AI lawyer: I know how you ruled next summer
Nearly 80% accuracy in human rights trials from reports
RotM Artificial Intelligence can predict the outcomes of European Court of Human Rights trials to a high accuracy, according to research published today.
The use of AI has is slowly seeping into many industries including the legal sector. AI can trawl through vast amounts of information at a faster rate than humans without slowing down, making it easier for lawyers to prepare for hearings.
The paper, published in PeerJ Computer Science, shows that the new software has gone one step further. It can judge the final result of legal trials based on the information in human rights cases to 79 per cent accuracy.
But before the law profession starts panicking over AI taking over a judge’s job, the researchers are quick to reassure them that the system has not been designed to replace human arbiters.
"We don't see AI replacing judges or lawyers, but we think they'd find it useful for rapidly identifying patterns in cases that lead to certain outcomes. It could also be a valuable tool for highlighting which cases are most likely to be violations of the European Convention on Human Rights," said Dr Nikolaos Aletras, lead-author of the research and researcher at the Department of Computer Science at University College London.
The team of researchers analysed 584 cases relating to Articles 3 (prohibits torture and inhuman and degrading treatment), 6 (protects the right to a fair trial) and 8 (provides a right to respect for one's "private and family life, his home and his correspondence).
The software uses natural language processing and machine learning to analyse case information from both sides, Aletras told The Register.
It’s helpful that the information in the judgments case reports are laid out in a specific structure, making it easier for the software to analyse. The AI doesn’t understand language in terms of grammar and syntax or understand what any of the words mean.
Instead, words are broken down into vectors in high-dimensional space, and represented as “n-grams”, where the jumbled sequences of characters that represent the same word are a single n-gram.
The top 2,000 n-grams that occur most frequently are counted and similar n-grams are clustered into “topics”. Machine learning algorithms process the different topic in terms of weights, “the violation cases as +1, while no violation is denoted by -1”.
A lot of information has to fed to the AI system through supervised learning before it can make a judgment. “Information from each case is split into 10 parts. 90 per cent will be used for training, and the remaining 10 per cent goes into testing,” Aletras explained.
Although the AI has achieved a high-level of accuracy, it's important to know the results of the judgments were known beforehand, and the software has not been tested for active cases.
“Previous studies have predicted outcomes based on the nature of the crime, or the policy position of each judge, so this is the first time judgements have been predicted using analysis of text prepared by the court.
“We expect this sort of tool would improve efficiencies of high level, in demand courts, but to become a reality, we need to test it against more articles and the case data submitted to the court,” Dr Vasileios Lampos, co-author of the paper and researcher at UCL, said. ®