My take on it is that the entire language parsing system should essentially be looked at as a human machine interface, albeit potentially the most useful and important HMI ever, once it is perfected. It's still a very long way off from passing a Turing test; answering queries correctly is a very long way off from being able to construct and understand the flow of a natural human conversation. Although many people over the last 50 years have pointed out that the Turing test isn't the be-all and end-all of artificial intelligence; having an AI that can be mistaken for human isn't nearly as useful a breakthrough as having an AI that is unmistakably machine but understands natural language queries perfectly.
Like sclitheroe says, Deep Blue basically overcame the chess-playing problem through a massive opening book and sheer processing power, which is sort of unfortunate because it didn't represent a massive leap forward in AI. Any problem that has a finite solution can simply be solved with advances in memory and processing speed. Language is much more interesting because it's essentially an infinite problem. No matter how much data a computer has access to and how fast it can query that data, understanding language in a meaningful way will still be an amazing feat of AI programming.
|