This chapter discusses the application of artificially intelligent systems to the
assessment of evidence in criminal cases and proposes a system for formalizing
argumentation schemes for machine learning. It draws on recent work on
argumentation schemes and adapts them to address common issues encountered in
criminal cases, such as the credibility of witness testimony, the probative value of
similar facts, and the evaluation of expert evidence. This is illustrated by systematically
setting out the arguments presented over the course of several decades in the trials of
Robert Earl Hayes and identifying critical questions to evaluate the evidence. Hayes’
case is a well-known miscarriage of justice that has been extensively investigated and
litigated. It therefore serves as a good model for evaluating the arguments put forward
in the case over the course of several decades, as the facts of the case are well-known
and reliable. The proposed system combines argumentation schemes with scenariobased reasoning. The arguments are separated into different scenarios put forward by
the parties. Each scenario is evaluated separately; in the final step, their relative
plausibilities are compared using an abductive argumentation scheme to evaluate the
most plausible scenario. The final result is a qualitative plausibility value that signifies
how justified we are in believing that a given scenario is the best explanation for the
available evidence, at the time of evaluation. The plausibilities proposed here are
qualitative, relative and are correlated with legal standards of proof. This system assists
in identifying and removing prejudices and cognitive biases that impede reliable
evaluations of evidence. It also assists in better formalizing and defining argumentation
schemes to evaluate evidence in criminal cases, so as to prepare the way for future
computational applications.
Keywords: Argumentation theory, Argumentation schemes, Artificial Intelligence, Forensic DNA, Law, Machine learning, Philosophy of evidence, Trial advocacy.
