Formalizing knowledge and expertise: where have we been and where are we going?

Younger readers may not be aware that the announcement of the ‘Fifth Generation’ computer project in Japan in 1982 created an unprecedented level of interest in AI and, its application wing, knowledge-based expert systems (http://en.wikipedia.org/wiki/Fifth_generation_computer). The establishment of the British Computer Society Specialist Group in Expert Systems (SGES) was a UK manifestation of this interest, and the proposal to establish a review journal aimed at fostering the development of an engineering discipline to underpin developments in the field was developed under the auspices of SGES and with the support of several of the members of the group's management board, notably its first two Chairmen Donald Michie and Alex d'Agapayeff.

One reviewer conjectures quite reasonably that this may be because ‘applications are no longer worthy of publication unless there is some novelty in them’. Given the excitement of 25 years ago, I am surprised by the implication that knowledge engineers have run out of ideas for new kinds of applications and novel techniques. My suspicion is that a new generation of knowledge engineering researchers simply has different interests. This is a pity because building challenging applications creates great opportunities to test current theoretical ideas and force the discovery of new techniques by confronting the complexities of the real world.

As an example, I would draw the reader's attention to work by Martin Beveridge who developed a medical expert system with a spoken language user interface, which addresses these requirements. More information about this promising technology, including demonstrations and links, can be found at http://www.openclinical.org/dm_homey.html.

Clinical decision support systems are a very hot topic in the medical world (see www.openclinical.org).

The essential idea here is a situation of interest (such as a fault in a device or a medical symptom that needs to be diagnosed) can be classified by linking it to a node in a suitable domain ontology, and then this node can be mapped to a node in another ontology of possible causes (e.g. failures, diseases).