Kakao Sets AI Ethics Principles, Signs Safety MOU

Kakao announced on October 15 that it has established AI ethics principles and an implementation guide covering the planning and operation of its AI services, according to techm.kr. The principles, named SPIRIT, consist of six areas: safety, privacy, inclusiveness, accountability, user agency, and transparency and reliability.

Kakao defines the method for translating the principles into practical implementation as NORMA. The company plans to apply an AI ethics implementation checklist within the year.

Kakao also signed a strategic memorandum of understanding with the AI Safety Institute for joint AI safety evaluation and the establishment of an evaluation framework. Under the agreement, Kakao will provide its AI models and AI agents subject to safety evaluation and supply the infrastructure needed for conducting evaluations and developing evaluation tools. The AI Safety Institute will handle evaluation execution and methodology advancement.

The two organizations will pursue cooperation in three stages: language model safety evaluation, expansion to multimodal language models and AI agents, and joint development of evaluation tools.

Earlier Ethics Disclosure and Prior Cooperation

The ethics principles were first presented at the 'if kakao 26' developer conference held on October 13 at Kakao AI Campus. The principles and implementation guide were prepared in early October with seven external advisors from the AI & Technology Ethics Subcommittee, which was launched in July. Last year, Kakao conducted a joint safety evaluation of its own language model, Kanana-1.5-9.8B, with the AI Safety Institute.