South Korea: a chatbot derails and turns out to be homophobic and sexist

 

For the South Korean chatbot Lee Luda, everything had started well. Launched in December, this conversational robot, allowing Internet users to chat with what would be a 20-year-old student, immediately met with enormous success, with up to 750,000 users. The reasons for this craze? The spontaneity and naturalness of the responses made by artificial intelligence.

Developed by Seoul-based start-up Scatter Lab to work via Facebook Messenger, Lee Luda responded using an algorithm developed from data collected from 10 billion conversations on Kakao Talk, the country’s leading messaging app. But the chatbot was quickly at the heart of controversy because of its hateful responses, to the point that its developers were forced to suspend it on Tuesday.

I despise them completely

On the screenshot of a conversation, we can see the virtual student asserting that she “despises” gays and lesbians. Asked about transgender people, Luda explodes: “You drive me crazy. Don’t repeat this question again. I said I didn’t like them. In another conversation, she explains that the people behind the #MeToo movement were “just ignorant,” adding, “I completely despise them. She also says that she “would rather die” than live with a disabled person.

Scatter Lab apologized for the comments, adding that they did not represent the values ​​of the company. This is not the first time that chatbots have been derailed. But the embarrassment is compounded by the fact that Lee Luda was operating on the basis of discussions that have existed, and his setbacks could be indicative of some ideas taking root in South Korean society.

Scatter Lab said it had worked to prevent these kinds of issues during the six months of testing leading up to the chatbot’s launch. “Lee Luda is artificial intelligence, like a child learning to have a conversation. She still has a lot to learn, ”the company said on Tuesday. “We’re going to educate Luda to exercise judgment on which responses are appropriate, rather than indiscriminate learning,” Scatter Lab continued without saying when the chatbot would be back in service.


You may also like