Artificial Intelligence taught to predict court decisions in cases of violation of human rights

Photo: wikipedia.org/CherryX

The combined team of experts at the University of Pennsylvania and Sheffield has created a weak form of artificial intelligence that can predict the decisions of the European Court of Human Rights (European Court of Human Rights, ECtHR, ECHR) with an accuracy of 79%. The system predicts a court decision after reading the case materials, and on the basis of these data, the AI ​​makes a conclusion. The developers published the details of their work in the journal PeerJ Computer Science.

“We do not believe that AI can replace judges or lawyers, but we think that they would find our system useful, because it can quickly identify those characteristics of the case that most likely lead to a definite verdict. Our system can also be a valuable tool for detecting cases that are most likely to violate the European Convention on Human Rights , ”says project head Nikolaos Aletras.

The Convention for the Protection of Human Rights and Fundamental Freedoms, which is also known as the European Convention on Human Rights, is one of the main documents of the Council of Europe. This international treaty was signed in 1950 and entered into force on September 3, 1953. The Convention establishes the inalienable rights and freedoms of every person and obliges the states that ratify the Convention to guarantee these rights to every person who is under their jurisdiction. And the jurisdiction of the ECHR extends to all Council of Europe member states, from Reykjavik to Vladivostok.
')
Last year, approximately twice as many complaints were filed against human rights violations than in 2014. Moreover, most of these requests were simply not accepted for consideration for one or another formal reason (for example, preparation of a document not according to the rules). In the text of many cases, lawyers did not find corpus delicti. Now the flow of appeals to the ECHR is growing, but the number of judges remains the same. They have to get acquainted with the materials of thousands of cases in order to find those that are worth the attention of the court.

It is because of the large number of appeals to the ECHR that the Aletras team decided to create a special algorithm that could analyze the text of the appeals and predict court decisions on the fly. Analyzing the texts of hundreds of cases of the ECHR, the scientists found that the judges of this court, first of all, do not pay attention to legal terms in the texts of appeals, but to a simple description of the facts. That is, the judges, according to the authors of the draft, are rather realists who pay attention to the moral side of the matter, and not the formalists who strictly follow each letter of the law.

When learning the algorithm, they gave “acquaintance” with 584 court orders at the suit of citizens in the ECHR. Scientists chose those rulings that related to the three articles of the Convention for the Protection of Human Rights. They include a ban on torture and degrading treatment, the right to a fair trial and the right to respect for private and family life. Under these articles, the court rendered the most decisions.

Analyzing the texts of the appeals, artificial intelligence delivered its own “violation” or “non-violation” verdict. In order for the experiment to be clean, the scientists gave the AI ​​to analyze an equal number of cases where violations were or were not established by the ECHR. “In an ideal situation, we needed to test our algorithm on complaints just lodged, and not on documents that have already been published. Unfortunately, we do not have access to such data, so we decided to carry out the work using court publications. ”


Convention for the Protection of Human Rights and Fundamental Freedoms

As mentioned above, the computer, using a number of precedents, analyzed each particular case, rendering a verdict on it based on the phrases found in the text of the case, facts, circumstances. “Previously, our system could predict court decisions only on the basis of the type of crime or the opinion of each individual judge. We were able to predict the decisions of the judges based on the analysis of the documents prepared by the court, ”said one of the study participants.

According to the developers, in order to become a truly effective tool, artificial intelligence needs to be run-in on a larger number of documents, including those that have not yet been considered by the court. The authors of the paper argue that the method can also be used to assess the testimony of witnesses in a particular case or the statements of lawyers.

Experts believe that some aspects of AI cases are simply unable to “understand”. “The AI ​​can guess pretty well, but without the use of context, the work of the system can give an erroneous result,” said Matt Jones, an analyst at Tessella.