The construction of arguments has long been viewed as a paradigmatic example of human reasoning and, as such, is an important ability for computer programs that attempt to model intelligent behavior. We explore the use of argumentation for deriving and justifying claims in domains where knowledge is incomplete, uncertain, or inconsistent, i.e., weak theory domains. Argumentation supports a notion of proof appropriate for reasoning in weak theory domains, e.g., a claim is proved if there is plausible, irrefutable support for the claim, and there is no such support for any counter-claim.

We present elements of a theory of argumentation involving two senses of argument, argument as Supporting explanation and argument as dialectical process. For argument as supporting explanation, we create argument structures that organize relevant, available support for both a claim and its negation. In dialectical argument, the format of a two-sided argument process is used to intertwine the strengths and weaknesses of support for competing claims, so arguments can be refuted and directly compared. Our account of dialectical argumentation includes a catalog of argument moves and a set of heuristics for selecting moves and thereby controlling argument generation.This model, which has been implemented in a computer program, is a flexible environment for exploring the representation and generation of arguments. We show how the program generates reasonable arguments for a set of example problems. We give an analysis of the program, including limits of the current model of argumentation.

For artificial intelligence programs, the ability to generate arguments provides a useful technique for reasoning in real world contexts. For argumentation researchers, artificial intelligence methodology offers a new way for evaluating theories of argumentation.