Taiwan's New AI Ten Major Constructions: Vision or Slogan? A Whitepaper

-

 

Taiwan's New AI Ten Major Constructions: Vision or Slogan? A Whitepaper

Executive Summary

Taiwan's "New AI Ten Major Constructions" initiative aims to position the nation as a leader in the global AI landscape, with ambitious goals such as achieving Asia's second-largest computing power, developing sovereign AI, and empowering a million enterprises. However, the initiative faces significant challenges, including the need for practical implementation, data availability, and legislative support. This whitepaper, inspired by insights from Rock Tsai, Chief Information Officer of Taiwan Mobile and Chairman of the IMA, explores whether this initiative is a transformative vision or merely a slogan. It delves into the concept of "Sovereign AI" versus "AI Sovereignty," proposes strategies for cultural integration through initiatives like the "Taiwan Tongues" project, and re-evaluates the approach to building computing infrastructure, advocating for a balanced, pragmatic, and globally integrated strategy.

Introduction

The "New AI Ten Major Constructions" initiative, announced by the Taiwanese government, seeks to leverage AI to drive economic and technological advancement. Unlike the infrastructure-heavy "Ten Major Construction Projects" of the 1970s, this modern initiative focuses on soft infrastructure, such as AI adoption across industries, national computing power, and sovereign AI development. However, questions remain about its feasibility and the practical steps needed to translate ambition into reality. This whitepaper examines the initiative's core components, challenges, and strategic recommendations, drawing on insights from industry leaders and ongoing projects like Taiwan Tongues.

Sovereign AI vs. AI Sovereignty

Understanding Sovereign AI

The concept of "Sovereign AI," popularized by NVIDIA's CEO Jensen Huang, refers to a nation's ability to develop and control its own AI systems, encompassing both software (AI models) and hardware (computing infrastructure). It is likened to a national deity—culturally resonant, locally controlled, and capable of serving national interests. For Taiwan, this translates into building a robust AI ecosystem that reflects its values, language, and cultural identity.

The Case for AI Sovereignty

Rock Tsai argues that the term "AI Sovereignty" may be more appropriate than "Sovereign AI." While Sovereign AI implies ownership of a world-class AI model, AI Sovereignty emphasizes a nation's ability to regulate and influence AI systems operating within its borders, ensuring alignment with local laws, values, and cultural norms. This approach is more achievable in the short term, given the commercial nature of AI development and the dominance of global players like OpenAI, Google, and Meta.

Strategies for AI Sovereignty

To achieve AI Sovereignty, Taiwan can adopt a three-tiered approach:

  1. Market-Driven Diversity: Allow market mechanisms to foster diverse AI applications, respecting Taiwan's open and free society.

  2. Regulatory Oversight: Enforce regulations ensuring AI systems used in Taiwan comply with local laws and values, particularly in sensitive sectors like finance and telecommunications.

  3. Cultural Integration: Proactively embed Taiwanese culture, language, and values into global AI models during their training phase, making compliance with Taiwan's norms inherent rather than enforced.

The Taiwan Tongues Project: Embedding Taiwanese Culture in AI

Overview

The Taiwan Tongues project, initiated by the IMA under Rock Tsai's leadership, aims to create a comprehensive Taiwan Common Data Set to train AI models with Taiwanese linguistic and cultural nuances, particularly in Traditional Chinese, Taiwanese Hokkien, and Hakka. This initiative addresses the current gap where global AI models, often trained on Simplified Chinese or English corpora, fail to capture Taiwan's unique identity.

Key Components

  1. Data Collection:

    • Literary Contributions: Over 60 writers have donated high-quality Taiwanese Hokkien texts, increasing the available corpus from 4 million to 10 million words in three months.

    • Citizen Co-Creation: A Taiwan-specific encyclopedia, modeled on Wikipedia, encourages public contributions in Traditional Chinese, Taiwanese Hokkien, and Hakka, ensuring open-source, consensus-driven content.

  2. AI Training and Fine-Tuning:

    • Develop open-source datasets and fine-tuning frameworks to adapt global AI models (e.g., Google’s Gemma 3) to Taiwanese contexts.

    • Use iterative translation and review processes to enhance AI model accuracy in local languages.

  3. Evaluation Mechanism:

    • Introduce the "Taiwan Tongues Goal," a benchmark to assess how well AI models understand and reflect Taiwanese culture and language.

Impact

By providing global AI developers with access to a standardized, high-quality Taiwanese dataset, Taiwan can influence the training of international models, ensuring they inherently understand Taiwanese culture. This approach not only enhances AI Sovereignty but also fosters innovation by lowering barriers for local startups to access quality data.

Rethinking Computing Power: Local vs. Global Strategies

The Goal of Asia’s Second-Largest Computing Power

Taiwan's ambition to achieve Asia’s second-largest computing power is a key pillar of the New AI Ten Major Constructions. However, the high cost of land, electricity, and cooling in Taiwan raises questions about the feasibility of building extensive local GPU-based infrastructure.

Challenges of Local Infrastructure

  • Economic Viability: Taiwan’s high electricity and land costs make local data centers less competitive compared to regions with cheaper resources (e.g., near hydroelectric plants or in colder climates).

