Herself’s Artificial Intelligence

Reasoning Programs and Common Sense

Programs that have commonsense are some of the earliest attempts at artificial intelligence. The problem with the commonsense programs is there is no way to easily define the knowledge in a way the computer can use with out the computer understanding language. Little success was made here, but the descendants of common sense programs, expert systems have been very successful at problem solving of the type people learn in college, like engineering, law and medical diagnoses. From these studies came the computer languages Lisp and Prolong.

“Common sense knowledge includes the basic facts about events (including actions) and their effects, facts about knowledge and how it is obtained, facts about beliefs and desires. It also includes the basic facts about material objects and their properties.” [John McCarthy]

John McCarthy’s paper on common sense shaped the artificial intelligence movement for several decades. The Advice Taker is a program that was to have commonsense. He states that ‘A program has common sense if it automatically deduces for itself a sufficiently wide class of immediate consequences of anything it is told and what it already knows’. So the goal is for the Advice Taker to be able to discover simple abstractions. McCarthy defines proof of a program having common sense to be: that all behaviors be representable in the system; interesting changes in behavior must be expressible in a simple way; the system must understand the concept of partial success; and the system must be able to create subroutines to be used in procedures.

Another attempt at a general problem solving program following this type of reasoning is the “General Problem Solver” [Newell, Shaw, Simon, Ernst]. The program was given objects and operators with which to compare objects, take actions, and develop situations. An initial situation and a goal are given to the program and the program determines how to get from here to there. Three basic steps are taken to do this:
Evaluate difference between current situation and the goal state
Find an operator that lowers the difference between current and goal state
If it is legal to apply this operator, do so, else determine a situation that can use that operator and set it as a short term goal.

Reasoning Programs are the next step in this line of research. They will need to be able to sense and or find information about their environment, to prove whether or not solutions exist to problems given them. They will need to be able to reason out steps from initial situations to goals. Importantly it will need a language to define objects, goals, operators, logic, and temporality, states of being and other things.

Situational states or states of being are data structures and often defined with matrices, lists, statements, sets, trees, or more commonly graphs. Situational calculus, a derivative of first-order predicate calculus was developed to allow for states to be described as data structures as well as the effect of actions on states.

Common sense is very different than intelligence or education. Some people have one, two, or all three of these qualities. Teaching and testing for common sense has not progressed well with people and will probably not do well with computers until we have a greater understanding of exactly what common sense is, how it is acquired and how it can be tested for. Some other problems developing these systems are putting common sense into a language that is easily understood by people and computers. A second major problem has been representing time and changes that occur over time. Common sense seems to be learned from doing rather than being taught, so it may be that agents may gain common sense about the computer network they exist on, or further down the line robots may gain a bit of what we consider common sense about our world.

See also:
Knowledge based expert systems