This initiative is a 2024 technology impact assessment of safe and trustworthy AI conducted by the Ministry of Science and ICT and KISTEP. It evaluates the potential societal effects of healthcare AI, humanoid robots and brain-computer interfaces and recommends future development and policy directions. The assessment involved expert panels, a citizen forum and surveys of 125 experts and 1,000 citizens. It identified key issues and policy tasks concerning privacy, algorithmic bias, AI incident li
Initiative overview
With the rapid advancement of artificial intelligence and the emergence of generative AI systems such as ChatGPT, societal concerns regarding the safety and trustworthiness of AI systems are increasing. Key issues include the unauthorised use of personal data, algorithmic bias, unclear liability for AI-related accidents, and the opacity of decision-making processes.
As advanced AI technologies, including healthcare AI, humanoid robots and brain-computer interfaces, increasingly affect citizens’ lives directly, there is a pressing need for pre-emptive assessment and response strategies.
Objectives
Based on Article 14 of the Framework Act on Science and Technology, the assessment examines the anticipated impacts of safe and trustworthy AI technologies on the economy, society, culture, ethics and the environment. Its primary goal is to develop policy tasks that maximise positive impacts while minimising negative consequences.
The initiative focuses particularly on:
- securing data fairness;
- Explainable AI (XAI); and
- ensuring AI-system robustness.
Together, these areas are intended to foster an AI environment that citizens can use with confidence.
Implementation and institutionalisation
The assessment findings are reflected in relevant ministries’ national R&D planning, and policy measures are developed to mitigate identified negative impacts. Nine core tasks were selected, including:
- a Basic Act on Future Technology Safety;
- human-centred technology ethics and safety guidelines;
- governance for technologies using sensitive information;
- R&D for future technology safety;
- clinical safety systems for healthcare AI;
- standardisation of medical AI data;
- medical AI demonstration platforms;
- physical safety standards for robots; and
- safety management systems for human-robot collaboration.
These tasks will be formalised through legislative amendments, policy development and R&D projects. A feedback system has also been established under which relevant agencies participate in adjusting and managing key policy tasks for at least three years following the assessment.
Other relevant details
Evolution and development: The initiative forms part of the annual Technology Assessment series conducted since 2003. In 2024, Safe and Trustworthy AI was selected following proposals from experts and citizens through the Technology Recommendation Committee and preference surveys conducted across ministries. The assessment was carried out over six months by a nine-member steering committee, 19 experts working across three subcommittees, and a citizen forum of 15 participants. Big-data analysis and generative AI-supported issue analysis were introduced for the first time. Looking ahead, the framework will be restructured to focus on core issues associated with technologies expected to proliferate in the short to medium term, considering both their probability of occurrence and potential impact.
Results, outcomes and impacts: The survey identified five high-risk issues: unauthorised use or leakage of personal data; unclear liability for AI-related accidents; weakened human autonomy and social interaction resulting from AI dependence; job displacement and wider social changes associated with humanoid robots; and addiction or dependence associated with brain-computer interfaces. Nine of 17 priority tasks were selected and grouped into legislative, policy and R&D categories, with arrangements for institutional feedback. The analysis also indicated growing media attention to personal data and security from 2019 onwards, followed by a shift towards safety, trust and ethics after the emergence of ChatGPT. Expert analysis initially generated 79 issues and 100 potential tasks. Generative AI was used to support deduplication and priority-setting. The findings are expected to guide AI safety R&D, clinical safety systems for healthcare AI, robot safety standards and policies supporting a trustworthy AI ecosystem.
Challenges and lessons learned: Reflection rates declined from 77.8% in 2021 to 68.8% in 2023, partly because recommendations were sometimes abstract and implementation involved long time lags. Mechanical evaluation processes and limited incentives for ministries also hindered the identification of core issues and sustained follow-up. Generative AI was used to analyse 79 issues and 100 proposed tasks and identify core concerns. The NTIS and K2Base databases were searched to refine nine key tasks, while 1,125 survey participants, comprising 125 experts and 1,000 citizens, contributed to priority-setting. A key lesson is that stakeholder management should continue for at least three years rather than relying on one-off feedback. Issues should also be categorised as risks, uncertainties or ambiguities so that tailored responses can be developed. Successful technology assessments should cover a complete cycle of detection, analysis and feedback. Institutional incentives and assessment during the “midstream” stage, when technologies are sufficiently developed to assess but not yet fully embedded, are crucial for achieving meaningful policy impact.
