Machine solving should not be like a magic making process. Along with result, it is equally important to know the process by which the results were obtained. The thinking process of the machine must be like the way we humans do. The way humans have patterns in their thoughts, it should be embedded in the machine as well so that they can perform the tasks just like we humans do. When a human solves a problem, he learns it through the solution. The same capability has to be raised with machines as well. However the machines will have to store these patterns in the form of data and need construction of an information processing language to process the data.
Differences and Similarities
Machines and humans differ in several ways. There are problems that humans can find difficult which machines can easily solve and vice versa. There are similarities and differences between the behavior of program and human subject.
The Problem Space
A machine should be able to identify the correct problem space with the given task. So essentially we have task environment and problem space. A machine should be able to identify what is rote and what is meaningful information for the task. Constructing the problem spaces can happen with various information sources. The task instructions, previous experience of same task, previous experience with analogous tasks, generalization of range of tasks, using task instructions for constructing new problem space and information accumulated while solving a problem. Just like the process people follow, machines should carry out selective search in a problem space and incorporate some of the structural information of the task environment.
All the effort lies in making a machine more like a human being.
Once when we have problem space, the machine finds the solution by visiting node after node (state after state) and each of the nodes can be ranked. The problem solver learns iteration wise and explores for the next right state of information. This exploration can happen by progressive deepening or scan and search strategies. It starts by asking a question and to the solution to it might consist of several other questions. The particular heuristic search system that finds differences between current and desired situations finds an operator relevant to each difference and applies an operator to reduce the difference is usually called the means-ends analysis.
Production Rules and Behavior
Iterations are evaluated based on the set of production rules. A production ties stimulus together with its response. It’s like creating the hooks to glue the next desired node of information. All this is carried out so that to approximate the human behavior. What we need is a program of primitive information processes that generates the desired behavior.
Note: This post is an inspiration and summary of:
Human problem solving: The state of the theory in 1970, Simon, Herbert A.; Newell, Allen, American Psychologist, Vol 26(2), Feb 1971, 145-159.