Phan Minh Dung described a model for argumentation in Artificial Intelligence in 1995 [1]. The model presents an argument framework as a collection of arguments together with an attack-relation defined on the set. A single argument A is acceptable if every argument that attacks A is itself attacked in the set. A set of arguments is admissible if and only if it is conflict free and every argument it contains is admissible.

This paper, while not criticizing the structural concept of Dung’s argument framework, uses examples to show that the framework is epistemologically flawed. Two kinds of examples are presented to show this. In the first example, the author shows that Dung’s principles are in conflict with the principle of judgment suspension where an argument is neither supported nor attacked in the framework, so that a principal would suspend judgment on this argument. In the second example, the author shows that the knowledge base interpretation of an argument framework fails to take into account that a rational agent must take into account foreign arguments that cannot be constructed from the proponent’s knowledge base.

The examples are lucidly explained and the paper poses an interesting challenge to a much-used construct.