To Err...

As noted science fiction author Robert Heinlein once said: "Man is not a rational animal, he is a rationalizing animal." [3] As recent work in behavioral economics has made clear, our irrationality is not random; we are irrational in predictable ways [4]. Despite this, many "human-centered" systems do not leverage this predictable irrationality in a way that is potentially productive for users. Hypertext research has its roots in the field of human computer interaction. Some of the most influential early writers in hypertext (e.g., Bush [1] and Engelbart [2]) described hypertext systems as ones that would extend and augment human capabilities. This foundational perspective on hypertext is still well-evidenced in some of the most recent research. Bush focused largely on extending human memory; Engelbart on augmenting human intellect. For these two pioneers, as well as a host of other hypertext researchers, the key aspect of hyper-text that allows it to augment humans so well is its rich under-standing of structure and the computations in and around it. It is practically axiomatic within the field that this rich structure is somehow closer to "how humans think," though, as with most axioms, exactly why or how this is so is never fully and satisfactorily explained. Instead, the hypertext field relies on argument from example: non-linear traversals of information spaces (as opposed to an artificial linearity imposed by technologies such as writing) or emergent spatial organization of information (as opposed to prematurely formal organizations such as outlines). These examples (and others) work from generally "correct" forms of human thinking. Hypertext researchers, for example, do not claim that non-linear traversals of information spaces augment human intellect because some of the traversed links are faulty. On the contrary, linear representations are in some sense "defective" because they cannot represent the inherently non-linear thinking engaged in by humans, this "defectiveness" apparently attributable to a failure to capture useful associations that are (or could usefully be) present in the mind of the traverser. However, if fidelity of representation is the basis of the belief that hypertext can augment human thought, should we not also build into our understanding of structure the predictable biases we know are present? In this talk, we will explore the idea of accounting for the types of predictable irrationality discussed by behavioral economists in the understanding of hypertext structure. How can we model our biases? Does this bring us closer to accurate representations of human thought? Can we, in turn, find new applications for hypertext based not only on non-linear, emergent and rich structures, but ones that are flawed in a predictable way?