Having realized that the world is a big, hairy place full of difficult problems, researchers pursuing the knowledge representation question made a strategic retreat into simplified, virtual worlds, mo
Having realized that the world is a big, hairy place full of difficult problems, researchers pursuing the knowledge representation question made a strategic retreat into simplified, virtual worlds, modeled inside computers. They attempted to design systems that could represent knowledge of a virtual world that was less messy than the one we live in; things like a box full of blocks which the computer could be asked to move around with a virtual arm. They did reasonably well - you could have little conversations, in limited English, with the controlling program. Ask it what was where in the box, how it got there, why it did things, and tell it to move the blocks around. The difficulty came in returning these systems to the real world.
All of the knowledge the systems had needed to represent in their neat and tidy little VR domains had a very specific context - the contents of the virtual environment and the things you could do in it. What researchers discovered was that, out in the real world, there is an enormous amount of information that doesnt seem to correspond to any context of interaction particularly well: common sense information about the way the world works generally. Even trying to capture this common sense knowledge, whatever it might be, much less have a machine process it seemed a pretty daunting prospect. Nonetheless, that is exactly what the researchers on the Cyc project have been doing for well over a decade now.