Legal Evidence and Metrics

In computer applications to legal evidence, metrics fits in diverse roles. In several forensic science disciplines, measurements and quantitative evaluations are important. In these two decades, there has been a great effort to apply formal methods and mathematical tools for assessing legal evidence from metric findings and testimony in order to make the court’s decision less arbitrary. By now it seemed clear that the typical tools of classical artificial intelligence, i.e. logic and probability, could actually make the judge’s decision fairer. In this article instead we claim that whether a kind of metric is appropriate or inappropriate depends upon what you are going to use it for. Bayesianism as applied to how legal evidence is weighted and evaluated, is the subject of a controversy; it is, indeed, indeed problematic. It would be wrong to embody probabilistic calculations if your software is to propose what the verdict ought to be in a criminal case. A ranking of the plausibility of scenarios other than probabilistically is to be sought and preferred. Probabilities are not as problematic, in such a context that (for example) the prosecution uses a computer tool in order to reach a decision as for whether a suspect is to be prosecuted, or alternatively the prosecution is to rather offer that suspect a plea bargain, or not to proceed, i.e., drop the case. Finally, Metrology and Artificial Intelligence are entering the legal test, and this paper is intended also as an introduction to this wide and varied field, which is rather scattered in publications belonging to scientific areas that are still too much far apart.