By implicit learning, Dr. Taatgen means that the knowledge of the expert system should continuously be adapted to the changing experience of the system. The cognitive ACT-R architecture, developed by John Anderson, solves this issue by saving the parameters of chance for possible recurrence in the future of all factual data. If a patient suffers from headache and fever, the probability of influenza normally is much higher than that of meningitis. If however a case of meningitis has occurred in the recent past, there exists a temporarily increased risk for meningitis. It is this kind of relevant chance information that has to be taken into account by the expert system by using the ACT-R parameters.
In order to solve complex problems, people tend to use former examples, that have been integrated into their implicit experience. Examples are applied to build up reasoning. This type of unconscious knowledge is extremely difficult for an expert to express in words. Since the knowledge of expert systems has been derived from the human expert who can't exactly define his skill to rely on past examples for current problems, it is hard to transfer this knowledge to the expert system which is typically based on rules instead of examples. By means of a well defined series of symptoms, the reasoning mechanism of the computer will search for the matching rules, possibly requesting some additional information, to finally come up with a diagnosis.
The general practitioner instead will base his decision on former experiences with the particular patient. Does this person only need to be put at ease or does he really need examination? It all depends on the implicit knowledge the doctor has built up with this patient. In turn, explicit learning consists of learning through learning strategies and there are quite of lot of strategies available to the human mind. The expert can solve a new problem by using analogy with a former similar problem, whether by focusing one particular aspect of the problem, whether by trying to gain more knowledge about the problem through the expansion of the problem domain. Strategies can be general but also very specific.
Again, this constitutes a challenge for the expert system builder. Indeed, the traditional knowledge system is based on the obsolete concept of a domain-independent reasoning module. In this regard, the ACT-R architecture for cognition is able to contribute a great deal to an alternative way of reasoning for expert systems. Its implementation of the three vital aspects of implicit and explicit learning, which are experience, examples, and strategy, can turn knowledge systems into valuable assistants for medical decision making. More information about the ACT-R cognitive architecture is available in Dr. Taatgen's thesis "Learning without limits: from problem solving towards a unified theory of learning".