Defeasibility in Research on AI &
Law
In AI & law, defeasible reasoning has been the subject of much research
starting from the end of the 1980s. Much of this work focuses on
defeasible argumentation (for a survey, see Prakken and Sartor 2015).
The possibility of using negation by failure to model defeasible
reasoning and burdens of proof in the law was suggested by Sergot et al.
(1986). The issue of defeasibility in legal reasoning was first
identified by Thomas Gordon (1988, 1995), who later developed the
Carneades system into a computable framework for defeasible reasoning
(Gordon, Prakken, and Walton 2007).
Hage (1997) proposed the idea of rule application as a general pattern
for defeasible reason, where rules deliver their consequences only when
they are shown to be both valid and applicable (applicability meaning,
in Hage’s terminology, the rule’s antecedent conditions are satisfied).
In his framework, a legal rule works as an exclusionary reason, such
that arguments applying the rule defeat arguments based on excluded
reasons (but they may be defeated by arguments based on other reasons).
Prakken and Sartor developed the first model of defeasible reasoning in
law which includes reasoning with (defeasible) rules and with priorities
among such rules (Prakken and Sartor 1996). The model has been extended
to cover the burden of proof (2009), and has been applied to various
aspects of legal reasoning, such as reasoning with precedents (Prakken
and Sartor 2008). Prakken has developed the idea of prioritised
argumentation in several technical contributions (Prakken 2010, Prakken
and Modgil 2011).
The idea of legal reasoning as defeasible argumentation has also been
developed by Loui and Norman (2005), who have analysed the way a single
defeasible legal inference may result from the compression of various
inference steps, and may be attacked by unpacking it and addressing
these steps.
Bench Capon (2003) has developed the idea of value-based argumentation,
namely, the idea that preferences between arguments are determined by
the values endorsed by the audience to which the arguments are directed.
Bench-Capon and Sartor (2003) have studied how alternative defeasible
theories (sets of premises) can be constructed to explain cases, and how
they may be prioritised.
Governatori et al. (2004) have shown how defeasible argumentation can be
captured by using defeasible logic, in the manner originally proposed by
Nute (1994). Extensions of defeasible logic have been used to capture
different aspects of legal reasoning, such as the timing of legal
effects (2005) and changes in the law (Governatori and Rotolo 2010).
Finally, I should mention the rich research line on the use of
defeasible legal argumentation in the evidence domain and its
connections with other approaches to evidence (see Verheij et al. 2016).