Then I watched the Nova episode on Watson last week, and my perspective has completely changed. If you missed the episode, I strongly encourage you to take time out and view it (I've embedded it at the bottom of this post).
What was so cool was to learn about how the Watson team used "machine learning" to "teach" Watson how to deconstruct questions to understand what is being asked, and then through many examples of what correct answers to questions are, used its vast database of raw unstructured data to piece together answers to these questions. (Yes, its Jeopardy, so actually it's reversed. Watson deconstructs statements, researches, then constructs questions.) The Nova folks did a fantastic job walking through the development history of Watson, showcasing the challenges the team faced, and describing the methods they used to overcome them. My only critique is that I wished they would have gotten more technical in their description of machine learning. The example of getting a computer to recognize the letter 'A' was great, but I'd love to understand more about the bits and bytes of how this works.
Watching the episode made a convert out of me. I'm completely amazed by what IBM has accomplished with Watson, and look forward to watching it compete on Jeopardy Monday - Wednesday this week (the DVR is all set to record). I guess I'd be surprised if Watson actually beats the two best Jeopardy contestants to ever play the game, but regardless of the outcome, the technology behind Watson holds amazing promise.
I have always been a sci-fi fan, and Isaac Assimov had me hooked at an early age on the idea of artificial intelligence. But where IBM talked about the potential real-life applications of Watson to things like medical treatment and financial analysis, my first thought was to applying this to human learning. Wouldn't it be amazing to have a system that learned about you--not to understand what movie you might want to watch next, or what product it could up-sell you on--but learned about your learning styles and objectives. Imagine it then dynamically assembling learning content from the vast resources of the internet in a manner that you would learn best. Imagine linking all of these systems so that the examples of how Joe and Susy and Ryan learned, could be referenced so that your system would get better and better at this over time.
Human learning has always been a keen interest of mine, so not surprising that's where my thoughts initially went. But then it hit me, what could possibly have more unstructure content than email and social sites like Connections?
What if Watson were put to use to read your email and prompt you on what you may want to do next, such as:
- Create an activity
- File the email in a folder that matches where you've put similar emails in the past
- Create a new folder and file the email
- Open a composite application to pull data to answer a question asked
- Start a reply to the email based on a template you've used in the past
- Recognize that this sender is not in your contacts, and create a pre-populated entry based on the person's email signature
- Create a new calendar entry based on the request to meet next Tuesday and populate it with the invitees
So yes, PLEASE, Watson, meet IBM Collaborative Software--I can't wait! (Oh yeah, and good luck on Jeopardy too.)