AAAI recognizes Watson for what it can, and will do
|
Director, Watson
Technologies
Dr. Eric Brown |
The Association for the Advancement of Artificial
Intelligence (AAAI) awarded IBM Research’s Watson team the 2013 Feigenbaum Prize “for
recognizing and encouraging outstanding Artificial Intelligence research
advances that are made by using experimental methods of computer science.” IBM
Watson Technologies Director Dr. Eric Brown, and his colleagues Drs. Chris
Welty and Jennifer Chu-Carroll accepted the award, and donated the monetary
portion of the prize to Wikimedia, on
behalf of the team.
Brown, Welty, and Chu-Carroll have spoken extensively about
Watson since its Jeopardy! win in 2011, with Brown most-recently creating a TED-Ed lesson on cognitive computing
that featured Watson’s latest work. They answered a few questions about their
work in AI, what’s next for Watson, and what it means to win the Feigenbaum
Prize.
Watson now works in healthcare and customer service. What
is the team “teaching” Watson, today?
|
Dr. Jennifer
Chu-Carroll |
Our team continues to focus on the medical domain to drive Watson’s
improvements. We like working in this area because one, it has important
societal benefits, but also because of how it forces us to think about how to
develop Watson’s technology.
Healthcare problems are complex. And every health issue a
patient deals with has years’ worth of background information associated with it.
Physicians can’t have all of this information assimilated into their brains. Watson
relieves them of that cognitive burden by accessing that information – helping
them make more informed decisions.
As we develop new and better ways to support how a doctor
makes a decision, we’re also thinking about how the system can replicate across
other domains. Because new domains require system modifications, a big part of
what we’re doing is coming up with automatic techniques that adapt to these new
domains.
|
Dr. Chris Welty |
Remember, when Watson played Jeopardy, while the clues
reached across subjects and required understanding of wordplay, ultimately it
needed just one correct answer. Applying the technology in something like healthcare is actually a dialogue. Watson produces possible answers, ranked in order of which ones it thinks are most correct, surfaces evidence of why it thought of those answers and can incorporate feedback from the user.
It’s an interesting challenge for us to make Watson a more
natural problem-solving collaborator.
Watson harnessed an amazing amount of already-available computer
horsepower to play Jeopardy! What will be needed, from a systems
perspective, to advance AI?
One of the reasons Watson succeeds
is that it’s an architecture designed to allow the integration of different
kinds of analytics techniques. From a software engineering point of view, this
lets experts from all different backgrounds to contribute to the system. To
that end, it allows our team to be very creative.
This ability gets to one of the
reasons for earning the Feigenbaum Prize. The prize recognizes
Artificial Intelligence advances based on an experimental approach that
emphasizes architectures, systems, and applications to real world problems. We’ve
certainly followed that approach with Watson and believe that an experimental
approach, especially one that leverages big data and a hybrid of rule-based and
statistical techniques, will continue to be the best way to drive advances in
AI.
Edward Feigenbaum is a pioneer
in AI research. What does it mean to the team to earn this Prize? And why donate
the prize money to the Wikimedia Foundation?
We were pleasantly surprised. It’s
an honor, and particularly significant that the entire AI community is
recognizing Watson’s accomplishment. The fact that the Prize looks for advances
in AI based on an overall systems approach is an affirmation of our approach to
applying Watson to solve problems.
We chose the Wikimedia foundation
to recognize the contribution that Wikipedia and WikiData made to the Watson system
that won on Jeopardy! The quality and coverage of Wikipedia was well suited as
a source for answers to Jeopardy!’s clues.
Labels: artificial intelligence, Feigenbaum Prize, ibmwatson, Wikimedia