For example, mining of historical data means that what we gleaned is not novel, because it is at least one step behind the latest state of knowledge/affair. Do you agree?
This comes in the purview of temporal data mining... It purely depends on what you looking for... you can divide time into intervals and try to compare data in one time frame with the other...
Not necessarily. In some cases mining large historical collections results in findings which the original authors would never have imagined. By correlating information mined from many different documents and different types of information and visualising them, it is possible to answer questions which would be difficult and tedious using traditional research methods.
Thanks Tarun and Beatrice. Perhaps I will rephrase the question by attempting to differentiate the word "new" and "novel", so as to add a sense of comparativeness. I think data mining can discover previously unknown insights that actually exist in current realm, but it cannot discover novel insights that exist outside the current realm, nor can it create insights that never exist before.
Data mining is by definition effort to understand "underlying laws" how a particular system works. The knowledge derived is invalid for for different system but if the system is same historical data can definitely create new insights about the underlying laws.
Think of "gravity" as a system and Imagine you currently do not know Newton's laws. You gather lot of data about objects on earth their acceleration, force, displacement, etc etc. Now you can definitely create the insight of Newton's 1st law, 2nd law etc. and within the Earth's given system it will hold true. Since you know similar should hold true for other celestial objects, it can also be tested on a different realm. but in general you will need to be careful porting knowledge from one realm to another.