Predictive Analysis of Judicial Behavior – Return to Mechanical Jurisprudence? Cover Image

Analiza predvidljivosti postupanja sudija – povratak mehaničkoj jurisprudenciji?
Predictive Analysis of Judicial Behavior – Return to Mechanical Jurisprudence?

Author(s): Dragutin Avramović
Subject(s): Law, Constitution, Jurisprudence, Philosophy of Law
Published by: Правни факултет Универзитета у Београду
Keywords: Predictive analysis; mechanical jurisprudence; jurimetrics; legal realism; artificial intelligence

Summary/Abstract: This paper is about the roots of computerized predictive analysis of judge’s behavior, and it investigates problems which predictive analysis is facing today, its prospects, as well as advantages and limits of that method. The author is of opinion that major limitation in predicting behavior of a judge with broad precision is emotional shortage of artificial intelligence. Following the line of legal realism the author concludes that unpredictability of judge’s behavior, namely specific and unique feature of their personalities, makes legal outcome not completely certain. Algorithms are today based mostly upon mechanical jurisprudence and syllogistic reasoning as computers are “fed” by man. In the future it might become possible, due to dynamic development of technology and with possibility of developing self-learned artificial intelligence, that machines could once replace judges or at least predict rulings with a very high percentage of probability. But, if the law is ars boni et aequi, and that ancient Roman postulate is not yet validly contested, then the range of mechanical jurisprudence is still limited. Machine may be able to gain equality, as it is basically and originally a mathematical principle and operation. But it cannot reach sensitivity of the second, correspondingly important, or even more fundamental legal principle – what is good in general and in concrete case. Good is a value, having its specific loading and feature in every single case. This is something that the machine cannot properly recognize and measure.

  • Issue Year: 2018
  • Issue No: 2
  • Page Range: 155-167
  • Page Count: 13
  • Language: Serbian