4. User Models
4.5 The User’s Mental Model
The point of these possible worlds is that they can be thought of as a set of hypotheses about the user’s model of the world (the system), since they are ’generated' from a user command which is, we assume, intended to work, i.e. a command which the user assumes is consistent with the actual world. The next step is to choose from among a number of candidate possible worlds one which is deemed most likely to correspond to the user’s model of the system. Logical proximity can be taken as an initial measure of best fit, since it credits the user with ’distorting’ the actual world as little as possible. In other words, it assumes the user behaves rationally.
However, logical proximity only gives us a partial aid to discriminating between possible worlds, since it only addresses one dimension of user psychology; one which we might call 'conservation'. By this is meant the natural tendency of the user's model to 'track' the actual world rather than radically diverge from it, stemming from the user's desire to maintain control of the interaction and to 'know what is going on’. This use of possible worlds has similarities with Nozick's (1981). Other factors, such as failures of short term memory affect the user’s ability to do this and these need to be taken into consideration so that a measure of 'psychological proximity' can be determined.
command as subsisting in a psychological space and, again, some will be closer to the actual world than others. Psychological proximity is the epistemic closeness of a possible world to the actual world. Thus we have to seek evidence which is held to support or undermine the user's belief in the various possible worlds compatible with the anomalous command. Of course, there will be a cut off point for proximity beyond which it is considered highly improbable that the possible world concerned could be believed in by any user. This will limit the search space by reducing the number of possible worlds requiring consideration. (An alternative would be to pursue a notion of epistemic possibility in order to limit the search space to those possible worlds which are compatible and consistent with what the user knows. However, this would require the creation and maintenance of an extremely detailed model of user belief, which would be a significant achievement in itself. Therefore this avenue is not pursued here).
We have noted elsewhere the difference between system state and system functionality. The latter relates to the rules of the system and is static in nature, while the former is constantly changing. This means that knowledge of function is more likely to be stable and less prone to short term memory limitations. As is evident in the protocol analysis described in chapter seven, short term memory problems are legion with regard to state information concerning the UNIX® file system. This leads us to two different types of justification for psychological proximity.
The kind of evidence required to justify the user's belief in a possible world where the functional rules of the system are other than the actual rules will relate directly to the user and will generally be discontinuing. If a user has negotiated a particular function (F) of the system several times successfully or has been notified of a certain rule on several occasions, then a possible world with an 'incorrect' rule (F) is less likely to be that believed in by the user. Thus a user model of competence or knowledge can be used as a source of evidence in disconfirming user belief in functionally distorted possible worlds. Note that a system function may embody several rules and
that a single invocation of a function may not involve all of the rules relating to that function. Thus functional knowledge should be represented in terms of these rules and not in terms of whole functions.
As state model mismatches relate more often to short term memory limitations, evidence in support of state-distorted possible worlds is unlikely to be derived from this source. However, past system states can provide such evidence of psychological proximity. What we seek is any justification the user might have for favouring one possible world over another and the fact that a particular possible world was actual at an earlier time may provide such justification. For example, if the change that needs to be made to the actual world in order to transform it into a certain possible world is equivalent to the 'undoing' of some past user action, this may mean that the user has forgotten that past action and believes the system state to be that which would have been the case had that action not been performed. Of course, these circumstances are also compatible with a possible world in which the past action does not have the effect of rendering the later action invalid and that possible world is also a candidate for the user's mental model and misconception. In order to resolve this ambiguity, we require evidence to support one candidate over all others. Failing this, all candidates may assume equal status and so it will be necessary to correct for all of them and allow the user to identify the information appropriate to his or her misconception.