US engineers and lawyers have taught artificial intelligence to predict decisions of the US Supreme Court in already closed cases. The system does this better than professional lawyers.
Experts from the Stanford Center for Legal Informatics, the Chicago College of Law in Kent, and the South Texas College of Law have created an artificial intelligence system that can predict the outcome of a lawsuit better than professional lawyers.
To teach AI to deal with legal collisions, the US Supreme Court database, which has been maintained since 1791, was loaded into the system. On the basis of her data, the system built an algorithm that predicts a judge's decision, based on 16 parameters: the field of law, time, information about a particular judge, etc. Additional factors are also taken into account - for example, whether the oral arguments of the parties took place.
Based on historical data collected between 1816 and 2015. and processed by machine learning algorithms, the researchers created a statistical model that the system uses to find patterns between the characteristics of the case and the court's verdict. Now, using this model, AI correctly predicts the outcome of 70.2% of cases out of 28 thousand, and the system predicted decisions of individual judges correctly in 71.9% of cases out of 240 thousand. By comparison, legal experts correctly predict a Supreme Court decision only 66% of the time. An article describing the algorithm was published in the journal PLOS ONE.
The achievement of American scientists is inferior to the work of their colleagues at University College London: last year they taught AI to predict decisions of the European Court of Human Rights with 79% accuracy.
Last year, IBM, based on the Watson supercomputer, created artificial intelligence capable of recognizing natural language and answering legal questions. The international law firm Baker & Hostetler bought the rights to use AI and entrusted it to communicate with its clients - searching for similar queries in the database, analyzing the relevant legal regulations and selecting recommendations.
Alexander Privalov