  • Rapid Obsolescence: AI hardware, particularly GPUs, becomes outdated within 2–3 years, making large-scale local investments risky.

  • Idle Capacity: Building excess capacity for peacetime use risks waste, while wartime scenarios require strategic reserves that may remain unused.

Alternative Approach: Hybrid and Global Infrastructure

Rock Tsai proposes a hybrid model combining local and global computing resources:

  1. Local Baseline Capacity: Invest in local GPU and CPU infrastructure for consistent, predictable demand, ensuring cost-effectiveness over 2–3 years.

  2. Cloud-Based Scalability: Rent cloud-based GPU resources from global providers (e.g., AWS, Azure, Google Cloud) for peak demand, avoiding overinvestment in local hardware.

  3. Strategic Overseas Investments: Build or lease computing infrastructure in cost-effective regions (e.g., near renewable energy sources or in colder climates) to optimize costs while maintaining access during geopolitical disruptions.

  4. Military-Civilian Dual Use: Design computing infrastructure to serve commercial needs in peacetime while allowing government requisition during emergencies, similar to Taiwan’s historical vehicle mobilization system.

Policy Recommendations

  • Subsidies for Local Investment: Offer subsidies for companies building local computing infrastructure, contingent on providing a portion of capacity for public or startup use at reduced rates.

  • Wartime Preparedness: Establish a clear framework for requisitioning private computing resources during national emergencies, ensuring alignment with defense needs.

  • Global Collaboration: Partner with international AI companies to integrate Taiwanese datasets into their models, enhancing global AI’s alignment with Taiwan’s interests.

Data Privacy and Ethical Considerations

Challenges in the Telecommunications Sector

As a telecommunications provider, Taiwan Mobile faces strict regulations under Taiwan’s Constitution and the National Communications Commission (NCC), prohibiting the use of customer data (e.g., call logs, SMS) for AI training without judicial authorization. This contrasts with social media platforms like Meta or Google, which leverage user-generated content under less restrictive agreements.

Opportunities for Ethical AI

  • Customer Consent: Use data from customer interactions (e.g., call center recordings) with explicit consent for training AI models, ensuring compliance with privacy laws.

  • Public Data Initiatives: Expand open-source datasets like Taiwan Tongues to provide ethical, high-quality training data without compromising individual privacy.

  • Regulatory Alignment: Advocate for updated regulations that balance privacy with innovation, addressing disparities between traditional industries (e.g., telecom) and internet-based platforms.

Conclusion

Taiwan’s New AI Ten Major Constructions represent an ambitious vision to position the nation as a global AI leader. However, translating this vision into reality requires moving beyond slogans to actionable strategies. By prioritizing AI Sovereignty over Sovereign AI, Taiwan can leverage its cultural and linguistic strengths through initiatives like Taiwan Tongues. A hybrid approach to computing infrastructure, balancing local investments with global partnerships, ensures economic viability and wartime preparedness. Through collaborative efforts between government, industry, and civil society, Taiwan can embed its values in global AI systems, fostering innovation while safeguarding its digital sovereignty.

Call to Action

  • Government: Invest in public data infrastructure (e.g., Taiwan Common Data Set) and update regulations to support ethical AI development.

  • Industry: Participate in initiatives like Taiwan Tongues to contribute data and expertise, ensuring AI models reflect Taiwanese culture.

  • Citizens: Engage in co-creation platforms like the Taiwan Encyclopedia to enrich AI training datasets with diverse, high-quality content.

By adopting these strategies, Taiwan can transform the New AI Ten Major Constructions from a lofty slogan into a tangible, impactful reality.

