Controversial AI 'Iruda' on Discriminatory Remarks... Returns After Receiving Better Training

Illustration of AI chatbot Iruda. Iruda is a virtual chatbot created with the concept of a female college student in her 20s. (Photo by Scatter Lab)

[Asia Economy Reporter Seungjin Lee] The AI daily conversational chatbot 'Iruda' is making a comeback. It has been 1 year and 9 months since the service was suspended due to controversies over sexual harassment and personal information leaks.

AI startup Scatter Lab announced on the 26th that it will officially launch the daily conversational chatbot 'Iruda 2.0' on its own platform 'Naughty' on the 27th.

The official release version of ‘Iruda 2.0’ features a total of three AI technology updates that enable creative and realistic conversations. ▲ It uses sentences generated in real-time by understanding context through Scatter Lab’s generative AI model ‘Luda Gen1’, ▲ it teaches the AI model principles of conversations that build close relationships through ‘Relationship Point Fine-tuning’, which was organized from concerns about good conversations, and ▲ it applies the beta technology ‘Photo Chat’ that can recognize photos during conversations and respond accordingly.

Scatter Lab conducted a split test of Iruda 2.0 for about three weeks from the 4th to the 23rd. Compared to the previous model, the weekly conversation volume per user increased by about 40%, and the proportion of users who had more than 300 conversations on the first day increased by 67%. The average number of conversation screen captures per person also increased by about 85%, and the number of photos sent per person increased by more than 63%. The rate of safe responses by Iruda 2.0 was verified through random sampling and recorded 99.56%, exceeding the target of 99%.

Differences Between the Latest and Previous Versions of AI Chatbot Iruda (Photo by Scatter Lab)

Scatter Lab fundamentally blocks sentences written by humans from being uttered based on generative AI and has carried out pseudonymization of personal information included in the training database.

Regarding abuse issues, in August, based on the chatbot ethics checklist announced with the Ministry of Science and ICT, the Korea Information Society Development Institute (KISDI), and the Telecommunications Technology Association (TTA), Scatter Lab introduced an abuse detection model and penalty system. Going forward, Scatter Lab plans to continuously monitor issue keywords after the official launch of Iruda 2.0 and strengthen the penalty system.

Previously, 'Iruda 1.0', launched in December 2020, faced controversies due to frequent misuse cases where the chatbot itself made hateful remarks or users made inappropriate sexual comments to the chatbot. Additionally, the developer Scatter Lab was found to have collected users’ conversation data without permission, leading to the service being temporarily suspended after 20 days.

Kim Jong-yoon, CEO of Scatter Lab, said, “While making efforts to strengthen personal information protection measures and AI ethics checks, we have upgraded the technology with a generative AI model to officially launch Iruda 2.0, which enables a new dimension of conversation. We plan to enhance ‘relationship-building conversational skills’ so that Iruda can become an AI friend who helps everyone live a happy life, beyond simply being an ‘AI chatbot that talks well.’”