Written by Hon-min Yau.

Image credit: AI Artificial Intelligence concept by Jernej Furman. / Wikimedia, license: CC BY 2.0.

In November 2022, the release of ChatGPT 3.0 by OpenAI amazed the world with its natural capability of communication and powerful knowledge of the social world. Since then, the potentiality of generative Artificial Intelligence (AI) technologies in society has come to people’s attention. In fact, the development of AI technology is a knowledge-intensive and capital-heavy activity, and much of the focus has been centred on resource-rich great powers, such as the US and China. Early in 2016, the US already announced the National Artificial Intelligence Research and Development Strategic Plan and substantially updated it in 2019 and 2023 to make AI a national priority. Likewise, in 2017, China also released the New Generation AI Development Plan and vowed to become a global AI innovation centre by 2030. However, either looking at AI from the geopolitical reason of tech competition or the political-economic perspective of establishing business dominance, people often forget that the AI pioneers of China, Lee Kai-fu, and famous CEOs of AI blue chips of the US Silicon Valley, either Jensen Huang from Nvidia or Lisa Su from AMD, all are originally from Taiwan. Hence, this article will deviate from a traditional great power perspective, review the historical development of AI capability in Taiwan, and examine how Taiwan plays a role in the AI industry.

Strategic Terrain of Global AI Industry

Generally speaking, the development of AI could be conceptually divided into three phases. The first phase was inspired by the British Scientist Alan Turing in the 1950s when he proposed the concept of “Computing Machinery and Intelligence,” which is categorized as strong AI, also known as Artificial General Intelligence (AGI). The main feature of this over-optimistic vision was to create a type of AI that replicates human logic of thinking, as proposed by scientists at the famous Dartmouth workshop in 1956, which marked the academic beginning of the AI field. Given that the human mind is more complicated than what machines could do at the time, by the 1980s, scientists eventually scaled down their ambition to less powerful AI, often categorized as weak AI. The second phase of AI systems, better known as expert systems, would be capable of answering questions in a particular domain. Taiwan did not play any noticeable role in both phases.

The third phase of AI development follows the objective of weak AI, and it came in the 1990s with the introduction of new techniques, namely neural networks, which enable machine learning and are later better known as deep learning. Instead of teaching machines how to “think” like a human mind, this is a process that machines could better “learn” to match between what is known and a most likely set of the predicted results. To allow machines to perform well in these imitation games, the advancement of three technological innovations in powerful processing power, huge quantity of data, and advanced algorithms are required. The successful case was in 2016 when the AlphaGo AI system outperformed South Korean professional player Lee Sedol in a board game called GO, which was traditionally considered that the human mind has the advantage since the complexity of the game is beyond the computation capability at the time. As this new progress of Artificial Intelligence does not particularly rely on natural resources or strategic material that often decides the success of an industry, small states, like Taiwan, could prevail as long as they have sufficient talents and good strategies.

Technical Development of AI in Taiwan

Sensing the incoming grand trend of AI development, Taiwan’s “Year 0” for AI came in 2017. At the time, AI-enabled devices such as smart speakers or smart TVs have been the focus of the consumer market, but other emerging AI-enabled services in fields such as object recognition, medical diagnosis, customer service, and vehicle self-driving were seen by Taiwan as new economic opportunities for Taiwan’s development. Traditionally, Taiwan has been strong in manufacturing semiconductor chips for the AI industry, which is capable of delivering powerful computation processing power. Still, the success of the homegrown AI industry requires a close synergy with two additional requirements, such as sufficient data for AI machine learning and good AI algorithms. Hence, Taiwan desperately needed to cultivate more talent to facilitate better AI development in these three fields. As a result of this thinking, the Ministry of Science and Technology (which was renamed the National Science and Technology Council in 2022) initiated the project “AI Grand Strategy for a Small Country” from 2017 to 2021 to build an AI local ecosystem by investing US$517.5 million. The project not only set up Research & Development centres for innovations of new AI services but also focused on setting up AI Infrastructure, such as Computer Data centres, for sufficient data and processing power. The project also provided funds to enhance Taiwan’s AI chip-making capability and organize AI competition events to inspire and cultivate the required AI talent. In addition, the Science and Technology Advisory Group of the Executive Yuan (Taiwan’s cabinet) established the AI Taiwan Action Plan 1.0 (2018-2021) to further broaden Taiwan’s AI capacity.

