I was a bit gleeful reading IBM's plans for a future in which they monetize an investment in "Watson" supercomputing. I suppose Watson's capabilities include meaning recognition but IBM's vision seems to be for better call center automation and (I would expect) better customer service automation. They are not thinking about automated-communication (auto-com) in general. They are not yet thinking about person-to-person automation - where I use a virtual agent to look for things online; or where my agent develops a crush on yours.
As I tried to say before, the secret of meaning recognition lies in the shared world which is the topic of the communication. The cool thing is that a computer program can have a model of that world and communication with the computer can be real. Not a parlor trick. My thought was that little bits of our world might get language-ized in open source and, after a while, different pieces of the world would coalesce into more and more complex domains of automated communication.
For simple example: buying a plane ticket. This is a situation where the "world" is a trip with dates, times, and places - an itinerary. A computer can know all possibilities and detect them in free-form communication with a person planning a trip, and have a real conversation. The point is not to replace Expedia, rather that this could be based on free form text and so could other things with a shared world.
Update: I want to extend a thought about how real communication is possible between a person and a computer if they share a world which is being discussed. Suppose, for contrast, that a robot was programmed to hand you a soft drink beverage. From your point of view getting a drink would be real. The communication does not need to be any more complicated. For one thing, the point of the communication might be to get it to give you a drink. For another, what more is there? The narrative which is exchanged in a communication must exist at both ends. But that is not necessarily complicated.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment