14 August 2023
CO23111 | How ASEAN can Contribute to International AI Governance Frameworks
SYNOPSIS
As artificial intelligence is crucial to ASEAN’s goal of becoming a digital economy and society, learning from the practices of other countries and participating in international efforts to establish governance frameworks may help the region plan better for its upcoming ASEAN Guide on AI Governance and Ethics set to be released in 2024.
COMMENTARY
On 18 July 2023, United Nations Secretary-General António Guterres urged the UN Security Council to adopt a universal approach on artificial intelligence (AI) governance. This came a month after ASEAN member states announced that they will be developing the ASEAN Guide on AI Governance and Ethics by 2024.
These developments are significant as ASEAN has signposted a goal to become a digital economy and a digital society by 2025. Whether AI could advance or retrogress that goal could depend on how ASEAN will approach and handle the rapidly advancing technology. It is therefore worthwhile investigating other AI governance frameworks around the world, not only for ASEAN to benefit from best practices and lessons learned, but also to investigate differences that could act as obstacles to standardised AI regulation.
AI Governance Frameworks Around the World
While there has been some movement among individual ASEAN states in terms of AI regulation, progress has generally been slow-paced.
Singapore is the most technologically advanced country in the region, but its approach on AI governance appears to be cautious. From its perspective, there is still much to learn about the new technology before decisions on regulation should be made. Singapore has therefore focused on developing tools to understand AI more, beginning with the AI Verify Toolkit – the world’s first AI governance and software testing toolkit – that allows for technical tests and record process checks of AI models.
In June 2023, Singapore also established the AI Verify Foundation, which seeks to give direction to the future of international AI standards through global collaboration. The Foundation is expected to support the development and use of the AI Verify Toolkit to assess and address risks associated with AI. This allows Singapore to carefully consider its next steps, having learnt from the global community, before committing anything to legislation.
Thailand has been making headway as it has already launched a public hearing for its AI Promotion Act and published its draft Royal Decree on Artificial Intelligence System Service Business earlier this year. However, legally binding AI regulation has yet to materialise.
Indonesia’s 2045 National AI Strategy, formulated in 2020 and dubbed Stratnas AI, is also a draft awaiting the president’s signature to become a legal framework for AI in the country.
The Philippines and Malaysia are still at their planning stages while Cambodia has participated in expert discussions. The other ASEAN member states have not announced any plans for AI regulation as yet.
The European Union AI Act released in June 2023 is the first attempt in the world at horizontal AI regulation. The proposed Act has a risk-based approach, categorising AI development and use based on risk levels and applying guardrails for high risk and therefore absolutely prohibited practices and being more lenient with low to minimal risk activities.
There is provision for accountability (for both providers and users of the technology), which together with safety, transparency, traceability, non-discrimination, and care for the environment, are all top priorities. It is also stated explicitly that these systems must be supervised by humans and not by automation to avoid harmful consequences, with a hint at human accountability in case of catastrophe. At this juncture, the EU AI Act leaves for itself some room for improvement. AI providers are given leeway to conduct risk-self-assessment, and there is no mechanism for complaint or judicial redress yet.
Whereas the EU was lauded for being quick off the mark in AI governance, the United States has been criticised for lagging despite the fact that much of recent AI developments were accomplished by US-based companies.
The US Blueprint for an AI Bill of Rights was published in October 2022, a year after the White House Office of Science and Technology Policy (OSTP) announced the initiative. The document attributed the year-long delay to the need to “listen to the American public”. The US Blueprint takes a rights-based approach, with principles safeguarding civil rights and promoting democratic values for the development, use, and governance of AI systems. Like the EU Act, it remains non-binding and is not accountable for US government policy.
As for China, there is much to learn from its content-based approach to AI governance. In 2021, it began regulating recommendation algorithms and placed targeted restrictions on both algorithms as well as AI.
China released its draft Measures for the Management of Generative AI Services in April 2023, which upholds socialist core values, and explicitly states that content from AI must not contain materials that subvert state power, overturn the socialist system, or incite separatism. However, the Measures also provide protection from content on ethnic hatred and discrimination, mental and physical harm, and personal data and trade secrets breaches. Accountability is placed on providers in terms of ensuring legality and truthfulness, not just of results but also of sources of generative AI product training, something that is yet to be seen in Western measures.
Providers are also required to include mechanisms for handling user complaints, as well as for guiding users to properly understand and use AI-generated content. Regulation of use has been placed in the hands of providers – with no legal accountability for users – as users who violate terms will merely have their services either suspended or terminated. There was also no mention of public consultation, which probably contributes to greater flexibility in the drafting and provision of guardrails. Unlike the EU and the US though, China has departments in place to impose penalties on non-compliance by providers in accordance with Chinese laws.