Traditionally, Taiwan was very robust in the semiconductor and hardware side of the ICT (Information and Communications Technologies) industry. Hence, Taiwan has a strong foundation in delivering the required hardware infrastructure for the AI industry. Nevertheless, Taiwan has no companies like Google, OpenAI, or Microsoft, who are known for building good AI software. In addition, it is even more important to develop customized AI services based on the foundation provided by these mature AI platforms. As such, AI Taiwan Action Plan 1.0 (2018-2021) endeavoured to intelligentize Taiwan’s existing industries by providing more financial resources to support their R&D activities. By 2021, Taiwan’s first AI strategy delivered fruitful results, such as building the TWIWANIA series of supercomputers, completing YOLOv4, a cutting-edge real-time object AI detection system, and establishing organizations like the Taiwan AI Academy and Artificial Intelligence Foundation to educate the younger generation.

Norms and Regulations of AI in Taiwan

Taiwan is not only working on enlarging its AI opportunities but also developing plans to mitigate its AI risks. In 2019, the Legislative Yuan first reviewed the proposal to establish an AI Development Basic Law to promote the AI industry while protecting individual privacy. However, the concern about AI has shifted rapidly from the promotion of AI to the concern of privacy, protection of secrecy, ethical principles of AI technology, intellectual property of AI-generated content, safety requirements of AI (to prevent harm to people), and security standards of AI (to reduce the vulnerability of AI systems). The Legislative Yuan had subsequent discussions in 2020 and 2022. Eventually, the Executive Yuan decided to draft the Artificial Intelligence Fundamental Act in the hope of regulating activities including but not limited to compliance requirements, testing & verification standards.

The creation of an overarching legislation requires time and effort in consensus building within Taiwanese society, but by 2022, the risk presented by generative AI in creating contents that have manipulative capability in documents, images, video, and voice is imminent. In addition, the use of generative AI services has the potential to endanger government secrecy as the protected data may be arbitrarily uploaded to these AI services by government employees for analysis. Hence, for the short-term effort, the Taiwan Executive Yuan quickly promulgated the Draft Guidelines for Use of Generative AI by Executive Yuan and its Subordinate Agencies in 2023 to manage the immediate risk of endangering data security.

For the long-term effort, Taiwan delivered a revised version of the AI Act, AI Taiwan Action Plan 2.0 (2023-2026), in 2023. While the previous version of the AI Taiwan Action Plan focused on establishing technical capability and capacity in AI, the revised version emphasized AI’s ethical standards and governance. In particular, given that commercialized AI services have generated various risks in domains such as finance, health, and transportation, as exemplified by the passing of the EU Artificial Intelligence Act on May 21, 2024, the National Science and Technology Council of Taiwan also vows to create an up-to-date umbrella regulation on AI by the end of 2024.

Conclusion

To sum up, Taiwan has a traditional strength in terms of providing AI infrastructure, including AI chip making and data centre building, based on its past reputation as a strong ICT industry. In addition, the government of Taiwan is also expediting its effort to commercialize more innovative and value-added AI services to aid and solve business problems in human society. However, Taiwan is insufficient in its foundation research of machine learning AI algorithms, such as coding deep machine learning models like the Generative Pre-trained Transformer (GPT) by OpenAI, as this requires years of talent enrichment, a mass repository of human resources, and a considerable number of financial capital. Furthermore, given that any machine-learning system will require a mass source of good data, Taiwan is also working on reinforcing its governance and regulation to facilitate a healthy AI ecosystem. Finally, as the pace of creating new social norms and regulations is always behind the development of AI technology, Taiwan also started thinking about how to better adapt itself to the advancement of technology while not hyping the challenges. As such, Taiwan is undoubtedly aware of both its strengths and weaknesses, but it is continuously investing in itself and looking to create more opportunities for the next generation of prosperity.

Dr Hon-min Yau is Director & Associate Professor of the Graduate Institute of International Security (GIIS), College of International & National Defense Affairs (INDAC), National Defense University, Taiwan. He received his Ph.D. from the Department of International Politics, Aberystwyth University. His research investigates the intermingled relations between security, policy, and technologies. He published a research article, ‘An Assessment of Cyberpower within the Triangular Relations of Taiwan-US-China and Its Implications’, in the International Journal of Taiwan Studies 2(2): 264–291.

This article was published as part of a special issue on ‘International Journal of Taiwan Studies: Contemplating AI & Taiwan Studies‘.

Adam Barnett is the coordinator of the Taiwan Insight-International Journal of Taiwan Studies Special Issue for August 2024.