NAVER Unveils AI Safety Framework to Build Secure Sovereign AI Ecosystem
NAVER Unveils AI Safety Framework to Build Secure Sovereign AI Ecosystem
NAVER Unveils AI Safety Framework to Build Secure Sovereign AI Ecosystem
- Identifying “severe disempowerment of the human species” and “misuse” as potential risks associated with AI, NAVER has developed a response system to address these risks
- As it takes its Sovereign AI global, NAVER aims to enhance its safety framework by incorporating cultural diversity
June 17, 2024
After establishing the NAVER AI Ethics Principles in 2021 and the Consultation on Human-Centered AI’s Ethical Considerations (CHEC) in 2022, NAVER has now revealed a more detailed AI Safety Framework (ASF).
Identifying “severe disempowerment of the human species” and “misuse” as potential risks associated with AI, NAVER has developed a response system to address these risks
NAVER Corporation (CEO Choi Soo-yeon) announced NAVER ASF via its technology channel, “Channel TECH.” This framework is a comprehensive response system that identifies, evaluates, and manages potential risks throughout the development and deployment stages of AI systems.
NAVER ASF identifies “severe disempowerment of the human species” and “misuse” as potential AI-induced risks and has designed strategies to counter these risks. To minimize the “severe disempowerment of the human species,” where human influence over the AI system is lost, an “AI Risk Assessment Scale” is utilized to evaluate and control the system’s risk regularly. AI systems that perform at the highest level are classified as “Frontier AIs,” with their risks assessed every three months. An extra evaluation is conducted if the system’s capabilities increase rapidly by more than six times.
To address another potential risk of AI systems defined by NAVER ASF as “misuse,” an “AI Risk Assessment Matrix” will be applied. This matrix tailors risk management approaches based on the AI system’s intended use and the degree of safety measures required. For instance, AI systems designed for specific tasks, such as biochemical material development, are made accessible only to users with specific qualifications to reduce risks. AI systems necessitating extensive safety measures, regardless of their intended use, are withheld from deployment until their risks are adequately reduced through further technical and policy-based safety protocols.
Taking its Sovereign AI global, NAVER aims to enhance its safety framework by incorporating cultural diversity
NAVER intends to develop NAVER ASF into a culturally diverse safety. The company plans to collaborate with governments and enterprises outside Korea to codevelop Sovereign AI, which will involve identifying the risks particular to different cultural realms and advancing benchmarks to reflect specific cultural traits.
NAVER has already broadened its AI safety-enhanced training datasets, originally developed in Korean, to a global audience. At the “2023 meeting of the Association for Computational Linguistics (ACL),” a premier natural language processing conference held last year, NAVER showcased a study that introduced a vast new Korean dataset designed to reduce biases in AI regarding delicate social issues, including religion and ethics. This study also suggested data design protocols to modify similar datasets for various linguistic contexts, considering the differing social sensitivities across cultural regions. This contribution was well received by the academic community for its global AI community engagement.
NAVER CEO Choi Soo-yeon remarked, “With the creation of ‘HyperCLOVA X,’ a Sovereign AI developed in Korea, NAVER has proven that cultural and geopolitical contexts, as well as local understanding, not only influence AI’s performance but also its safety. As we expand our Sovereign AI initiatives into the global market, we are committed to NAVER ASF’s ongoing enhancement. Our goal is to actively foster a sustainable AI ecosystem where diverse regional cultures and values are respected and coexist securely.”