コメント

コメントを投稿

このブログの人気の投稿

修仙を極めた僕が量子理論で世界を救うまでの恋愛記録

-
イメージ
第一章：「 量子場の支配者 」 【オープニングナレーション】 伝説によれば、天地が分かれる以前、混沌が支配していた太古の昔、虚空を貫く目に見えぬ力が存在した――人々はそれを「原初の場」と呼んだ。 万物が形を成す前から場は存在し、粒子が生まれる前から波は既に潜在していた。 【主人公：星波悠真（ほしなみ ゆうま）・蒼雲宗外門弟子】 「師匠、私は毎日修行していますが、丹田に何の変化もありません。私の才能が低いのでしょうか？」 【師匠：玄場師（ゲンジョウシ）】 「お前の才能は非凡だ。ただ理解していないだけだ――この世に本当の『粒子』など存在しない。あるのは場の現れだけだ。」 悠真は驚く。「場？どんな場ですか？」 玄場師は空に手を差し伸べる。「あの瞬く星が見えるか？あれは星ではなく、場の振動だ。あの囁く風が聞こえるか？あれは風ではなく、波の囁きだ。」 「粒子？ハハ、それは場が空間に顔を出して『こんにちは』と言い、また眠りに戻るだけの点に過ぎん。」 🌌【嵐の夜】 悠真は偶然、宗門の禁地に迷い込み、「場源古陣」に落ちてしまう。彼の体内のあらゆる「場」が一斉に共鳴し始めた。 ⚡ 電子場が舞い、ヒッグス場が燃え、強い相互作用場が戦い、弱い相互作用場が怠け、重力場がふざけているのを目撃する…… 心の中に声が響く： 「『量子場視界』が開かれた。万物の真の姿が見えるだろう――全ての粒子は場の夢、全ての現象は波の舞い。」 🔥【三年後、悠真再臨】 「場の主宰者」と呼ばれる。霊気や剣に頼らず、ただ一念で天地万場を揺るがせ、軽く手を振るだけで敵の「量子重ね合わせを暴走」させ自滅させる。 敵は恐怖に叫ぶ：「彼は……我々と戦っているのではなく、宇宙の根源を操っている！」 🎬【エンドクレジット】 「真の達人は粒子を作らない――場を振動させる。」 「修行の極致は昇天ではなく、共鳴である。」 第二章：「絡み合う双魂・量子真我」 【オープニングナレーション】 悠真が「量子場の主宰者」となって以来、天地の万象は全て弦のような波動として彼に映る。しかし、量子場道の本当の恐ろしさは現象ではなく――「絡み合い」にあると、まだ気づいていなかった。 🌙【場面：蒼雲宗・外天谷】 雷鳴が轟き、山壁に神秘的な符号が現れる。悠真は虚空を凝視し、独り言を呟く。「この波形...
続きを読む

凡人修真の一念永恒（原典・呪文注釈付き）

-
イメージ
第一章《量子場尊》 📜【開場旁白】 傳聞上古之時，天地尚未開闢，混沌未明，有一股無形之力橫亙於虛空——世人稱之為「本源之場」。 萬物未形，場已存在；粒子未生，波已潛動。 【主角：李粒子，青雲宗外門弟子】 「師尊，我每日苦修，丹田毫無波動，是我資質太差嗎？」 【師尊：玄場子】 「你天資不凡，只是不懂——這世上從來沒有真正的『粒子』，只有場的顯化。」 李粒子震驚：「場？什麼場？」 玄場子抬手指天：「你看那星辰閃爍，不是星，是場之振動；你聽那風聲細語，不是風，是波之低語。」 「粒子？哈哈，那只是場在空間中的某一點，跳出來打個照面，然後又退回去繼續打瞌睡罷了。」 🌌【一晚雷雨之夜】 李粒子誤入宗門禁地，跌入神秘的「場源古陣」，體內各種「場」瞬間共振。 ⚡ 他見到了電子場在飛舞、希格斯場在閃耀、強交互場在打架、弱交互場在偷懶、重力場在摸魚…… 一道聲音在他腦中響起： 「你已開啟『量子場視界』，可見萬物實相——一切粒子，皆為場之夢；一切現象，皆為波之舞。」 🔥【三年後，李粒子重出江湖】 人稱：「場尊」。他不靠靈力，不靠寶劍，只用一道念力，震動天地萬場，隨手一撥，令敵人「量子疊加失控」，自我崩潰。 敵人驚呼：「他……他不是在打我，他是在操縱宇宙本源啊！！」 🎬【結尾字幕】 「真正的強者，不創造粒子——他震動場。」 「修仙修到極致，不是飛升，而是共振。」 第二章：《糾纏雙魂．量子真我》 📜【開場旁白】 自從李粒子踏入「量子場尊」之境，天地萬象於他心中皆如弦線波動。然他尚未明白，量子場之道，真正可怕的，不是顯化，而是——糾纏。 🌙【場景：青雲宗·天外谷】 風雷交加，山壁浮現神秘符文，仿若一道古老程式碼在空間中解碼。  李粒子凝神望向虛空，自語道：「這段波動……不是自然生成，是……另一個我？」 ⚛️【忽然虛空顫動，一道與他完全一模一樣的身影憑空出現！】  李粒子：「你是誰？」 👤「我，是你在另一個場域的投影，你呼喚我，我便顯現；你震動，我便共鳴。」  原來，李粒子無意中激活了場中被稱為「量子糾纏之印」的古老咒陣，使他的意識與另一個平行場域中的「真我」互相牽連，永世共振。 📖【回憶插敘】 玄場子曾語：「當你看見第二個自己，不要怕，那不是幻象，而是——場的多重態疊加所產生的共振靈魂。」 ...
続きを読む

Exploring Quantum Computing: Principles and Applications

-
イメージ
Quantum Computing: Theoretical Foundations and Applications Since Richard Feynman proposed the idea of “simulating nature with quantum systems” in 1981, quantum computing has gradually moved from theoretical blueprints into real-world laboratories. Tech giants such as IBM and Google have since taken the lead in developing cloud-based quantum platforms and conducting early-stage commercial testing. Feynman recognized that classical digital computation cannot efficiently simulate most quantum systems. As an alternative, he introduced the concept of a quantum simulator or quantum analog—a controllable quantum system designed to emulate another quantum system that we aim to study but cannot directly manipulate. His core insight was that  “nature cannot be simulated efficiently by a classical computer, but perhaps it can be simulated by a quantum mechanical system.” Unlike traditional computers that use bits as the basic unit of information, quantum computers are...
続きを読む