ArgKit

ArgKit is a Java software library for integrating argumentation into third party software applications. It is distributed under an open source license and available on sourceforge. It's main module is called Dungine, a module that implements several algorithms for Dung style argument evaluation.

The underlying model for Dungine, Dung’s acceptability semantics, is interesting to people who work in AI and are trying to capture how people reason and debate, particularly with inconsistent evidence and automate reasoning in this context. In COSSAC we use ideas like this to try and improve clinical decision support systems that:

  • recommend diagnoses for a patient with a history and a set of symptoms and
  • recommend treatments for a diagnosed condition.

Clinicians don’t always agree, and neither does the evidence upon which they base their opinions. If we can adequately represent their arguments, then this software might allow us to automatically reason about which of those arguments we should accept.

In the business world one might use Dungine to:

  • accept a recommendation from a source on the web or
  • represent a decision about granting a loan.

There are many alternative approaches to supporting these examples, including statistical methods or applying a set of prioritised rules. Argumentation is probably most useful in situations where it feels natural to represent contradicting reasons for doing something (accepting the recommendation or granting the loan) but this is a new area, and there are too few examples of it’s application in the wild. So ArgKit has been built to help people try it out, and in doing so gather empirical evidence to help us understand whether this form of argumentation is a practical and useful technique.

Links