System 'learns' to spot holes to aid police, prosecutors, judges

Shanghai is testing an artificial intelligence system that helps police officers, prosecutors and judges check the validity of evidence in criminal cases, as part of an effort to prevent wrongful convictions.

Over the past month, the system has reviewed 60 cases－including homicides, burglaries and telefraud－and correctly identified 48 flaws in evidence, the Shanghai High People's Court said on Monday.

Technicians entered information into the AI system from 17,000 documents related to old cases, such as case files, judgments and notices requesting that police reinvestigate. The system used the information to "learn" how to spot potential problems.

"It will continue to make progress if more learning models are established, and more materials are input for it to acquire a stronger ability to identify doubtful evidence through repetitive learning and exercise," said Guo Weiqing, vice-president of the court.

According to the Supreme People's Court, 34 wrongful criminal convictions have been overturned since 2013, drawing nationwide attention.

One reason for wrongful convictions is that facts are unclear and evidence is insufficient, said Cui Yadong, president of Shanghai High People's Court.

"The AI system was designed to shoulder two missions," he said. "One is to ensure that the standard of evidence is consistent in all cases. The other is to see if all the unknowns in a case have been verified by existing evidence and to find blemishes in evidence－and contradictory evidence－in a timely manner, and to alert officers handling the case to guarantee that all evidence can stand the test of the law and curb subjectivity and randomness in case handling."

The key in such a system is setting up a standard for it to learn what is essential evidence, what constitutes a complete evidence chain and whether the evidence is capable of proving the case, according to Jin Zemeng, a product manager at iFlytech, an information technology company involved in the pilot project.

The standard of evidence will differ depending on the case, said Xu Shiliang, vice-presiding judge at a criminal tribunal of the Shanghai court, noting that standards for 18 criminal charges have been set.

"For example, we came up with 30 indispensable pieces of evidence and 235 standards to verify the evidence based on the archives of nearly 600 major cases of homicide, intentional injury, robbery and kidnapping," he said.

Ye Qing, president of East China University of Political Science and Law in Shanghai, said that AI can be applied in many ways in the judicial field to help reduce judges' enormous workload and improve the quality of their work.