-
mology, etc. to refer to our largely algorithmic work, reliably confused the hell out
of people who actually knew what those terms meant.
In my 12-year career at IBM, I found myself shifting with the times as a
revolution was happening in AI. Many researchers roundly rejected the assumptions
of the KR field, finding the focus on computation rather than data to be problematic.
A new generation of data scientists who wanted to instrument and measure
everything began to take over. I spent a lot of my time at IBM trying to convince
others that the KR technology was useful, and even helping them use it. It was a
v
评论1