A human may be thought of as an intelligent agent who observes percepts from his or her environment and produces actions based on those percepts. If the goal, then, is to maximize knowledge about the world, then actions that yield more knowledge would be preferable.
Consider the action to read a book and the action to look at a crowd of people. Surely, the density of information in the crowd of people is greater, right? However, the book contains a model of the world built from many years of observation. That book is also built upon other books and so on. Furthermore, the model in that book may be based on percepts that cannot be observed in the current time (e.g., comets, natural disasters) or with instruments available to the agent (e.g., microscope, telescope, invisible waves).
In most superintelligence scenarios, without some additional sense augmentation, reading books would probably give you the largest acceleration of knowledge acquisition. However, it's not without flaws sense the models described by the books may be erroneous or incomplete. So, it is likely that a combination of first-hand and second-hand experiences would combine to become greater than the whole.
If you consider Will, his primary goals seem to be friendship and romance with knowledge acquisition being some n-ary goal spurred by curiosity. So, he probably sates his curiosity with books rather than deduction from first-hand experiences since it frees up his actions to focus on his more pressing goals.