We elaborate our recent thesis  stating that computation is a process of knowledge generation. We give two conditions for a process to be computational, i.e. to be a knowledge generating process. First, the epistemic domain in which the computation is carried out must be known, and second, there must be evidence that the generated knowledge is indeed derivable within the underlying domain of ... [Show full abstract] discourse by the rules governing the domain and the underlying computational mechanism. The fulfillment of these conditions may be decided by an observer which, again, is modeled as a computational process according to our definition. As a consequence, our definition of computation is observer-relative. The viability of our definition is scrutinized by several examples of computations considered widely in the literature. Among them, we consider the question whether a rock can compute as well as some aspects of Searle's Chinese room thought experiment. The examples illustrate that the epistemic approach to computation brings valuable new insight into the nature of computation and helps to resolve some classical problems related to these examples.