The real world is often uncertain. To model this world, expert systems exploit different methods for uncertain inference. Dempster-Shafer theory is considered to be a promising basis for such methods. Although computationally expensive, it has been believed to provide a natural and sound model of human uncertainty handling. Voorbraak’s paper questions this belief. It claims that the range of applicability of Dempster-Shafer theory is rather limited. Also, the author argues that the requirements mentioned in the theory as necessary to justify Dempster’s rule of combination are not sufficient. Therefore, he describes some additional conditions.

For expert system developers, it is important to take this message into account. Those interested in Dempster-Shafer theory or in the comparison of different methods for uncertain reasoning may find it useful to read this concise paper.