AI 안전을 위한 규제와 거버넌스

Abstract

AI safety is a policy goal to ensure the safety of the people from various harmful situations or problems that may appear in the process of developing or using AI. Until now, voluntary efforts have been required to secure safety through AI ethics. However, as in the EU AI Act, it is shifting beyond ethical principles to a regulatory focus. There are various methodologies for achieving AI safety. This is because various matters for safety can be embodied systematically or technically. Above all, it is securing transparency in the process of using AI. It is to ensure fairness without bias against results. The social value that can be obtained through AI safety is the safety of the people and the reliability of AI. To this end, the transition between no-fault liability and the burden of proof can be an effective safety measure. The government should present specific policy goals for AI safety. Whether to establish a regulation-oriented goal or a principle-based self-regulatory goal may vary depending on social consensus on AI safety. This is because the goal and task of AI safety is to ensure the safety of the people in various problem situations that may arise in the process of using AI. From a technology-neutral point of view, a concrete discussion should be made on whether or not the business model is regulated by technology. For this, continuous monitoring of civil society is necessary. This is because although technology was developed in good faith, misuse in the process of using it may appear unexpectedly. The review of the framework for AI risk management is expected to play a meaningful role in AI risk management and safety evaluation. In addition, as an institutional and technical means to ensure AI safety, the right to request explanation and explainable AI(XAI) will be able to reduce the risk of the black box effect. Above all, governance for safety can be more concrete by the participation of public-private cooperation and civil society that conducts continuous monitoring. In addition, legislation of AI Act needs to be supported in order to maintain a continuous and consistent safety policy. Safety according to technological advancement is a must, not an option, and this is especially true for AI, an unknown technology.