Scatter Lab “Achieved, I couldn’t filter all the real names in the sentence”… Do not destroy (general)

We admit that personal information was leaked and officially apologize… “Strengthening the anonymization process”

Users continue to ask for destruction… The hate controversy “looked like sympathy”

(Seoul = Yonhap News) Reporter Hyo-Seok Lee = Apologizes, acknowledging that ScatterLab, a startup that developed the AI ​​chatbot’Iruda’, “filtered the real name with an algorithm, but there was a part where the name remained depending on the context.” did.

Through a press release on the afternoon of the 12th, Scatter Lab said, “I have learned the text data of’Science of Love’,” and “Personal information such as the speaker’s name has been deleted.”

The science of love is an app that shows the level of affection by analyzing conversation patterns such as reply time when you enter the KakaoTalk conversation you have with your partner. It was launched by Scatter Lab in 2016.

Scatter Lab has revealed that it has developed Eruda using about 10 billion KakaoTalk conversations collected as the science of love.

Science users of dating have not clearly announced that Scatter Lab is using it to develop chatbots such as Eruda while collecting KakaoTalk conversations, and are pursuing a class action suit saying that they have not properly handled anonymization.

In the data that day, Scatter Lab explained, “It is impossible to identify an individual by combining DB sentences,” saying, “It is possible to respond to users with sentences contained in a separate database (DB) that is not linked to member information.” .

As the science of love, about 10 billion KakaoTalk conversations were collected, and 100 million sentences that could be used in Eruda were selected to create a separate DB, and since 100 million sentences were stored in an independent form through anonymization, Eruda’s remarks were combined. It explains that it is not possible to specify an individual.

However, Scatter Lab said, “Information such as real names, alphabets, and numbers in each sentence-by-sentence conversation was deleted through algorithms and filtering, but depending on the context, the name of the person remained.” “I apologized.

Scatterlab repeatedly apologized, saying, “After the launch of the service, the internal monitoring team was filtering and responding to the parts determined to be sensitive immediately after discovery,” he said. “However, we could not filter out all the anomalous methods using Hangul.”

They emphasized that “Reinforcement of real name/address filtering algorithms, enhancement of de-identification through random transformation of conversational data, and overall improvement of algorithms to prevent exposure of sensitive information will be updated.”

Scatter Lab did not disclose its intention to destroy KakaoTalk conversation data.

Science users of romance are demanding that the data be destroyed entirely, saying that Scatter Lab has inappropriately collected KakaoTalk conversation data and has not properly protected it.

Regarding the controversy over discrimination and hate against women, the disabled, LGBTI people, and black people, Scatter Lab put forward a position that “the user first appeared to be in harmony with the context of discrimination and hatred.”

Eruda selects an appropriate answer based on about 10 conversations exchanged with the user, and if the user attempts a conversation that may result in a hateful or discriminatory answer in the context, it may appear that they are in harmony in the process of attempting to achieve the conversation. This is the company’s explanation.

Scatter Lab commented, “We removed hate expressions and derogatory words from certain groups before service,” and commented, “We used keyword-based response as a realistic measure, but in the long run, we need a way to help AI algorithms learn right and wrong.” .

They added, “I think that AI can establish ethical consciousness or moral standards by itself if we learn more refined data with algorithms,” and added, “We will improve this with AI that contains universal values ​​in society.”

However, ScatterLab said that it was only sympathetic to users’ discriminatory conversation attempts. Did not explain.

In a press release, Kim Jong-yoon, CEO of Scatter Lab, said, “We stopped our first step (due to the suspension of the service), but we don’t want to stop our dream of creating an AI like a friend who can communicate as well as people. I will take it.”

[email protected]

Unauthorized reproduction-prohibition of redistribution>
2021/01/12 17:00 sent