iFlytek accused of using humans to power its AI translator
iFlytek is China’s leader in voice-related technology. It’s one of just four tech companies enlisted by the Chinese government into the "national AI team", companies that will supposedly lead the country’s development of artificial intelligence, and produces tools like voice recognition and translation.
But last week iFlytek was accused of misleading people into thinking its AI software was automatically translating a speaker at a conference -- when in fact it was transcribing a human interpreter instead.
iFlytek claims that it never promised simultaneous translation, and that it was all a misunderstanding.
One of the human interpreters, Bell Wang, told his story on Quora-like site Zhihu.
Wang was doing simultaneous voice translation at a government conference where a Japanese professor spoke in English. During the speech, two different text feeds were displayed: One to transcribes the professor’s English speech into English text, and the other apparently translates that into Chinese text. Both displays were marked with “Xunfei Tingjian” or “iFlytek Hears”.
But when Wang read both the English transcription and the Chinese version next to it, he noticed that the two didn’t match: Errors in the English transcription were not being translated into Chinese. Instead, Wang realized that the Chinese text was a transcript of what his fellow interpreter was saying -- and he was furious.
“I was so mad that my hair stood up,” he said in the post. “This is a flat-out scam!”
Netizens joined Wang’s outrage. The post received more than 75,000 upvotes and was widely reported by Chinese media, with many commenters sneering at what they saw was dishonesty by iFlytek.
But iFlytek says it’s always been honest about the capabilities of its software. In a statement, it said it was intentional: It offered conference organizers the choice of either translation or voice-to-text transcription, and the conference picked the latter. It also said that it asked that interpreters were notified.
The company also argued that it’s never said it could provide simultaneous translation -- and pointed out that it has, in the past, urged the public not to get too excited about the translation abilities of AI.
In a follow-up post, Wang said that an iFlytek executive admitted to him in a phone call that there was a lack of communication and transparency from iFlytek to interpreters, who weren’t given any credit and felt that their rights were violated.
Wang said that he could have missed iFlytek’s announcement that it will transcribe their translation, and says he doesn’t think the company actively faked anything.
We reached out to Wang, but he declined to give us more details.