The rapid development and spread of advanced artificial intelligence (AI) has led to legislative responses around the world to find suitable and adaptable governance measures. On 15 July 2024, the National Science and Technology Council of Taiwan (NSTC) published the draft bill of the Basic Act on AI (人工智慧基本法草案; the Bill) for a 60-day public consultation. The Bill itself is available here. 

Key Takeaways

• The Bill establishes core principles for regulating the development and application of AI and outlines the government’s role in developing and governing AI and its related industries.
• There will be a review of existing laws and further adjustments to the responsibilities of various government agencies. It is not clear whether a new governmental agency will be created to regulate AI development and coordinate with other agencies after such a review, or whether an existing governmental agency will be designated to do so. Before such adjustments are made, the central competent authorities in charge of relevant industries will work with the NSTC to interpret and implement the provisions of the Bill.  
• The Bill promotes public-private cooperation and international collaboration and would establish a regulatory sandbox for research, development, and application of AI.
• The Bill also commits to education of AI literacy, building human capacity while preparing for labor market transformation.
• The Bill promotes the relevant agencies to establish an AI risk classification framework that interfaces with the international community.

Definition of “AI”

The Bill defines AI as “a machine-based system that has the ability to operate autonomously and, through inputs or sensing, and with machine learning and algorithms, can provide output such as predictions, contents, recommendations, or decisions influencing real or virtual environments for explicit or implicit objectives.” Such a definition is mostly in line with the current trends in AI laws and regulations in the European Union, the Organization for Economic Co-operation and Development (OECD), and the United States.

Core Principles

The Bill requires that the government shall follow seven core principles in promoting the development and application of AI:

Sustainable Development and Well-being

Referencing the Hiroshima Process Code of Conduct for Organizations Developing Advanced AI Systems, AI stakeholders shall engage in responsible AI management to benefit humanity and the planet. The Bill focuses on promoting social fairness and a sustainable environment. This includes narrowing the digital divide by providing suitable education and training to prepare the people for changes that will come with the development and application of AI. 

Human Autonomy 

Human rights and other civil rights and cultural values must be respected. There should be human supervision and respect for rule of law and human-centered values. The government should identify, assess, and reduce the risks of the use of AI. While promoting research and development and the application of AI through appropriate standards, guidelines, or directives, it should also assess potential weaknesses and abuse based on risk profiles to improve the verifiability of AI decision-making and human supervision. This principle references the OECD Recommendation on Artificial Intelligence, supporting autonomy and fundamental rights. 

Privacy and Data Governance

As the development of AI relies on vast amounts of data, key concerns in AI development relate to the collection, processing, and use of personal data, as well as data security and privacy. Referencing the US Blueprint for an AI Bill of Rights (“US Blueprint”) published in October 2022, the Bill commits to data minimization and protection of personal data while promoting openness and re-utilization of nonsensitive data. In the process of AI research and development and application, unnecessary collection, processing, or use of personal data should be avoided, and personal data protection should be promoted and incorporated into preset and design-relevant measures or mechanisms.

Security and Safety

Referencing the US Blueprint and Singapore’s Proposed Model AI Governance Framework for Generative AI (“Singapore’s MGF-Gen AI”), the development and application of AI should ensure the security and safety of AI systems and provide protection against AI-related security threats and attacks.

Transparency and Explain-ability

The decision-making capabilities of AI may materially impact stakeholders. Referencing EU Ethics Guidelines for Trustworthy AI 2019, the Bill requires suitable disclosure and labeling for AI-generated results to evaluate potential risks and impacts and thereby improve the trustworthiness of AI.

Fairness and Nondiscrimination

Referencing the US Blueprint, the development and application of AI must be conducted in a manner that to the greatest extent possible eliminates risks of bias and discrimination being incorporated into the algorithm. The AI-generated results should take into account the diversity of the society and must avoid discriminating against specific groups. 

Accountability

Assumption of corresponding responsibilities, including internal governance responsibilities and external social responsibilities, should be ensured. This principle is formulated by referencing the Singapore’s MGF-Gen AI.

Missions of the Government

As the basic law for AI, the Bill sets out the government’s role and missions to promote AI development under the core principles above. Further laws, regulations, and directives can be expected to implement these missions:

• The government should promote the development, application, and infrastructure of AI and allocate suitable resources, including providing tax or financial incentives, investment, funding, benefits, and counselling to AI-related industries;
• There should be a regulatory sandbox for the development and application of AI;
• The government should promote public-private collaboration and international cooperation in AI; 
• The government should buttress the quality and quantity of materials for AI use to ensure that training results maintain the country’s multicultural values and protect intellectual property rights;
• The Ministry of Digital Affairs (MODA) and other relevant agencies may provide or recommend tools or methods for assessment and verification of weaknesses and abuse;
• The MODA should refer to international standards or standard development of AI data security protection, risk classification, and management, and promote an AI risk classification framework that interfaces with the international community. The government may formulate classification standards based on a risk-based approach;
• The government should establish a data openness, sharing, and reuse mechanism to improve the availability of data used by AI and regularly review and adjust relevant laws and regulations;
• The government should prevent the application of AI from causing damage to citizens’ lives, bodies, freedom, or property or to safety, social order, and the ecological environment, as well as from violating relevant laws and regulations by causing or generating conflicts of interest, bias, discrimination, false advertising, or misleading or falsified information, etc.;
• The government needs to reconcile the regulatory interpretation and implementation of relevant laws and regulations so as not to hinder the new technologies within the core principles provided under the Bill;
• The government should increase the trustworthiness of AI applications through mechanisms such as standards, verification, testing, labeling, disclosure, traceability, or accountability based on AI risk classification and establish relevant conditions, responsibilities, relief, compensation, or insurance for AI applications. However, to avoid affecting the freedom of academic and industrial research and development, activities conducted before application of AI only need to comply with the core principles and shall not be subject to the regulations related to application accountability.

Conclusion

This is not the first attempt of the Taiwan government to grapple with regulating AI. In 2023, Taiwan’s Executive Yuan (the executive branch of the central government) published directives on the use of generative AI for itself and its subordinate agencies. Political parties submitted their own versions of the draft AI bills to the Legislative Yuan (the legislative branch of the central government) but made no advancement of such bills. Other directives are currently under consultation with the relevant industries. Nevertheless, the Bill serves as a new effort by the government to establish core principles for regulating and governing AI, which will lay a foundation for all the laws, regulations, and directives that follow. Industry players should closely follow this development and prepare to align their operations with the evolving regulatory landscape.