It's the Name of the Computer Winning Jeopardy

What is Watson?

In the last two days, I've been paying attention to an IBM artificial intelligence project that has a computer playing Jeopardy against the two best Jeopardy champions ever (Brad Rutter and Ken Jennings) over the course of 3 days (Feb. 14-16). The intriguing aspect of this project is not that a computer can hold a wealth of knowledge and find answers to clues. No, the really cool part is that Watson is designed to process natural language - Jeopardy clues make sense to people because we can pick up on clever word play and puns, etc. This is different than a computer playing chess, because that's about weighing best moves and possible outcomes. Processing natural language is an incredibly difficult task. I was just thinking about how, whenever I have to call the customer service department of a bank or an airline, I immediately try to go to the "agent" option because it's a lot easier to talk (natural language) to a person than to "say out loud one of the following options," to which the computer usually responds, "I'm sorry. I didn't get that. Please say out loud one of the following options..."

This hasn't made big news here where we are, but I was still able to read some news articles and watch the Jeopardy episodes online.

As a language learner, this has me even more intrigued. We are able to communicate so much through how we phrase a statement, or through our tone of voice, or even through intentionally unspoken words that rely on the context of the situation. These are things that the a.i. field, I'm sure, is still many years away from being able to tackle. (Watson is unable to see or hear, but is fed the clues as text.)

Also, reading up on Watson adds to my fascination with the complexity of the human mind. Watson is made up of a cluster of 90 servers with a total of 720 core processors, holding 16 TB of RAM. Watson is not connected to the internet, but has been provided a massive amount of reference material to serve as its knowledge-base. Granted, Watson is beating Rutter/Jennings after 2 days - but still, that's the amount of computing power needed to process language and retrieve information somewhat like humans do.

On the side, I'm also curious about IBM's algorithm for Watson's wagering. The best part of Day 2 was when Watson, leading his opponents by $10K, hits a Daily Double and wagers $6435. How did he come up with that number? And then, in Final Jeopardy (where Watson got the answer wrong), Watson wagered only $947, safely maintaining the lead over his opponents, who got the answer right. But, why $947?

Looking forward to Day 3 of the challenge.