Below are links to a series of videos on IBM’s website that describe Watson and how it works. I highly recommend watching these, while keeping in mind the possible implications for job automation.
One thing I have tried to emphasize here and in my book, The Lights in the Tunnel, is that a true thinking machine is really NOT required in order to automate a large proportion of knowledge-based jobs: all that’s required is a sufficiently powerful specialized (or narrow) artificial intelligence application. I think Watson is a pretty strong indication that things are trending in that direction.
Watson is not based on custom databases of information, but rather uses machine learning techniques to parse and analyze information in “as is” form — in other words, books, magazines, encyclopedias, etc. (It’s worth noting that the real purpose behind Google’s massive book scanning project is reportedly to someday make the information accessible to an AI application, so IBM is certainly not the only company working on this problem).
Watson uses a massively parallel architecture that launches multiple (perhaps thousands) of queries using different algorithms simultaneously. The best answer—and a confidence level—is then selected using experience acquired via machine learning (in other words, by practicing with sample questions).
Here’s a quote from the IBM website: “Watson’s first test will be on Jeopardy!, but the real test will be applying the underlying data management and analytics technology across different industries.”
Here are the videos:
Building Watson (Includes explanations of how the technology works and practice matches showing improvement over time.)