Researchers at and working with IBM have created a computer and software that played and beat the best human players of the TV game show Jeopardy. This is an event that raised a lot of questions. For computer scientists it raises question along the lines of “how does it work?” and “How is this work applicable to other domains?” For the wider audience, but also critically important for computer scientist, there are also questions about what does this mean for artificial intelligence? Are computers catching up or even passing humans in intelligence? Is Watson smart? A genius? Is Watson actually thinking? Should be be afraid of the machines from the Terminator movies moving a step closer to reality? Ah, probably not.
I read a book by Ken Jennings (Brainiac: Adventures in the Curious, Competitive, Compulsive World of Trivia Buffs) which talks about Jeopardy and similar quiz games and the people who compete in them. It doesn’t take a genius to do well at these games. It does take some skill at understanding the questions and a good recall for huge amounts of data. But its not about doing great things with the data or making complicated decisions. Watson does things with a lot of work that people tend to do without thinking about them. Which brings up another point. It took 4 years, 25-30 people and millions of dollars to create a machine what people can do with a few minutes (or less) on instructions. The resulting computer/software can only play one game – Jeopardy. It would talk a lot of work to change Watson to play other games. People are much more versatile. Watson is not going to be making decisions based on its database anytime soon.
The big skill of Watson (for a computer) is to understand the natural language questions and then do a data search based on the results. People are still better at that. The Watson miscue on the Final Jeopardy question when Watson identified Toronto as a US city (more or less) while both humans correctly answered Chicago is a demonstration of that. So the computer is not yet that smart – not scary smart anyway.
What is thinking anyway? Is it just parsing a statement and doing a data search? I think not. I think it is more about things like making more choices based on insufficient data – making a jump of faith if you will. It is about how you use the data once it has been retrieved. Right now Watson is just a sort of search engine that is a lot better than average at parsing natural language questions. This makes is great for retrieving data but raw data is only as useful as the people making decisions based on it. I’m sure some look at a lot of computerized systems and saying “those machines are making decisions based on data” and they’d be right. To a point. The computers are making decisions based also on rules and guidelines that people have developed and programmed into the machines. They are not making the decision on, for example, what tolerances are ok for a manufacturing process. Or on what makes a flaw minor or major.
Thinking involves some level of creativity as well. I don’t see Watson having any creativity either. So it’s a great project. I think it advances the art of natural language processing and data mining. I am not so sure that we can say this is a super smart machine compared to humans. We’ve got a ways to go for that. What do you think? Are their other questions to ask and answer? I wonder what Watson would reply if asked questions about itself.