Challenges to a Universal AI Framework
Achieving universal AI regulation in the region and around the world will take time to materialise. The EU has been attempting to have all its member states sign the encompassing EU AI Act, and its modest goal of achieving that by the end of 2023 reflects the challenges it faces in having member states on board in the governance of a technology that is rapidly developing and not well-understood yet.
One evident challenge would be geopolitical interests, as tension and competition between the US and China (in AI development, among others) would make it difficult to come to an agreement, and ASEAN countries may find it difficult to avoid taking sides. Political differences may also come into play, as democracies and autocracies might not agree on what is permissible or beneficial. Another challenge would be self-interest, as countries lagging behind in technological innovation may prioritise development over regulation, while those ahead might prefer not to be constrained by new rules.
What should ASEAN do?
This is not to say that all efforts for standardisation and unification will be futile. There are steps which ASEAN member states should take, individually and as a regional grouping, to be at the forefront of AI governance frameworks.
First, it will be advantageous for ASEAN to join the UN organisation that will oversee AI governance among states, although the establishment of this body may take time in view of the challenges discussed.
Second, ASEAN member states would, in the meantime, benefit from more regional cooperation, collaboration, and information sharing, especially among government entities overseeing the development, use and regulation of AI. Singapore’s AI Verify Toolkit may be a useful reference or starting point.
Third, where applicable, expert and public opinion in individual states should be analysed, not just for broader understanding of the technology, but also the specific contexts where AI will operate.
Lastly, learning from other existing frameworks, a balanced approach that handles risk, recognises rights, and oversees trustworthiness of content without stifling innovation may be the way forward in ASEAN’s consideration.
Despite the challenges, ASEAN’s participation in the global effort at AI governance is an important goal in case self-regulation, national regulation, and even regional cooperation fall short. It is necessary for ASEAN to move fast on this to capitalise on the region’s attractiveness as a destination for digital and high technological development and innovation, as well as foreign direct investment.
About the Author
Karryl Kim Sagun-Trajano is a Research Fellow at Future Issues and Technology (FIT), S. Rajaratnam School of International Studies (RSIS), Nanyang Technological University (NTU), Singapore.
As artificial intelligence is crucial to ASEAN’s goal of becoming a digital economy and society, learning from the practices of other countries and participating in international efforts to establish governance frameworks may help the region plan better for its upcoming ASEAN Guide on AI Governance and Ethics set to be released in 2024.
SYNOPSIS
As artificial intelligence is crucial to ASEAN’s goal of becoming a digital economy and society, learning from the practices of other countries and participating in international efforts to establish governance frameworks may help the region plan better for its upcoming ASEAN Guide on AI Governance and Ethics set to be released in 2024.
COMMENTARY
On 18 July 2023, United Nations Secretary-General António Guterres urged the UN Security Council to adopt a universal approach on artificial intelligence (AI) governance. This came a month after ASEAN member states announced that they will be developing the ASEAN Guide on AI Governance and Ethics by 2024.
These developments are significant as ASEAN has signposted a goal to become a digital economy and a digital society by 2025. Whether AI could advance or retrogress that goal could depend on how ASEAN will approach and handle the rapidly advancing technology. It is therefore worthwhile investigating other AI governance frameworks around the world, not only for ASEAN to benefit from best practices and lessons learned, but also to investigate differences that could act as obstacles to standardised AI regulation.
AI Governance Frameworks Around the World
While there has been some movement among individual ASEAN states in terms of AI regulation, progress has generally been slow-paced.
Singapore is the most technologically advanced country in the region, but its approach on AI governance appears to be cautious. From its perspective, there is still much to learn about the new technology before decisions on regulation should be made. Singapore has therefore focused on developing tools to understand AI more, beginning with the AI Verify Toolkit – the world’s first AI governance and software testing toolkit – that allows for technical tests and record process checks of AI models.
In June 2023, Singapore also established the AI Verify Foundation, which seeks to give direction to the future of international AI standards through global collaboration. The Foundation is expected to support the development and use of the AI Verify Toolkit to assess and address risks associated with AI. This allows Singapore to carefully consider its next steps, having learnt from the global community, before committing anything to legislation.
Thailand has been making headway as it has already launched a public hearing for its AI Promotion Act and published its draft Royal Decree on Artificial Intelligence System Service Business earlier this year. However, legally binding AI regulation has yet to materialise.
Indonesia’s 2045 National AI Strategy, formulated in 2020 and dubbed Stratnas AI, is also a draft awaiting the president’s signature to become a legal framework for AI in the country.
