There is an undergoing discussion on how do we define an 'expert'. Here it is my view on the topic:
(Note: originally posted on http://bit.ly/cTydOg)
The basic assumption behind the concept of being an ‘expert’ — which, based from the above comments, it seems everyone agrees — is that there should be a *learning process* that makes such entity distinguished from the ones who haven’t acquired the same level of knowledge or skill.
Thus, one can conclude that an ‘expert’ is a system who have improved its performance based on its learning experience. Note that this is exactly the standard definition one would find by studying “Machine Learning”, a subarea of AI and statistical pattern recognition. However, how does one assess the ‘performance’ of a system? Although that’s not so simple, it’s not that difficult, as there are a variety of metrics one can use to precisely measure the ‘reliability’ of a diverse range of claims made by an ‘expert’.
In this sense, I fully agree with @Openworld comment above in that we should focus on the outcome of the process itself. That is, pick a definite preformance metric and we won’t need to rely on ‘common sense’ to know whether one really knows what he/her is ‘talking about’.
Everyone can do that at home.