Several authors have noted the need for some form of embedded intelligence in large hyperbooks. Such an expert system could suggest possible links based on past choices, analyze visited nodes for recurring patterns, assist in searches, and so on. Oren has written of such "adaptive technologies" in the context of the Memex (321). Halasz notes that "[i]n many ways, hypermedia and knowledge-based systems are a natural fit" (847).
Expert systems may be classified as being generalized or specialized, and active or passive. They may be available through scripting or simple interfaces accessible to non-programmers.
Koch has written of the possibility of using neural nets to provide intelligence. A neural net is a simulation in software of the structure of the brain. Individual elements (neurons) are assigned a number of weighted inputs (stimuli) to which they generate an output (response). A network of such elements is trained by allowing it to make a series of guesses as to the correct response where the response is known. By various methods (eg. back propagation), the most ideal values for the weights are determined. Over time, as correct responses are rewarded, the net becomes more accurate.