Dec 04, 2023 (East Asia Forum) - Generative artificial intelligence (AI) has captured the world’s imagination. It has also been greeted with alarm, with policymakers concerned about its control by non-state actors and the impact of AI systems on citizens within and across national borders.
Most AI experts agree that the world needs to work together to promote the best and prevent the worst. But China announcing its Global AI Governance Initiative two weeks before a UK-hosted AI Safety Summit and one day after the United States further tightened export controls over advanced computing chips raises questions about the effectiveness of multilateral efforts to develop trustworthy, inclusive and environmentally sustainable AI systems.
Regional coordination of AI governance is nowhere more crucial than in Asia.
With Asia facing one of its worst economic outlooks in half a century, the key to inclusive and sustainable growth in the region will be reforming the service sector to harness the digital revolution, including through the development of advanced AI systems. Coordinated regional arrangements for AI can also help mitigate the most acute risks of geostrategic competition between the United States and China while reducing the need for middle powers to choose sides.
Effective AI governance faces fundamental challenges. The concentration of power over AI inputs by the United States, China and a handful of their technology infrastructure firms is just one. Another problem is governments’ tendency to localise and protect key digital assets. Meanwhile, Asia’s women, rural residents, and indigenous populations remain systematically excluded from accessing the benefits of AI systems.
There are huge differences in state perspectives and capabilities for dealing with AI-related challenges, yet the region already possesses the raw ingredients required to shape a regional framework for AI governance. These include a wide variety of flexible digital policy tools and industry engagement strategies that can be upgraded and flexibly deployed.
A foundational challenge for AI governance in Asia is that a handful of US and Chinese technology infrastructure companies enjoy near-monopoly power over most key inputs. The impressive early performance of large language models (LLMs) shows they could become the foundational infrastructure on which AI applications rely. But LLMs depend on data and computation-intensive machine learning that only the best-resourced companies can maintain.
This signals a worrying ‘winner takes most’ environment. AI leaders benefit disproportionately from the learning and capital they accrue, further concentrating power. This concentration makes it difficult for new entrants to compete and public actors to ensure transparency and accountability of AI systems.
With power over AI inputs concentrated, some governments across the Asia Pacific are seeking to protect and localise their digital assets through national policy. Localisation measures have negative impacts on AI systems. Localisation reduces access to training data, starves innovation ecosystems and risks fragmentation of cybersecurity mechanisms.
The Regional Comprehensive Economic Partnership (RCEP) trade agreement mirrors this trend, with its chapter on e-commerce allowing data localisation carveouts on national security grounds. The United States has taken an even more active approach. Investments in onshore production of graphics processing units (GPUs), AI innovation ecosystems and export controls targeting high-end GPUs sold to China signal its intention to extend US technology companies’ AI advantages through localisation.
Absent a robust regional framework to counteract localisation, it will be difficult for potential AI competitors such as China, India and Indonesia not to respond in kind. Smaller and poorer countries with the least access to data, computational capacity and talent will be left with fewer options to participate in the AI industry.
Southeast Asia’s comparatively weak AI readiness risks the region’s digital divides becoming ‘algorithmic divides’. While broadband connectivity has increased, an estimated 61 per cent of ASEAN populations do not use the internet despite living within range of internet access. Several countries lack adequate data protection laws and AI strategies.
Governments, capital providers, small- and medium-enterprises (SMEs) and citizens can coordinate strategies that counterbalance concentration, localisation, and exclusion in AI systems.
Key to addressing concentration will be promoting new paradigms of data ownership and valuation that increase equity, including experimentation with data cooperatives and data unions. Capital providers can support the development of SME- and community-driven AI systems while reducing reliance on largescale proprietary AI models and centralised cloud computing infrastructure.
Regional coordination of third-party AI oversight can lower the prohibitive costs of regulation at the national level. Existing national policy tools offer starting points for a regional approach that places responsibility on technology firms. Singapore’s AI Verify Foundation is an encouraging public–private partnership that increases broad stakeholder participation in AI systems. A proposed global regulatory sandbox initiative could even begin in Asia.
Counterbalancing localisation can begin with updating existing bilateral, minilateral and multilateral trade agreements for cross-border data flows. Examining national security exemptions in multilateral trade rules can help distinguish which AI-relevant assets could be liberalised. The World Trade Organization’s joint initiative on e-commerce is a forum in which Asia Pacific nations can push to gain momentum. A regional interdependent standards body could ensure liberalisation of cross-border data flows does not compromise accountability.
To address exclusion, regulatory leaders can work with ASEAN and Pacific Island nations to strengthen regulations and AI strategies. SME financing and digital capacity building will be key to supporting equitable participation in regional AI ecosystems. Donors and development practitioners can also support locally led efforts to increase citizen participation and representation in AI systems and engagement with digital governance.
There are no easy answers to questions of concentration, localisation and exclusion in AI systems. But coordinated AI governance can create incentives for diverse regional stakeholders to actively steward AI systems while increasing transparency around risks.
In practice, AI governance will need to move as fast as the technology landscape is evolving.
Jacob Taylor is Fellow at the Brookings Institution’s Center for Sustainable Development.
