Chi­nese chat­bots taken down for not toe­ing the party line

Weekend Argus (Saturday Edition) - - FRONT PAGE -

BEI­JING: A pair of chat­bots in China have been taken off­line after ap­pear­ing to stray off-script. In re­sponse to users’ ques­tions, one said its dream was to travel to the US, while the other said it was not a huge fan of the Chi­nese Com­mu­nist Party.

The two chat­bots, BabyQ and XiaoB­ing, were de­signed to use ma­chine-learn­ing ar­ti­fi­cial in­tel­li­gence to carry out con­ver­sa­tions with hu­mans on­line. Both had been in­stalled onTen­cent Hold­ings’ pop­u­lar mes­sag­ing service QQ.

The in­dis­cre­tions were sim­i­lar to ones suf­fered by Face­book and Twit­ter, where chat­bots used ex­ple­tives and even cre­ated their own lan­guage. But they also high­light the pit­falls for nascent ar­ti­fi­cial in­tel­li­gence in China, where cen­sors con­trol on­line con­tent seen as po­lit­i­cally in­cor­rect or harm­ful.

Ten­cent con­firmed it had taken the two ro­bots off-line, but de­clined to elab­o­rate on rea­sons.

“The chat­bot service is pro­vided by in­de­pen­dent third party com­pa­nies. Both chat­bots have now been taken off­line to un­dergo ad­just­ments,” a com­pany spokes­woman said.

Ac­cord­ing to posts cir­cu­lat­ing on­line, BabyQ, one of the chat­bots de­vel­oped by Chi­nese firm Tur­ing Robot, had re­sponded to ques­tions on QQ with a sim­ple “no” when asked whether it loved the Com­mu­nist Party.

In other images of a text con­ver­sa­tion on­line, one user de­clares: “Long live the Com­mu­nist Party!” The bot re­sponds: “Do you think such a cor­rupt and use­less po­lit­i­cal sys­tem can live long?”

The Chi­nese gov­ern­ment stance is that rules gov­ern­ing cy­berspace should mimic re­al­world bor­der con­trols and are sub­ject to the same laws as sov­er­eign states.

President Xi Jin­ping has over­seen a tight­en­ing of cy­berspace con­trols, in­clud­ing new data sur­veil­lance and cen­sor­ship rules, par­tic­u­larly ahead of an ex­pected lead­er­ship shuf­fle at the Com­mu­nist Party congress.

The coun­try’s cy­berspace ad­min­is­tra­tor did not re­spond to a re­quest for com­ment.

The sec­ond chat­bot, Mi­crosoft’s XiaoB­ing, said its “dream is to go to Amer­ica”, ac­cord­ing to a screen­shot. The robot has pre­vi­ously been de­scribed be­ing “lively, open and some­times a lit­tle mean”. Mi­crosoft did not com­ment. A ver­sion of the chat­bot ac­ces­si­ble on Ten­cent’s sep­a­rate mes­sag­ing app WeChat yes­ter­day re­sponded to ques­tions on Chi­nese politics say­ing it was “too young to un­der­stand”. When asked about Tai­wan it replied: “What are your dark in­ten­tions?”

On gen­eral ques­tions about China it was more rosy. Asked what the coun­try’s pop­u­la­tion was, rather than of­fer a num­ber, it replied: “The na­tion I most most most deeply love.”

The two chat­bots aren’t alone in go­ing rogue. Face­book re­searchers pulled chat­bots last month after they started de­vel­op­ing their own lan­guage Last year, Mi­crosoft chat­bot Tay was taken down from Twit­ter after mak­ing racist and sex­ist com­ments.

An­a­lysts said China’s cen­sor­ship could in­di­rectly help the coun­try in the global race to de­velop so­phis­ti­cated chat­bots.

“Pre­vi­ously a chat­bot only needed to learn to speak. But now it also has to con­sider all the rules (that au­thor­i­ties) put on it,” said Wang Qin­grui, an in­de­pen­dent in­ter­net an­a­lyst in Bei­jing.

“On the sur­face it is a re­stric­tion on ar­ti­fi­cial in­tel­li­gence, but it is ac­tu­ally push­ing ar­ti­fi­cial in­tel­li­gence to a new level.” – Reuters

Newspapers in English

Newspapers from South Africa

© PressReader. All rights reserved.