AI doing work in place of people: How can it be controlled? Kakao Corp. sets out ethical principles

[Financial News] As an era approaches in which artificial intelligence (AI) independently carries out tasks such as making reservations and searching for information on behalf of people, it is becoming increasingly important to establish safeguards to control AI behavior. In response, Kakao Corp. is setting up a system to assess in advance the authority granted to AI, the scope of its access to personal information, and emergency control measures. The company plans to identify risks from the planning through the operation of AI services and implement a related checklist within the year.
Kakao Corp. said on the 15th that it had unveiled new ethical principles and an implementation guide for autonomous AI agents at the developer conference “if(kakao)26,” held on the 13th at Kakao AI Campus.
The guide focuses on identifying in advance the risks that may arise as AI sets its own goals and performs tasks, and ensuring that people can control it. While conventional AI has focused on answering users' questions or generating content, autonomous agents can access external information and services to carry out multistep tasks. This, the company explained, has increased the need to assess whether the authority granted to AI is appropriate and whether actions that deviate from users' intentions can be prevented.
To this end, Kakao Corp. divided the risks posed by autonomous agents into five areas: control authority, goals, permissions, reference resources, and interactions. Taking into account situations in which AI acts contrary to a user's instructions or accesses information beyond the permitted scope, it linked response measures such as human oversight, emergency controls, permission controls, and data protection.
Kakao Corp. also drew up an “AI Ethics Implementation Checklist” to apply these standards during the actual development and operation of services. The checklist consists of three stages—identifying service attributes, preliminary risk screening, and reviewing and assessing control items—and is due to be implemented within the year. Its distinguishing feature is that it turns ethical principles into a concrete process for assessing risks specific to each service, rather than stopping at declaring principles.
The protection of children and adolescents is also among the key areas to be assessed. In addition to the “Digital Child and Adolescent Protection Checklist” it has operated since last year, Kakao Corp. plans to refer to the OECD and UNICEF's “5Cs,” a framework for classifying online risks to children, and incorporate risks specific to AI services into the new checklist.
The new ethical principles, “SPIRIT,” comprise six elements: safety, privacy, inclusion, responsibility, user agency, and transparency and reliability. Kakao Corp. used the three core values and seven principles of the “Republic of Korea's AI Ethics Principles” as key implementation standards and adapted them to the characteristics of Kakao Corp.'s services. It named the framework for putting these principles into practice across the organization “NORMA.”
The principles and guide were developed earlier this month with seven members of the external advisory group to the AI and Technology Ethics Subcommittee, which was launched last July. The external advisors reviewed alignment with domestic and international AI standards, while Kakao Corp.'s internal teams responsible for AI safety, personal information, information security, and services assessed the practical feasibility of implementation.
At the session “How Should We Respond to Autonomous Agents That Break Ethics?”, Ha Jin-hwa, Kakao Corp.'s senior AI communications lead, stressed that the greater AI's autonomy becomes, the more important it is to clarify human judgment and responsibility and reflect ethical principles in service design and operation.
Kim Se-woong, chair of Kakao Corp.'s AI and Technology Ethics Subcommittee, said, “As AI becomes more autonomous, we will make human judgment and responsibility clear and build AI services that users can use with confidence.”

[email protected] Choi Hye-rim Reporter