The Philippines and Malaysia are still at their planning stages while Cambodia has participated in expert discussions. The other ASEAN member states have not announced any plans for AI regulation as yet.
The European Union AI Act released in June 2023 is the first attempt in the world at horizontal AI regulation. The proposed Act has a risk-based approach, categorising AI development and use based on risk levels and applying guardrails for high risk and therefore absolutely prohibited practices and being more lenient with low to minimal risk activities.
There is provision for accountability (for both providers and users of the technology), which together with safety, transparency, traceability, non-discrimination, and care for the environment, are all top priorities. It is also stated explicitly that these systems must be supervised by humans and not by automation to avoid harmful consequences, with a hint at human accountability in case of catastrophe. At this juncture, the EU AI Act leaves for itself some room for improvement. AI providers are given leeway to conduct risk-self-assessment, and there is no mechanism for complaint or judicial redress yet.
Whereas the EU was lauded for being quick off the mark in AI governance, the United States has been criticised for lagging despite the fact that much of recent AI developments were accomplished by US-based companies.
The US Blueprint for an AI Bill of Rights was published in October 2022, a year after the White House Office of Science and Technology Policy (OSTP) announced the initiative. The document attributed the year-long delay to the need to “listen to the American public”. The US Blueprint takes a rights-based approach, with principles safeguarding civil rights and promoting democratic values for the development, use, and governance of AI systems. Like the EU Act, it remains non-binding and is not accountable for US government policy.
As for China, there is much to learn from its content-based approach to AI governance. In 2021, it began regulating recommendation algorithms and placed targeted restrictions on both algorithms as well as AI.
China released its draft Measures for the Management of Generative AI Services in April 2023, which upholds socialist core values, and explicitly states that content from AI must not contain materials that subvert state power, overturn the socialist system, or incite separatism. However, the Measures also provide protection from content on ethnic hatred and discrimination, mental and physical harm, and personal data and trade secrets breaches. Accountability is placed on providers in terms of ensuring legality and truthfulness, not just of results but also of sources of generative AI product training, something that is yet to be seen in Western measures.
Providers are also required to include mechanisms for handling user complaints, as well as for guiding users to properly understand and use AI-generated content. Regulation of use has been placed in the hands of providers – with no legal accountability for users – as users who violate terms will merely have their services either suspended or terminated. There was also no mention of public consultation, which probably contributes to greater flexibility in the drafting and provision of guardrails. Unlike the EU and the US though, China has departments in place to impose penalties on non-compliance by providers in accordance with Chinese laws.
Challenges to a Universal AI Framework
Achieving universal AI regulation in the region and around the world will take time to materialise. The EU has been attempting to have all its member states sign the encompassing EU AI Act, and its modest goal of achieving that by the end of 2023 reflects the challenges it faces in having member states on board in the governance of a technology that is rapidly developing and not well-understood yet.
One evident challenge would be geopolitical interests, as tension and competition between the US and China (in AI development, among others) would make it difficult to come to an agreement, and ASEAN countries may find it difficult to avoid taking sides. Political differences may also come into play, as democracies and autocracies might not agree on what is permissible or beneficial. Another challenge would be self-interest, as countries lagging behind in technological innovation may prioritise development over regulation, while those ahead might prefer not to be constrained by new rules.
What should ASEAN do?
This is not to say that all efforts for standardisation and unification will be futile. There are steps which ASEAN member states should take, individually and as a regional grouping, to be at the forefront of AI governance frameworks.
First, it will be advantageous for ASEAN to join the UN organisation that will oversee AI governance among states, although the establishment of this body may take time in view of the challenges discussed.
Second, ASEAN member states would, in the meantime, benefit from more regional cooperation, collaboration, and information sharing, especially among government entities overseeing the development, use and regulation of AI. Singapore’s AI Verify Toolkit may be a useful reference or starting point.
Third, where applicable, expert and public opinion in individual states should be analysed, not just for broader understanding of the technology, but also the specific contexts where AI will operate.
Lastly, learning from other existing frameworks, a balanced approach that handles risk, recognises rights, and oversees trustworthiness of content without stifling innovation may be the way forward in ASEAN’s consideration.
Despite the challenges, ASEAN’s participation in the global effort at AI governance is an important goal in case self-regulation, national regulation, and even regional cooperation fall short. It is necessary for ASEAN to move fast on this to capitalise on the region’s attractiveness as a destination for digital and high technological development and innovation, as well as foreign direct investment.
About the Author
Karryl Kim Sagun-Trajano is a Research Fellow at Future Issues and Technology (FIT), S. Rajaratnam School of International Studies (RSIS), Nanyang Technological University (NTU), Singapore.
