In several papers, e.g. [McC88] and [McC89], I discussed the common sense informatic situation and contrasted it with the information situation within a formal scientific theory. In the latter, it is already decided what phenomena to take into account. In the former, any information possessed by the agent is available and potentially relevant to achieving its goals. Elaboration tolerance seems to be the key property of any formalism that can represent information in the common sense informatic situation.
Elaboration tolerance1 is the ability to accept changes to a person's or a computer program's representation of facts about a subject without having to start all over. Often the addition of a few sentences describing the change suffices for humans and should suffice for computer programs.
Humans have considerable elaboration tolerance, and computers need it to reach human-level AI. In this article we study elaboration tolerance in terms of logical AI. However, researchers pursuing other AI methodologies will also have to face the problem of elaboration tolerance; maybe they just haven't noticed it yet. The relation to belief revision will be discussed briefly in section 8.
Humans represent information about the world in natural language and use background knowledge not ordinarily expressed in natural language and which is quite difficult to express. 2 The combination of linguistic and non-linguistic knowledge is what gives us humans our elaboration tolerance. Unfortunately, psychological research hasn't yet led to enough understanding of the background knowledge, so it is hard to study elaboration tolerance in humans. However, it is easy to give plenty of examples of human elaboration tolerance, e.g. those in this article.
The Drosophila3 of this article is the missionaries and cannibals problem (MCP).
After describing the original MCP, we give a large number of elaborations. Humans tolerate these elaborations in the sense that we can use the sentences expressing one of the elaborations to get a modified problem. People will agree on what the modified problem is and will agree on whether a proposed solution is ok.
Then we consider logical formalizations of the original problem and discuss which elaborations different formalisms tolerate. Our goal--not achieved in this article--is a formalism for describing problems logically that is as elaboration tolerant as English and the associated background knowledge. However, some logical languages are more elaboration tolerant than others.