Learning Computer Plays Jeopardy: The Intelligence of Watson and What it Means
Previously published on Feb 19, 2011
The advent of Google changed how we organize and search for information on the Web, but even its advanced search engine technology is lacking when it comes to machines understanding what people really want. Since before the concept of artificial intelligence came into existence, scientists, inventors, and science fiction enthusiasts have dreamed of creating a mechanical companion for mankind. We even believe our society will eventually decode human intelligence and its many elements by incrementally building better machines. The capacity of machines to communicate in human terms is a vital step forward in that mission. One major leap forward comes from IBM's Watson, a computer designed to play Jeopardy.
Although a long standing popular misconception defines intelligence as a large knowledge base, true intelligence revolves around what an entity can do with the information it stores. It also involves the ability to adapt to novel situations and respond to outside influences. Being the first test for supposed computer intelligence, chess serves as a good example of where a person, who probably does not know every possible move, exhibits intelligence, yet a computer likely fails to show intelligence because it relies on static algorithms to select a winning series of moves. Of course, this is not to suggest the performance of Deep Blue over chess master Garry Kasparov was anything but a historic moment with powerful implications.
Furthermore, a game like Jeopardy pushes the limits of human memory and recollection, which is a problem easily solved for computers by adding more storage space, but Jeopardy also tests the intelligence of players by forcing them to decode often cryptic clues and select best-fit answers over a varying array of categories. Fortunately, the last few decades has yielded learning software, which we find implemented in common services that predict what music or books we might like with uncanny success. By showing a computer enough examples of a correct answer, computers can use learning software to eventually self-write algorithms that identity how a human would respond, i.e. learn what people would most likely say, write, think, or feel.
Because Watson has terabyte upon terabyte of data available to it and enough resources to wade through relevant segments of its memory very quickly, it has more data available to it than human Jeopardy players. The trick to winning Jeopardy comes from Watson's analysis of millions of Jeopardy questions, which enabled it to learn the language of the game, and programmed tweaks that allow it to learn from the successes, as well as the failures, of human players and itself. The result is a computer that can actually interpret and select a proper response in human language. The most impressive aspect of this marvel is that Watson can actually cope with puns and other poetic devices instead of relying on user input that is tailored specifically for a computer.
Mankind's greater struggles with technology result from our inability to directly communicate in plain human language. In order to build even basic computers, humans had to invent and learn machine languages that rely heavily on logic statements. Unfortunately, this has meant human expressions must be translated into machine language, which can be quite burdensome and inefficient, while we simply have not been able to communicate many human ideas to our electronic friends. This has inhibited machines from responding to our exact needs and left programmers to painstakingly develop creative ways to gap the limits of software. With breakthroughs like Watson, we see the very opposite where the computer is learning to speak in human.
Computers are extremely good at doing a lot of things at once, relatively quickly. That is, electronic hardware has a greater capacity to accomplish assigned tasks than human hardware. Recalling the original promise of search engines like Ask Jeeves, software has fallen short when answering our questions. Even Google, which is very user friendly, works only because computer literate people have learned to phrase our queries properly. If the Watson software can be adapted to communicate beyond the game of Jeopardy, it may well be the basis for a truly interactive search engine. In fact, Watson could be the first incarnation of an electronic assistant who can help us tailor every detail of our electronic life. Moreover, Watson may not be human intelligence, but it certainly is intelligent in a very practical sense.