MDDI 演講稿 · 2025-08-28
楊莉明部長在 Google Cloud AI 亞洲活動上的開幕致辭
要點
- • Google 與新加坡的 AI 合作鏈：早期的「AI Trailblazers」+ 今年的「企業算力倡議」（ECI）+ Google Cloud 的「AI Cloud Takeoff」。
- • 使用者案例：Mirxes（早期癌症檢測試劑初創，估值已超 10 億美元）借力 MedGemma 等 Google 開源醫療模型；Gill Capital（H&M、迪卡儂零售運營商）將 AI 智慧體嵌入電商平臺，每週節省 200+ 小時——計劃推廣到海外市場。
- • 智慧體（Agentic AI）從「簡單助手」到「能理解高層指令、自主拆解、試錯與持續學習」——能力越強，治理問題越尖銳：許可權、人在環、問責。
- • GovTech × Google Cloud 沙盒：政府將早期接入 Project Mariner Computer Use API（推理、規劃、任務管理），藉此測試與評估、把經驗分享給新加坡 AI 實務社群。
- • 現狀：約 1/3（約 5 萬名）公職人員常態化使用「Pair」（政府 AI 聊天機器人）；約 18,000 個機器人由公職人員自建。
完整譯文（繁體中文）
MDDI 英文原文譯文 · 翻譯日期: 2026-05-03
Google Cloud 執行長 Thomas Kurian 先生，
各位同事與朋友：
早安。
感謝邀請我出席「Google Cloud AI 亞洲」活動——很高興在這裡。
在最近的「國慶群眾大會」上——黃循財總理談到了 AI 的重要性、以及它對新加坡意味著什麼。一如以往的技術與創新浪潮——從計算機到網際網路——我們的方法始終是「支援公民把握機會、應對挑戰」。
這也是 2023 年 12 月推出的《國家 AI 戰略 2.0》（NAIS 2.0）的精神。自那之後——我們的 AI 生態變得更強、更有活力。
我們的夥伴做出了實實在在的貢獻。Google 是其中之一。通過「AI Trailblazers」（AI 先鋒）倡議——新加坡早期採用生成式 AI 的企業獲得了 Google Cloud 的算力、開發工具與工程師支援，構建原型應用。這在參與組織內部催生了良好勢能——也讓他們的同行也坐起身子來。
今年早些時候——我們推出了「企業算力倡議」（ECI），鼓勵企業藉助雲服務商已開發並嵌入的 AI 工具——啟動或加速 AI 採用。Google Cloud 現已加入 ECI 合作伙伴行列——其「AI Cloud Takeoff」計劃旨在幫許多新加坡企業構建 AI 能力、開發 AI 加持的產品與服務。
已經有幾家公司從這個專案中受益。
一個例子是 Mirxes——新加坡的生物技術初創，開發用於癌症與其他疾病早期檢測的試劑盒。Mirxes 估值已超過 10 億美元——但他們沒有止步。為通過研發提升競爭力——他們藉助 Google 的開源醫療模型，比如 MedGemma。
另一個例子是 Gill Capital——全球零售生活品牌（如 H&M、迪卡儂）的運營商與分銷商。該專案幫助 Gill Capital 構建嵌入新加坡電商平臺的 AI 智慧體——這些智慧體也幫本地零售店經理改進商品分類、推薦補貨時機、基於客戶資料制定銷售策略。Gill Capital 因此每週節省超過 200 小時的生產力。公司也計劃把這些在新加坡開發的智慧體推廣到海外市場。
我們將繼續以「合作」的方式推動 AI 採用。從我分享的案例可見——尤其在新領域探索時，這種合作格外有用。今年備受關注的一個領域，是「智慧體 AI」（Agentic AI）。
智慧體 AI 在「人與 AI 的互動方式」上開啟新的層級。
智慧體已經能扮演簡單的助手——按我們的指令分析資料、起草郵件、瀏覽網站搜資料。
更精細的智慧體可以在複雜工作流裡執行一系列指令——把任務串聯起來，並基於前序結果決定下一步。
再之後——可能出現這樣的智慧體：它能理解我們的高層指令，自主地把它們拆解成更小的任務，嘗試不同的方法，並基於既往成果持續學習與適應。
隨著精細度上升——生產力提升與價值放大的潛能也在增長。
我們可以設想——把 AI 智慧體引入公共服務的改進。
比如——企業有時會被「為獲得許可證要打交道的多家機構」繞暈。他們可能要打很多電話、查很多網站，才搞清楚正確的辦事順序。需要檢查時——還要安排預約。這些工作流——很適合讓一個智慧助手來幫忙。
另一個潛在應用是社會服務。求助者面對一張為他們提供支援的「機構與專案網路」時常常迷失。一個智慧助手可以幫社工為客戶提供基礎指引——比如滿足行政要求——讓社工騰出時間聚焦更有影響力的活動，如諮詢與發展合作。
和所有新興技術一樣——政府認為我們必須先理解它如何運作、錯誤為何發生。對 AI 智慧體——「非預期行為」是合理的關切——我們要在治理上更加用心。智慧體應被賦予什麼許可權？什麼時候人要在環（in the loop）？如果事情沒按預期發生，誰擔責？
這些問題值得認真思考。MDDI 將積極發揮作用——確保智慧體能力以安全、負責任的方式被開發與部署——好讓公職人員與公民能自信地與這些工具共事。
