This is an abstract of Marvin Minsky’s Frame System Theory paper, a classic paper in defining the problem space for an intelligent artifact. The paper itself is a pretty easy read, and it sheds some light on the sort of problem that humans are really good at doing without even realizing they’re good at it.
Minsky proposes a method of problem solving based on frames, data structures representing a generalized situation which store information regarding how to work with the situation. To make use of these in problem-solving, a situation being observed must be broken into sub-frames. The system attempts to match the sub-frames with an overarching frame in its database. If a near-match is found, that near-match provides links to related frames that may be a better fit. The system must be able to symbolically represent what it is experiencing, make “excuses” (inferences regarding why a particular frame does not quite fit, but should still be selected), and come up with novel frames if no match can be found. Finally, frames should be grouped into classes, with each frame related to its neighbors and the most connections passing through very common or general forms of the problem, so that a large variety of situations can be matched quickly.