The Role of Open Texture and Stare Decisis in Data Mining Discretion

The use of knowledge discovery from database (KDD) techniques such as neural networks in the legal domain can alter the practice of law by introducing a non-expert mechanism for predicting and analyzing Court outcomes. Legal domains that are well suited to KDD are those in which a high level of judicial discretion exists particularly where the source of the discretion stems from a principle statute that presents a list of factors to be taken into account without specifying the relative importance of each factor. This kind of discretion, sometimes referred to as shopping list discretion contributes to the characterization of law as open textured in a way that is different to situations more often considered open textured by the artificial intelligence and law community. Whereas forms of automated reasoning such as non-monotonic logics appropriately model other forms of open texture, KDD techniques are particularly well suited to modeling open texture characterized by shopping list discretion. This is understandable if a distinction between traditional, local and personal stare decisis is drawn. KDD is suitable for discretionary tasks because local and personal stare decisis often dominates traditional stare decisis in these tasks. This has two immediate ramifications for KDD that are discussed in this paper. The first involves a conceptualization of judicial error in local/ personal stare decisis on empirical grounds. The second involves the inclusion in KDD training sets of cases that are not interesting in that they set no precedent and/or are not appealed. These points are illustrated by articulating assumptions made in applying KDD to the prediction of property outcomes in Australian family law. This is a domain that involves considerable judicial shopping list discretion in addition to an emphasis on local and personal stare decisis.