The Frame Problem
It is common for cognitive scientists to underestimate the magnitude of the frame problem. Surely it is merely a matter of learning. However, Dennett (1984, p153) remarks that the reason that AI forces the frame problem to the surface is that when we try to compute intelligent behaviour we start at zero. A robot has to be told everything; things that we (as intelligent beings) take for granted. We know literally trillions of things (for example, that knives don't dissolve on contact with butter, and that opening a fridge door doesn't cause a nuclear holocaust in the kitchen). When deciding which information is relevant to a particular task we could not possibly (even using parallel processing) go through an exhaustive list of things that we know, ticking each off as being irrelevant like R2D1; rather we seem to just know what to consider and what not to consider. A classic example of the frame problem is seen in the thinking processes of chess grand masters. Although it may seem that grand masters consider more positions than a naive player when thinking about a chess position, in fact the reverse is true. Grand masters simply ignore certain moves (note that this is subtly different to briefly considering a move and then rejecting it) that lesser players would waste mental resources contemplating.
There must be in any intelligent being some highly efficient, partly productive, system for storing and recalling all of the information known to the being. The need for such a system is driven by space limitations (since our brains are not really very large) and by time limitations (since we live in a time-pressurised environment). What is in fact required is a system that stores a vast amount of information, but that genuinely ignores most of what it knows, and operates with a well-chosen portion of its knowledge at any moment.
