As a result of introducing artificial intelligence (AI) to detect online sexual exploitation of children and adolescents, it was found that even slang expressed with initial consonants was identified, and the daily throughput per person in charge was increased by about 10 times.
On the 10th, the Korea Women's Human Rights Agency's Central Digital Sexual Crime Victim Support Center demonstrated the artificial intelligence (AI)-based child and youth online sexual exploitation detection and reporting system introduced in April this year to the press corps visiting the Ministry of Gender Equality and Family and disclosed the operational results.
The AI system was designed to overcome the limitation that it is difficult to keep up with the speed of spread using the traditional method, in which workers at the Central Digital Sex Crime Victim Support Center search keywords one by one for numerous posts posted on social networking services (SNS) to identify sexually exploitative material.
As a result of learning about 40,000 photos and posts of sexual exploitation keywords, detection and reporting manuals, and accumulated photos and posts established by the Central Digital Sex Crime Victim Support Center, it was possible to identify sexual exploitation materials with a high probability.
For example, in one post collected at 9 a.m. on December 16th of last year, there was an expression such as 'ㅈㄱㅁㄴ', which uses only the initial consonant for conditional meeting, and the AI system determined that this post was 100% sexual exploitation.
In addition to whether it was sexually exploitative, the AI system analyzed whether it preyed on children and adolescents, and whether it was a new post or a post that had been reported multiple times.
After completing this preliminary judgment, the post is sent to a worker at the Digital Sex Crime Victim Support Center, who reexamines the AI analysis and decides whether or not to report it.
With the introduction of the AI system, worker work efficiency has increased significantly. In the six months from April to September this year, the Central Digital Sex Crime Victim Support Center supported the deletion of a total of 96,210 cases of sexual exploitation of children and adolescents, a 112% increase compared to last year's performance (45,438 cases).
The average daily processing volume per worker also increased by 872%, from approximately 20.8 to 202.1.
The AI system also masked the collected photos, alleviating the psychological burden on workers who are repeatedly exposed to sexually exploitative material.
The number of digital sex crime victims who received support from the Central Digital Sex Crime Victim Support Center steadily increased from 7,979 in 2022 to 8,983 in 2023, 10,305 in 2024, and 10,637 last year.
As of September this year, 9,712 people are receiving support, and if this trend continues, the number of victims is expected to reach 13,000 by the end of this year.
The Central Digital Sex Crime Victim Support Center plans to further upgrade its AI system to respond to this growing risk of digital sex crimes.
Specifically, we plan to increase the number of platforms subject to detection and reporting from 74 to 139 by next year, and improve the model so that it can detect new and circumvented expressions by reading not only the words used in sexual exploitation but also the context and flow.
In addition, an AI system will be distributed to local digital sex crime victim support centers and an AI chatbot will be introduced that can provide necessary support information to victimized youth in real time.
However, as the rate of failure to respond to deletion requests from the Central Digital Sex Crime Victim Support Center increased from 24.4% in 2022 to 28.5% last year, industry cooperation is also becoming important in responding to digital sex crimes.
Kim Mi-sun, head of the Central Digital Sex Crime Victim Support Center, said, "Digital sex crime damage is increasing, especially among teenagers and those in their 20s who actively use SNS. To prevent this, the standards for platforms and chatting apps need to be strengthened."
