AI may help avoid wrong­ful con­vic­tions

Sys­tem ‘learns’ to spot holes to aid po­lice, pros­e­cu­tors, judges

China Daily (Hong Kong) - - FRONT PAGE - By ZHOU WENTING in Shang­hai zhouwent­ing@chi­nadaily.com.cn

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, vi­cepre­sid­ing 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.

Newspapers in English

Newspapers from China

© PressReader. All rights reserved.