Watson Day 3: Watson 2.0?

by Gabrielle DeMarco on February 16, 2011

Last night we really got to see Watson stretch his FLOPs. The machine is walloping the competition after a particularly brutal end to the first game of Jeopardy! The IBM Challenge. But, you could also see the potential for the second game to quickly spiral out of Watson’s control. A well-timed “Daily Double:” to Ken could be hard to catch up to and Brad seems to be getting the hang of beating Watson to the buzzer. The frustration is visibly building in both human players as Watson unknowingly tromps all over their impressive Jeopardy! pedigrees. A human would feel a little guilty at this point.
I was in the audience at EMPAC last night with over 800 fellow Watson fans. Between the massive screen and the shouts of encouragement from around the Concert Hall, it was a fun night of play. The panelists were in their element and we learned a lot more about how the Watson system actually functions.
One thing that came out during the panel is that Watson learns on its own. IBMers and Rensselaer graduates Chris Welty and Adam Lally shared some great insight into how the machine learns from it mistakes. Similar to the way a toddler learns how to walk through concerted trial and error (read: falling on face), Watson can learn after getting wrong answers. As an example, take Watson’s silly response to the “Final Jeopardy!” question under the category “U.S. Cities”. As a response, Watson uncertainly guessed, “What is Toronto?????” The question marks were an adorable, humanizing addition, but the response sent the audience into giggles. But, given what we now know about Watson, if this had been one question in a larger category about U.S. Cities, it is possible that Watson may have learned from the mistake during the course of play by analyzing the other correct answers offered by its opponents in the category and figuring out that all the responses were referring to U.S. Cities. This “thought” process would make a large Canadian city a very improbable response. Just to be clear. This wouldn’t be a programmer going in and tweaking the system. This would be a MACHINE doing the equivalent of thinking on its feet.
The ramifications of such a sophisticated level of computer thinking are really exciting, which brings us to our third and final night of the three-day viewing party at EMPAC. Tonight, the panelists will look into the future of Watson and its descendants. Outside of Jeopardy!, what can be done with this technology? Is Watson a step toward an AI that is intellectually on par with the human mind?
Joining Welty and Lally in the discussion tonight are Rensselaer professors Selmer Bringsjord, Deborah McGuinness, and Sanmay Das. Trust me when I say this: Given the subject matter and the profs involved, the conversation will get very interesting. The discussion tonight will be led by Rensselaer Vice President of Research Francine Berman. Before joining RPI, Berman was the High Performance Computing Endowed Chair at the University of California, San Diego, and director of the San Diego Supercomputer Center. She is also an expert in cyber infrastructure.

Bringsjord is head of Cognitive Science here and director of the Rensselaer AI and Reasoning (RAIR) Laboratory. He spends much of his time developing and studying ever-smarter computer intelligence. He literally taught Welty, Lally, and principle investigator for Watson, David Ferrucci, much of what they learned on the topic during their time at RPI. Among his accomplishments is “E” the “evil” computer, which is the first computer system designed to deceive its human user. But before you tag him an evil genius, he also developed “Eddie,” a 4-year-old child in Second Life who can reason about his own beliefs to draw conclusions in a manner that matches human children his age. As the mother of a particularly precocious four-year-old boy (he negotiated all morning on how many rounds of “Angry Birds” he could play on my phone this morning), I know that “Eddie” is no small task. As a teaser for the type of thinking Bringsjord is likely to bring to discussion of AI’s future, he recently wrote an entire paper entitled, “Honestly Speaking, How Close are We to HAL 9000?”

McGuinness is professor in the Tetherless World Research Constellation. McGuinness is among the world’s foremost experts on Web languages, ontologies, and ontology environments. The knowledge that went into creating Watson falls squarely in McGuinness’ playing field. And as an expert on the World Wide Web and its uses, it will be exciting to hear her talk about Watson’s future as it relates to the Web. Watson currently has to rely on the knowledge stored in its hardware. Imagine if it had access to the knowledge contained on the entire Web!

Das is a professor of computer science who studies what is known as collective intelligence. Collective intelligence (or sometimes collective stupidity) occurs when large groups of people act together to reach an outcome. One good example from Das’ research is financial markets. As we all know, something like the New York Stock Exchange involves the very complex interaction of millions of people as well as machines. What could a machine like Watson bring to such group? Would Watson be a good financial analyst? Could Watson uncover someone trying to game the system?

Join the final night of play and get a free glimpse into the future of computation.