Distribution of Damages in Car Accidents throught the Use of Neural Networks.
In: Cardozo Law Review, Vol. 13: pp. 987-1000
After a traffic accident the damage has to be fairly divided
among the parties involved, and a ratio has to be determined.
There are many precedents for this, and judges have developed catalogues
suggesting ratios for common types of accidents.
The problem that "every case is different," however, remains.
Many cases have familiar aspects, but also unfamiliar ones. Even if
a case is composed of several familiar aspects with established ratios,
the question remains as to how these are to be figured into one
ratio. The first thought would be to invent a mathematical
formula, but such formulae are rigid and speculative. The body of
law has grown organically and must not be forced into a sleek system.
The distant consequences of using a mathematical formula
cannot be foreseen; they might well be grossly unjust.
I suggest using a neural network instead. Precedents may be
fed into the network directly as learning patterns. This has the
advantage that court rulings can be transferred directly and not via
a formula. Future modifications in court rulings also can be
adopted by the network. As far as the effect of the learning patterns
on new cases is concerned, a relatively safe assumption is that
they will fit in harmoniously with the precedents. This is due to
the network's structure—a number of simple decisional units,
which are interconnected, tune their activity to each other, thus
achieving a state of equilibrium. When the conditions of such an
equilibrium are translated back into the terms of the case, the solution
can hardly be totally unjust.