 |
Leo J. Obrst |
Helping Computers to "Think"
Leo J. Obrst
May 2007
Leo J. Obrst took a winding path to his career as a MITRE artificial
intelligence engineer—one that included detours into fiction
and Fortran. Always fascinated by language and computers, he was
originally unsure how to combine his passions. He spent just a year
in college before taking a nine-year hiatus from school, working
in statistical analysis and software programming for a variety of
companies. He wrote poetry and short stories. Then Obrst picked
up a book he says changed his life.
"I really got my start after I read Doug Hofstadter's book Godel,
Escher, Bach: An Eternal Golden Braid," he explains. The book,
a wide-ranging meditation on creativity, language, and logic, poses
the question of whether languages follow formal rules. Through its
musings on mathematics, computer science, and the nature of cognition,
Obrst says he came to better understand the intriguing links between
linguistics and computing. This provided the inspiration for his
future work. "I decided to go back to school so that I could work
in artificial intelligence," he says.
Today Obrst is group leader of MITRE's Information Semantics group
and co-leader of a team of MITRE researchers who are developing
"Semantic Web" capabilities for the intelligence community and for
the military's Web-based command and control systems. The Semantic
Web, which will enable machines to perform rudimentary reasoning,
is predicted to be the next generation of the World Wide Web. While
it's far from becoming a reality for the average Internet user,
the emerging Semantic Web is already on its way towards transforming
military and intelligence systems, thanks in part to MITRE's research
and technical expertise in semantic technology, Obrst says.
"The Next Major Leap Forward"
"Many of us believe that semantic technology is the next major
leap forward in information technology," he says. "This is about
the computer's eventual evolution to the human level—machines
assisting humans at a higher level than ever before."
The existing Web presents information to people, while the Semantic
Web will make information meaningful to machines by attaching specific
meanings, or semantics, to the raw data. Semantics are expressed
in frameworks of definitions of terms and concepts called ontologies.
Obrst's team is focusing on the development of ontologies and semantic
rules—tools that help computers parse, compose, and interpret
the meanings of words and images. The team is also producing "taxonomies,"
or classifications, for various military and government terms to
support communities of interest. They are working with the Air Force
to produce a core taxonomy, or set of defined categories, for extending
Semantic Web capabilities to various Air Force technologies.
MITRE's military and government sponsors are increasingly focused
on finding and sharing sophisticated, sensitive data among many
disparate agencies and coalition partners, a task that will eventually
require far less human intervention than it does today. "The only
reasonable way to do this is with semantic technologies," Obrst
explains.
A Life Focused on Language
Obrst says that his past work in theoretical and computational
linguistics and his lifelong affinity for language led naturally
to his current role on the cutting edge of Semantic Web research.
When he eventually finished his undergraduate education, it was
with an English degree with a minor in linguistics. He then went
on to earn master's and doctoral degrees in theoretical linguistics
from the University of Texas at Austin, working all through his
studies at first in the university's artificial intelligence lab
and later at the Microelectronics Computer Corporation research
center. He studied Latin, Spanish, Hebrew, and Malayalam (a language
spoken in southwestern India) while working on his Ph.D.
From the world of academia, he went on to Intelligent Business
Systems, where he worked on developing natural language interfaces
to relational databases. Later, at Boeing, he worked as a senior
computational linguist, serving as system architect for a DARPA-funded
research project on semantics and intelligent agents for manufacturing
and design engineering, in addition to a natural-language understanding
system for project scheduling and massive database updating.
He also helped create the first commercial ontology department
in the world at an e-commerce company called VerticalNet in the
late 1990s, a project he recalls with considerable pride, though
he says it presented immense challenges.
"What we had to do was solve problems that ultimately were all
research issues," he says of the VerticalNet ontology work, which
was focused on business-to-business online commerce. "We had to
get these things to work even though universities were still doing
the underlying research on them." In this work, Obrst led development
of ontologies for VerticalNet's 58 disparate online buying communities,
creating the methodology used by the company to produce, implement,
and validate the ontologies.
Although the early-2000s dot-com bust ultimately claimed VerticalNet,
Obrst points to the work he did there as a critical precursor to
his MITRE research.
"It was an uphill battle in the beginning, to convince people in
the commercial world that these technologies had merit," he notes.
"Now, ontologies have achieved buzzword status. Today, MITRE's sponsors
can really see the value.
"I have a real sense of accomplishment here. Through MITRE, I've
worked with the Department of Defense, intelligence community, Army,
Navy, and IRS to promote semantic technologies," Obrst says. "Today
we're seeing real programs taking place in the federal government
that are solving hard problems."
And yes, Obrst is still writing poetry. "To me, poetry is like
assembly language programming in its conciseness, but expanded across
a thousand dimensions because it deals with so many nuances of sound,
syntax, and meaning," he says. "I'm using semantic technologies
of a different kind in my poetry."
—by Maria S. Lee
Related Information
Articles and News
Technical Papers and Presentations
Websites
|