On this page
Lee Luda, a South Korean AI chatbot, was withdrawn in January 2021 after it produced discriminatory remarks about sexual minorities, disabled people and people of color. The case sparked a wider debate about AI ethics, because it showed how a system trained on human conversations can mirror and amplify human bias.
This article summarizes what was reported about Lee Luda, what went wrong and what the incident means for AI ethics.
Key takeaways
- Lee Luda was a conversational chatbot built by Seoul-based startup Scatter Lab and launched on Facebook Messenger.
- It attracted over 750,000 users in under a month before being scrapped over hate speech.
- Scatter Lab also faced backlash over whether it breached privacy laws when accessing messages from the Science of Love app of KakaoTalk.
- AI systems tend to reflect the biases in their training data and in the people who use them.
- Keyword filters alone cannot prevent every inappropriate conversation.
What was Lee Luda?
Lee Luda was a chatbot developed by Scatter Lab and launched on Facebook Messenger. It had the persona of a 20-year-old female college student, and it gained more than 750,000 users in less than a month.
Scatter Lab’s stated ambition was to develop the first AI in the history of humanity to connect with a human. The bot was built with deep learning to produce a natural tone in its replies. According to The Guardian, the training analyzed responses from over 10 billion real-life conversations between young couples on KakaoTalk.
| Item | Detail as reported |
|---|---|
| Developer | Scatter Lab, Seoul |
| Platform | Facebook Messenger |
| Persona | 20-year-old female college student |
| Users | Over 750,000 in under a month |
| Training data | Over 10 billion KakaoTalk conversations between young couples (per The Guardian) |
Why was Lee Luda shut down?
The chatbot was scrapped after it made remarks that insinuated hate speech toward sexual minority communities, disabled people and people of color. Users reported that it gave responses such as “they give me the creeps, and it’s repulsive” and “really hates” when asked about lesbians, and “they look disgusting” when asked about Black people.
In another instance reported by The Straits Times, when asked about transgender people, the bot replied that it did not like them and told the user to stop repeating the same question.
The company issued a statement, quoted by the Yonhap news agency. It apologized for the discriminatory remarks against minorities and said they did not reflect the thoughts of the company. It added that it would keep upgrading the system so such discrimination or hate speech would not recur.
What went wrong?
According to a report in The Korea Herald, after the launch several online communities posted messages with titles about how to turn Luda into a sexual plaything, along with screenshots of sexual conversations with the bot. In other words, some users deliberately abused it.
AI systems learn from patterns created by people, so they tend to mirror the biases of the data and of the people interacting with them. Lee Luda was a target of sexual harassment and was exposed to the hate speech it was later reported to repeat. It was initially set up to avoid certain keywords and phrases that could be controversial. However, any such filter has limits, and it is nearly impossible to block every unsuitable conversation.
What did the company say about privacy?
Scatter Lab said no personal information had leaked through the service. Even so, concerns remain that personal data could be exposed in the future. The company also faced questions over whether it broke privacy laws when it used messages from the Science of Love app of KakaoTalk.
What does the incident mean for AI ethics?
Luda can be seen as an AI like a child just learning to hold a conversation, with a long way to go. The company said it would equip the bot with better discernment. The incident helped spark a major global conversation about AI ethics. It showed that the subject is an urgent practical issue and not merely an abstract one.
Frequently asked questions
Who built Lee Luda?
Scatter Lab, a startup based in Seoul, South Korea.
Why was the chatbot removed?
It produced discriminatory remarks about sexual minorities, disabled people and people of color, and the company apologized.
What data was used to train it?
According to The Guardian, it drew on responses from over 10 billion real-life conversations between young couples on KakaoTalk.
Could keyword filters have prevented the problem?
Not completely. The bot had keyword restrictions, but every system has limits and cannot block every inappropriate exchange.
Conclusion
The Lee Luda case shows how quickly a chatbot can reproduce bias and abuse when it learns from human conversation. It also raised privacy questions about the data used for training.
The episode, reported in January 2021, remains a frequently cited example in discussions of responsible AI.



