Soft­ware to weed out vi­o­lence and porn

Weekend Argus (Saturday Edition) - - MEDIA& MARKETING -

WASH­ING­TON: “She got caught up in the likes” was how a prose­cu­tor de­scribed the 18-year-old Ohio teen ac­cused of live- stream­ing the al­leged rape of her 17-year-old friend. There is no ques­tion that live-stream apps, such as Periscope and Face­book Live Video, carry the risk of ex­pos­ing au­di­ences to heinous im­ages. In rare but har­row­ing in­ci­dents, users have watched as sui­cide, rape or do­mes­tic vi­o­lence un­fold in real time be­fore their eyes.

The job of watch­ing and re­mov­ing vi­o­lent or porno­graphic con­tent from th­ese live apps, as well as video sites like YouTube, has pri­mar­ily been a hu­man un­der­tak­ing. The sheer amount of con­tent makes it a chal­lenge for man­power alone. Now, ar­ti­fi­cial in­tel­li­gence is poised to help with this task.

Dex­tro, a New York- based start-up, uses video recog­ni­tion AI to eas­ily search through con­tent on live-stream apps. The tech­nol­ogy or sim­i­lar soft­ware could eas­ily be used to weed out porn or vi­o­lence.

Co-founder David Luan said the chal­lenge lies in cre­at­ing soft­ware that can in­ter­pret not just still im­ages but mov­ing im­ages, au­dio and other “sig­ni­fiers”.

“What is chal­leng­ing about th­ese videos is that they are much more com­plex than just a sin­gle pic­ture. Even though it is a se­ries of im­ages all one af­ter the other, there’s the mo­tion el­e­ment, the au­dio, so much of that gets thrown away if you just an­a­lyse im­age af­ter im­age,” he said. “So you re­ally need to treat it like a whole piece and an­a­lyse that.”

Dex­tro’s soft­ware can recog­nise ob­jects and sig­ni­fiers in a frame with­out hu­man in­ter­ven­tion, such as a gun in a po­ten­tially vi­o­lent video. And AI’s recog­ni­tion soft­ware can an­a­lyse a video within 300 mil­lisec­onds of post­ing.

Face­book con­firmed the com­pany does not use AI to fil­ter out porno­graphic videos and de­clined to com­ment on whether soft­ware was be­ing de­vel­oped.

But in the af­ter­math of videos and live-streams of the deaths of Phi­lando Castile and Al­ton Ster­ling at the hands of the po­lice, the ques­tion of when cen­sor­ship is eth­i­cal or ap­pro­pri­ate poses a chal­lenge for tech de­vel­op­ers in the role of con­tent mod­er­a­tor. How would a ma­chine han­dle those videos, if it even­tu­ally takes over as the prime mod­er­a­tor? – Wash­ing­ton Post

Newspapers in English

Newspapers from South Africa

© PressReader. All rights reserved.