我們已有不錯的基礎——可以支撐 AI 的有意義採用。在我們 15 萬名公職人員中，約 1/3 現在常態化使用「Pair」——政府的 AI 聊天機器人——以增強生產力、寫作、研究。公職人員也在學習自己搭機器人——目前已有約 18,000 個機器人。這些機器人幫助處理多種工作流——比如有效回應公眾諮詢、研究資料分析、政策評審。
[公告] 但我們可以更進一步。GovTech 將與 Google Cloud 共建一個沙盒——為公共部門用例實驗智慧體 AI，並測試相應方案。這將通過對資料保護的增強與「Gemini 通過 Google Distributed Cloud 提供」的方式實現——稍後 Thomas 會再多講。在這個沙盒裡，Google 將讓 MDDI 旗下機構早期接入 Google 的「Project Mariner Computer Use API」——一款能推理、規劃、管理任務的智慧體 AI 工具。機構將有機會測試與評估最新的智慧體能力、評估風險、開發緩釋措施，並把所學分享給新加坡更廣義的 AI 實務社群。我們希望——通過這個沙盒——更好地理解如何與智慧體 AI 互動，並建立信心、為公共利益捕獲其價值。
我們今天宣佈的沙盒——是新加坡「與 AI 一同學習與成長」努力的一部分。我們將繼續尋找夥伴，去拓展 AI 採用的邊界。
我們感謝 Google 在新加坡建立的活動廣度與深度——並期待支援你們的區域與全球業務擴張。
祝大家今天的活動圓滿成功。
英文原文
MDDI 官網原始記錄 · 抓取日期: 2026-05-02
Mr Thomas Kurian, CEO, Google Cloud,
Colleagues and Friends,
Good morning.
Thank you for inviting me to the Google Cloud AI Asia event. I am delighted to be here.
At the recent National Day Rally, Prime Minister Lawrence Wong spoke about the importance of AI and what it means for Singapore. As with past waves of technology and innovation – from the computer to the internet – our approach has always been to support our citizens to make the most of the opportunities and respond appropriately to the challenges.
This was also the spirit that guided the National AI Strategy 2.0 (NAIS 2.0) which we launched in December 2023. Since then, our AI ecosystem has grown stronger and more vibrant.
Our partners have made a definite contribution. Google is one of them. Through the “AI Trailblazers” initiative, early adopters of GenAI in Singapore were given access to Google Cloud’s compute resources, development tools, and engineers to build prototype applications. That created good momentum within the participating organisations. It got their peers to sit up too.
Earlier this year, we introduced the Enterprise Compute Initiative (ECI). This was to encourage companies to kickstart or accelerate their AI adoption, through using AI tools already developed and embedded in the offerings of cloud service providers. Google Cloud is now onboard as one of the ECI partners. Its “AI Cloud Takeoff” programme aims to help many Singapore-based companies build AI capabilities and develop AI-enabled products and services.
