The government of Singapore has released an updated version of its national artificial-intelligence (AI) strategy, termed National AI Strategy 2.0 (NAIS 2.0). The original 2019 strategy, which NAIS 2.0 builds upon, was largely successful, with Singapore ranked third in the Global AI Index 2023 and second in the Government AI Readiness Index 2022.
The new strategy underscores the country’s commitment to being at the forefront of developing and adopting the latest AI models, including large language models (LLMs). While more than 60 countries have now launched their respective national AI strategies, Singapore continues to set itself apart by being one of the select few to embark on a second iteration.
NAIS 2.0 signals a significant shift in Singapore’s approach to AI. While the 2019 AI strategy treated AI as an emerging technology to be trialled in specific testbeds, the new strategy views AI as an enabling technology that should be deployed across all sectors.
This evolution in Singapore’s AI strategy highlights three key approaches that can inform AI strategies in other countries.
Anchoring LLMs in Local Languages and Contexts
Alongside the release of NAIS 2.0, AI Singapore – a national programme set up in 2017 to enhance Singapore’s AI capabilities – launched SEA-LION (Southeast Asian Languages in One Network), a family of LLMs that are foundational to all generative-AI development. Trained on data sets in 11 regional languages, SEA-LION’s models will cater better to South-East Asian contexts.
Although LLMs can understand and generate language, their ability to generate contextually appropriate responses is dependent on whether they are trained on input data that is from the relevant (i.e. local) contexts. Yet much of the data that existing LLMs such as ChatGPT are trained on are largely from economically developed, Westernised contexts – leading to accusations of cultural biases at best, and the promotion of certain values at worst.[_]
The launch of SEA-LION thus heralds a potentially significant shift in the way AI is developed and deployed in the Asian context. Similar developments, for instance in Sweden,[_] strongly suggest that LLMs should be trained for and tuned to local contexts in order to be most effective.
Developing an Agile, Pragmatic Approach to AI Regulation
Over the past few years, Singapore has developed a series of voluntary, non-binding guidelines and frameworks to inform its AI-governance approach. These include AI Verify (the world's first AI-testing framework and toolkit),[_] the Model AI Governance Framework[_] and the Advisory Guidelines on the Use of Personal Data for AI.[_] In alignment with its focus on building a trusted ecosystem, the country will continue to develop capabilities in privacy-enhancing technologies (PETs), enabling businesses to access data sets while upholding data privacy and effectively expanding the pool of data from which business can derive useful insights.
The rapid pace of AI’s development and our evolving understanding of its risks and beneficial use cases have underscored the importance of policymakers striking the right balance between protecting public interests and supporting innovation.[_] Although largely orchestrated and led by the government, AI development in Singapore is often carried out in tandem with the private sector to gain buy-in and facilitate knowledge sharing, ensuring the country remains at the cutting edge of global AI advancement.
To account for new knowledge and risks that emerge, these governance frameworks will be regularly reviewed and adjusted. Coupled with their “living” nature,[_] these frameworks allow for a degree of experimentation and exploration – thereby granting the government considerable agility within its regulatory approaches.
Creating an Ecosystem That Can Generate Value From AI
Singapore’s new AI strategy sets out a multi-pronged approach to building the critical infrastructure needed to generate economic value from AI.
First, by looking to deepen its substantive partnerships with major compute players and engaging with global multi-stakeholder fora, the country seeks to secure local access to compute capacity and to partake in global discussions about AI ethics and governance.
Second, to bolster the growth of the local AI research scene and start-up ecosystem, the government is partnering with Google Cloud to launch the AI Government Cloud Cluster (AGCC).[_] This dedicated cloud-computing environment will provide critical access to an AI technology stack, including pre-trained generative-AI models.
Third, to build its pipeline of future tech talent, Singapore has announced plans to increase its AI workforce to 15,000 (a three-fold increase from the current pool) and to increase the availability of AI training programmes and resources.
Singapore’s agile regulatory environment and key strengths in government policies, research and development, human capital, data connectivity and cloud infrastructure collectively contribute to its thriving AI ecosystem.
Countries that have experienced early and broad adoption of AI share a common feature: a national strategic direction for AI defined by their governments. Around 70 per cent of countries with an AI strategy are developed economies, and their priorities tend to include the development of ethical AI frameworks, investment in education and workforce, and global collaborations to sustain their AI leadership.[_] In contrast, developing economies tend to prioritise capacity building, fostering local talent and seeking international collaborations to bridge the AI technology gap.[_]
Singapore’s new AI strategy reflects both sets of priorities, and so has the potential to inform policymaking discussions elsewhere. While the emphasis on localised LLMs mirrors a growing international consensus, the country has also demonstrated the potential of voluntary guidelines and “living” frameworks as effective pillars of an agile AI governance approach. Combined with multi-pronged efforts to build a trusted AI ecosystem that’s geared towards generating value, Singapore’s latest AI strategy serves as a noteworthy demonstration of what effective AI governance can look like on the world stage.