Address the problem of ``what to do next''. Consist of search and control techniques. Hierarchical systems eg. trees can be searched using:
Eg. of heuristic search in problem space representation - the Tower of Hanoi puzzle: too many alternative paths between initial and goal states. Exploring all of these would take too long. So heuristics are needed.
Eg. of a heuristic - Means-end analysis: note the difference between the current state and goal state; establish a sub-goal to reduce the difference; find an operator to take you from sub-goal to goal ... continue thus till the gap between s-g and g is reduced to one step. Then apply the operators (in the reverse order in which they were generated.) In complex problems these comparisons might be carried out simultaneously along several dimensions.
Means-end analysis is an extension of the cybernetic principle of feedback: compare information about the current state to the desired state of the system and then nudge the current state in the desired direction.
Heuristics work with a narrow class of well-defined, puzzle-like problems requiring little prior knowledge. Real-world problems are less well-defined, goal state is not so clearly specified, they are semantically rich.
The heuristic approach is usually contrasted with the propositional approach in which mental representations are seen as propositions, or sentences in natural language consisting of a syntactic structure and semantic content. Eg. SHRDLU.
``strong AI'': the computer is really a mind and can be said to understand and to experience other cognitive states.
Searle's Chinese Room conundrum