After reading media coverage for the IBM Watson challenge to the greatest past Jeopardy! champions, I couldn't help but recall my own experience as part of the IBM team behind the Deep Blue chess matches against the world champion at that time, Gary Kasparov.
Much has changed in the public’s perception of computers and their capabilities since the Kasparov vs. Deep Blue contests in 1996 and 1997. I distinctly remember the media’s coverage of the matches. It included cartoons depicting the computer as a fiend that would resort to all kinds of tricks to get back at humans for losing the first match, or excel at other human activities after winning the second match. In any case, the general sense was of the computer posing a “threat” to humans.
The reaction to the Jeopardy! event reflects the higher level of comfort the public has gained with technology. No longer is the computer seen as a threat, but rather a tool which can, for example, assist medical practitioners in their search for knowledge to produce a correct diagnosis.
Watson and Deep Blue are not only different in the hardware they use (strictly general purpose for Watson, as opposed to specialized hardware accelerators for Deep Blue), but also in their approach to exploiting artificial intelligence techniques. Watson uses natural language processing to try and understand a statement, then uses reasoning and machine learning coupled with a massive knowledge base to try and come up with the correct response, also “spoken” in English. (In Jeopardy! the “questions” are phrased as statements and the responses must be the corresponding “questions.”) On the other hand, Deep Blue was in essence a powerful game-tree search engine, interacting with the rest of the world through chess coordinates entered and displayed on a computer screen.
Machines are not infallible (remember, they run programs with bugs, sorry, “features”) and in the machine learning domain, quality and quantity of “training” is critical (how human is that?). Deep Blue lost the first match with Kasparov, and individual games in the second match. The only reason Deep Blue did not lose any games after beating Kasparov is because it was retired.
Discussion abounds on whether “intelligent machines” are a threat to humankind, and the media tends to zero in when one of them fails to dominate a human opponent (as in the recent congressman vs. machine Jeopardy! challenge in Washington, DC). Still, one message has remained the same since the 1996 event: the victor in these challenges has been humanity. Even though individuals may have lost in competition with a computer (a fact generation Z is familiar and comfortable with), the triumph of the machine also reflects the capabilities of the human intellect that built it.
In my next blog I will highlight the use of machine learning, which has become a more mature technology in recent years, in some of our CA Labs projects.