27. What Are the Effects of Obtaining or Renouncing a Piece of Evidence?

fancyfantasicAI and Robotics

Nov 7, 2013 (3 years and 5 months ago)

88 views

E. Nissan
Computing, Criminal Investigation, Legal Evidence, and Argumentation


27. What Are the Effects of Obtaining or Renouncing a Piece of Evidence?


Levitt & Laskey (2002) combined Bayesian inference networks with knowledge representation from AI, in
order to structure and carry out an analysis of evidential argumentation a
bout an infamous murder cases. In
their Section 1.5, they enumerated desiderata concerning current and future tools for supporting judicial proof:
such tool should allow to represent hypotheses and supporting or denying evidence (Section 1.5.1), to c
ompare
beliefs between alternative hypotheses (Section 1.5.2), to update belief based on incrementally accrued
evidence (Section 1.5.3), and to examine variations of the same hypothetical

evidential scenario (Section
1.5.4). In Section 1.5.4, they w
rote (pp. 381

382): “A unique application of sensitivity analysis” (the
sensitivity analyses reported in their article, in Section 1.4.4., being performed by Keung
-
Chi Ng) “has been
provided for” Bayesian networks (BNs) “that provides a quantitative
ly powerful, scientifically meaningful,
and qualitatively intuitive measure of the relevance and importance of evidence to the truth or falsity of a target
hypothesis or BN query”. Gathering evidence may be costly and time
-
consuming. Does what is to be
proven
really require a given piece of evidence to be obtained?


Before evidence is observed, we can assess the impact of observing any possible state of an
evidence variable, as well as the impact, if relevant, of failing to learn anything about the
e
vidence available. We can examine the degree to which the probabilities in which we are most
interested would be affected by observing the evidence. We can rank the different evidence
-
gathering strategies by impact on the conclusion and prioritize eviden
ce gathering accordingly.

When there are monetary or other costs to evidence gathering, these techniques provide tools
for balancing the information gain of evidence gathering against the cost. [...] (Levitt &
Laskey,
ibid.,

p.

382).


An important con
sideration we can make is that if the purpose is to assess beforehand whether to incur the costs
of obtaining this or that evidence, then this is not as vulnerable to a Bayesio
-
skeptic critique in the controversy
about Bayesianism, as actually trying to
model the strength of the evidence for conviction in the actual trial,
and proposing such statistics as evidence, would be. In fact, making use of Bayesianism for a costs/benefits
analysis so that a decision is made, during fact investigation or the p
reparation of a trial, to either obtain or
renounce some evidence, belongs in the economics of one party in the trial preparing its case. A private party
turning to the courts, or called to respond, are within their rights to adopt a flawed method,
or consult an
astrologer, if they so wish, as well as to mismanage their case, and therefore have no case to answer if charged
with methodologically unwarranted assumptions by the Bayesio
-
skeptics. When it comes to the prosecution
preparing its cases,
they are more vulnerable to critique, if they demonstrably grossly mismanaged a case.
Yet, especially in the adversarial system of procedure from Anglo
-
American jurisdictions (unlike, until
recently, in many countries of the European Continent), in
criminal cases the prosecution has the discretion
whether to prosecute at all. Therefore, Bayesian methods being used in a costs/benefits analysis of whether to
obtain some evidence, or then of whether to prosecute or offer a plea bargain instead, are
immune from
Bayesio
-
skeptic attacks.