Several companies are already benefitting from this programme.
One example is Mirxes. It is a biotech startup in Singapore that develops test kits for early detection of cancer and other diseases. Already valued at over a billion dollars, Mirxes is not staying still. To sharpen its competitive edge through R&D, it turned to Google’s open-source healthcare models such as MedGemma.
Another example is Gill Capital. It is an operator and distributor of global retail lifestyle brands, such as H&M and Decathlon. The programme helped Gill Capital build AI agents that are embedded in their e-commerce platforms offered in Singapore. The agents can also help the retail store managers here improve product classifications, make recommendations on when to replenish stock, and develop sales strategies based on customer data. This has helped Gill Capital to unlock productivity savings of more than 200 hours per week. The company expects to expand the use of these agents developed in Singapore to their overseas markets too.
We will continue to take a partnership approach in promoting AI adoption. As seen from the examples I’ve shared, this is especially useful when exploring newer areas. One area attracting much attention this year is Agentic AI .
Agentic AI opens up new layers of possibilities in the way humans interact with AI.
Agents can already act as simple assistants – following up on our instructions to analyse data, draft emails, or navigate websites to search for information.
More sophisticated agents can implement a series of instructions within complex workflows. They can combine tasks and determine the next steps based on the results of the earlier tasks.
And then there may be agents that understand our high-level instructions, autonomously break them down into smaller tasks, experiment with different approaches, and continuously learn and adapt from earlier efforts.
With increased sophistication, the potential for productivity improvements and value enhancements also grows .
We can imagine introducing AI agents to help improve public services.
For example, businesses are sometimes confused by the multiple agencies they must deal with to get licenses approved. They may call each department or check many websites to figure out the right order of proceedings. When inspections are required, there is the added challenge of making appointments. Workflows like these could do with the help of an intelligent assistant.
Another potential application could be in social services. Help-seekers can feel lost navigating the network of organisations and programmes designed to support them. An intelligent assistant could help social workers provide basic guidance to their clients, such as in meeting administrative requirements. This could free up social workers’ time to focus on more impactful activities like counselling and developing partnerships.
As with all emerging technologies, the Government believes we need first to understand how they work and why mistakes happen . With AI agents, there are valid concerns about unintended actions, and we need to pay even more attention to governance. What permissions should agents be given? When should humans be in the loop? If things do not go as expected, who should be held accountable?
These questions deserve careful consideration. MDDI will play an active role in ensuring that agentic capabilities are developed and deployed in a safe and responsible way , so that our public officers and citizens can work with these tools confidently.
We are building on a good foundation that already supports meaningful adoption of AI. About one-third of our 150,000 public officers now regularly use Pair, the Government’s AI chatbot, to enhance their productivity, writing and research. Public officers are also learning how to build their own AI bots, of which there are now about 18,000 bots. These bots help with a variety of work processes such as responding effectively to public queries, research data analysis, and policy reviews,
[Announcement] But we can go further. GovTech will be working with Google Cloud on a sandbox to experiment with Agentic AI and test agentic solutions for public sector use cases. This will be made possible by enhancements to data protection and how Gemini will be offered through Google Distributed Cloud, which Thomas will say more about later. Through this sandbox, Google will give MDDI agencies early access to Google’s Project Mariner Computer Use API, an Agentic AI tool that can reason, plan and manage tasks. Agencies will have a chance to test and evaluate the latest agentic capabilities, assess the risks, develop mitigation measures, and share the lessons learned with the broader community of AI practitioners in Singapore. From the sandbox, we hope to better understand how to interact with agentic AI and build confidence to capture its value for the public good .
The sandbox we are announcing today is part of Singapore’s efforts to learn and grow with AI. We will continue to seek out partners to push the boundaries of AI adoption.
We thank Google for the breadth and depth of activities you have built up in Singapore and look forward to supporting your business expansion regionally and globally.
I wish you a most successful event today.