AI under the test of “beyond any reasonable doubt” in interpreting criminal law

The paper illustrates the possibility of using AI not only as an instrument to check the facts in a criminal trial but even to interpret and resolve significantly legal matters. Currently, the “expert systems” developed for this purpose take advantage of the judicial precedents as the basis of knowledge: given that, it is argued that these algorithms in a Constitutional State can not work based on the statistical rule of “more likely than not”, but they should be programmed according to the “political” alternative rule of ‘beyond any reasonable doubt’, which should be extended even to the doubt in interpreting the law. Thus, in the case of opposing judicial precedents, AI systems should suggest the most favorable interpretation for the defendant, and the judge should dissent only by explaining why he does not hold plausible the most favorable judicial precedent.