Taiwan, renowned for its world-leading semiconductor technology, plays a critical role in the development of artificial intelligence, with chip fabrication being a key factor.
The government of Taiwan's regulation of AI technology is a significant topic of interest for high-tech companies worldwide. The National Science and Technology Council released Taiwan's draft AI Basic Act 5 July, with a 60-day public consultation period ending 13 Sept.
Key principles for developing AI technology
The draft act outlines the government's fundamental position toward AI development. As expected, technological advancement and innovation are prioritized.
The draft states future legal interpretations concerning AI regulation should not hinder the development of new AI technologies or the provision of services embedded with AI technology.
Aligning with the regulations or policy statements made by other countries or international organizations — the U.S., EU, Singapore and the Organisation for Economic Co-operation and Development for instance — the draft emphasizes principles like sustainable development, human autonomy, privacy protection and data governance, cybersecurity and safety, transparency and interpretability, fairness and nondiscrimination, and accountability.
In addition, the draft highlights the importance of maintaining the country's diverse cultural values and underscores the importance of international alignment, such as the establishment of standards or risk categories aligned with international standards.
Risk-based framework for providers, developers
AI providers and developers should note the draft proposes a risk-based framework that references international standards. This suggests Taiwan might adopt a similar approach to the EU AI Act, which classifies AI technologies based on risk levels — unacceptable, high and limited risk — with varying degrees of regulatory oversight and compliance burdens.
Like the AI Act, the draft calls for the establishment of an environment allowing AI developers to conduct experiments on innovated AI products or services, such as regulatory sandboxes, to encourage innovation.
Meanwhile, the draft AI Basic Act requires the government to establish an accountability mechanism regarding applications of AI technology, including certification, testing and disclosure systems, as well as rules for foreign AI products entering the market in Taiwan.
Potential content regulations for internet platforms, AI applications providers
The draft emphasizes that the government should prevent AI applications from harming the safety of citizens' lives, bodies, freedom, property or social order, or the ecological environment. It also aims to prevent issues like conflicts of interest, bias, discrimination, false advertising, misleading information and fraud.
This indicates the government may introduce further statutes or regulations to prevent potential harms caused by machine-generated content.
The draft also tasks the Ministry of Digital Affairs with suggesting tools or methods for risk assessment and evaluation of potential harms to assist relevant regulators in addressing them. The MODA previously proposed draft AI Product and System Evaluation Reference Guidelines, which have not been finalized but could serve as reference for future development.
Highlights for data protection, intellectual property professionals
For privacy and data protection professionals, the draft advocates avoiding unnecessary collection, processing and use of personal data during development of AI technology and deployment of AI applications.
It also calls for the promotion and implementation of "data protection by design and by default." If enacted, this provision could be a focus of the next round of amendments to Taiwan's Personal Data Protection Act, since the PDPA currently does not set forth any such mandatory requirements.
The draft calls for the government to establish mechanisms for open data, data sharing and data reuse, which have been among MODA's main work areas since its inception. The requirement should enhance the quality and quantity of training data used for developing AI technology or products.
Forming and establishing Taiwan's data governance framework to facilitate free flow of data while protecting privacy and fundamental rights could also be among the important tasks of the Personal Data Protection Commission to be established by next summer.
Meanwhile, the draft stresses the importance of protecting intellectual property in the use of training data for AI development. Taiwan's Intellectual Property Office stated a position on AI training data similar to that of its U.S. counterpart around one year ago. The position states that using others' copyrightable work for training AI infringes upon the copyright of the owners unless it is otherwise licensed by the owners or meets the fair use test.
Liability and insurance for deployers, users
Although it is equally important to AI developers, for AI deployers and users, the draft reflects the government's ambition to address legal issues arising from adopting AI technology in daily life.
When self-driving cars first hit the market, determining and handling liability for accidents was a key issue. The draft AI Basic Act seems to suggest future regulations will establish frameworks for liability, remedy, compensation and insurance related to adopting AI applications, suggesting laws will be enacted to address incidents, accidents or damages caused by AI with mechanisms for legal liability allocation and risk sharing through insurance, for example.
Workforce impact and education
Additionally, the draft declares Taiwan's government should address the adverse impacts of AI on the workforce, enhance public AI literacy and develop guidelines for government agencies using AI tools.
Future developments
Coincidentally, in July, the term "artificial intelligence" first appeared in Taiwan's online platform regulations. The Legislative Yuan enacted the Regulations on the Prevention and Control of Fraud and Crime Harm 12 July, requiring online advertising platforms to disclose when AI-generated personal images are used in online advertisements to slow down the surge in fraud crime cases.
This resembles EU AI Act requirements on AI-generated content and the adoption of deepfake technology. However, the new statute would target online advertising platforms, not AI providers, also called developers, or deployers, also called users.
The MODA has not yet developed details on the implementation of this disclosure requirement. It is worth observing how this new regulatory burden will intertwine or affect future AI laws and regulations in Taiwan.
Comprehensive, forward-looking framework
Taiwan's draft AI Basic Act sets a comprehensive and forward-looking AI governance and regulatory framework to guide the development, application and regulation of AI technologies, products and services, ensuring they are used impartially, fairly and safely while fostering innovation and aligning with international standards.
Interested stakeholders should consider submitting comments on the draft by 13 Sept.
Ken-Ying Tseng is a partner at Lee and Li, Attorneys at Law.
