MDDI 演讲稿 · 2025-10-02
杨莉明部长在新加坡理工大学与英伟达 AI 中心开幕式上的演讲
Speech by Minister Josephine Teo at the Opening of SITxNVIDIA AI Centre
要点
- • 我们追求的不是「为 AI 而 AI」——是 AI 如何改善生活。NAIS2.0 启动 18 个月以来已建立近 50 个 AI CoE。
- • Razer 的 AI QA Companion 把游戏 QA 时间砍半；Grab 的 AI 不只为内部生产力，还助力商家做生意。
- • MIT 研究：放射科医生使用 CheXpert AI 工具时，诊断准确率反而下降——AI 工具单独使用时性能优于 2/3 的医生，但与人协作时设计不佳就会拖累专家判断。
- • 「不要把第一直觉放在『让 AI 强到取代人』，而要放在『让 AI 强到能增强每个人』」。AI 应作为「队友」（teammate），不是替代品。
- • 三条人才路径：①扩展 AI 实务者（SNAIC AI Programme 用 NVIDIA DLI 课程，3 年培训 200+）；②培育「AI 双语者」（domain expert 学 AI 这门「第二语言」）——SIT 启动应用 AI 博士培训中心；③产业 + 研究 + 政府生态合作。SNAIC 18 个月已与 70 家公司交付 50 个方案。
完整译文（中文）
MDDI 英文原文译文 · 翻译日期：2026-05-02
新加坡理工大学（SIT）董事会主席 Bill Chang 先生，
SIT 校长 Chua Kee Chaing 教授，
SIT-NVIDIA AI 中心联合主任 Ng Aik Beng 博士与 Daniel Wang 博士，
产业伙伴与各位同事：
早安。我很高兴回到 SIT——出席 SNAIC 的正式启用仪式。
你们听 AI 听得也许已经发腻了。
「一切都关乎 AI 吗？」
「除了 AI 就没别的了吗？」
我先说一句——我们不是「为 AI 而 AI」。
我们更感兴趣的是——这项技术如何能改善生活——无论是为个人，还是通过组织。
如果只谈「AI 多伟大」，但它不在真实世界里带来收益——那就没有用。
你们应该听过黄循财总理在最近的「国庆群众大会」（NDR）上分享的——AI 如何在新加坡被部署来支持公民、把握机会的几个例子。
在我们启动更新版《国家 AI 战略》（NAIS2.0）后的 18 个月里，已经建立了近 50 个 AI 卓越中心（CoE），还有更多例子。
我一直在走访这些 AI CoE——理解 AI 在组织里的影响——令我印象深刻的是它如何被用来提升生产力、为我们的人扩展机会。
我分享几个最近的例子。
雷蛇（Razer）的 AI QA Companion 几乎把新游戏 QA 测试与缺陷检测的耗时砍半。其 AI 套件中的其他工具，也将赋能游戏开发者及更多人——设计个性化的游戏体验。
Grab 把 AI 用于代码开发——并且超出内部生产力，帮助平台上的商家做生意——通过智能助手提供商业洞察与运营支持。
AI 显然有一个「竞争维度」——公司想用 AI 取得优势，国家也一样。
两者都会驱动 AI 的采用——但 AI 也应有「协作维度」。
比如 AI 治理——问题众多，没有人拥有全部答案。
我刚从纽约的联合国大会（UNGA）回来——「全球 AI 治理对话」（Global Dialogue on AI Governance）正是在那里启动的。
新加坡持续在这种对话中做出一致而坚定的贡献。
今年早些时候——新加坡承办了「国际表征学习大会」（ICLR）——这是 ICLR 在美国之外举办过的最大一届——吸引了 11,000 名代表。
在 ICLR 边线——我们也举办了「新加坡 AI 大会—国际科学交流」（SCAI-ISE）——主题是「安全与负责任 AI 的研究优先事项」。它促成了「新加坡共识」（The Singapore Consensus）——至今仍被关心 AI 治理的人时常提起。
明年 1 月——「人工智能促进协会」（AAAI）也将在新加坡举办年会。
所有这些合作都旨在——找出更好的方式，最大化 AI 部署的上行、最小化下行。
另一个重要的协作领域——是发展我们的人。
聚焦发展技术容易；但聚焦于「构建并使用 AI 的人」往往不够。
认真考虑「AI 与人类的关系」、以及我们希望两者如何共事，对我们而言至关重要。
我相信我们的第一直觉不应该是「如何让 AI 好到取代人」——而应该是「如何让 AI 好到增强每一个人」。两者只有一线之差。
MIT 的研究者发现了一件支持我这个信念的事情——
当放射科医生被允许使用一款叫 CheXpert 的 AI 诊断工具时——他们的诊断准确率反而下降了——尽管这款 AI 工具单独工作时表现优于 2/3 的放射科医生。
为什么这款表现更好的 AI 工具——在被专业人士使用时反而拉低了结果？
问题不在于 AI 的技术能力。
问题很可能是——这款工具不是为「与人类专家协作」而设计的。
它对自己的推理几乎不透明——以致放射科医生即便对 CheXpert 的诊断有疑虑，也搁置了自己更好的判断。
更根本的是——它把专家意见简化为「图像扫描」——没有纳入医学史、医患对话、与其他专家的交流等其他必要考量。
换句话说——我们在思考如何使用 AI 时必须谨慎。
AI 在很多任务上确实能高效自动化——价值损失不大或为零。这种情况下，自动化是合理的。
但这并非铁律。某些情形下——AI 不够好，为了「最后一点点性能提升以替代人」而追求，是没意义的。
在那些情形下——我们应当自律地把 AI 当作「与人协作」的工具——让两者作为一个整体共事。
这就是我最近谈到的「AI 作为队友」（AI as a teammate）的理念——AI 工具增强人的表现、AI 助手执行与人互补的任务。
我们怎样才能做到这一点？
扩大 AI 实务人才管线
第一——继续扩大 AI 实务者池子。这些人具备数据科学、机器学习等深厚技术能力。
在生成式 AI 时代——他们也将具备构建或微调大模型（LLMs）、利用「检索增强生成」（RAG）来编排 AI 智能体团队的能力。
更多 AI 实务者——意味着我们部署「AI 作为队友」的容量更大。
今天——我很高兴宣布——一项新的「SNAIC AI 计划」将直接为此目标做贡献。
在 IMDA「TechSkills Accelerator」（TeSA）的支持下——这一计划将在未来 3 年内培训 200 多名应届生与中职转型者——围绕先进 AI 系统与 AI 应用构建。
学员将以两个月的密集升级开始——基于 NVIDIA Deep Learning Institute 的课程。
随后是 4 个月由导师指导的产业实操项目——结业后能为产业做出有意义的贡献。
这将进一步壮大我们正通过其他既有项目培育的 AI 实务者池。
培育跨学科的「AI 双语者」
新加坡的「AI 梦之队」不会只有 AI 专家——还需要我所谓的「AI 双语者」（AI bilingualists）。
他们是放射科医生、会计师、技师、律师、创作者等领域专家——已经精通自身领域，也就是他们的「母语」。
他们能为团队带来宝贵知识——帮助团队把 AI 用得好——他们能提供数据科学家与机器学习工程师还不具备的语境与洞察。
为了让这支队伍合作良好——这些领域专家最好学一门「共同语言」——也就是 AI 这门「语言」。
新加坡人凭经验都知道——学两门语言并不容易——但并非不可能。
「一起学、彼此练」更有趣。许多人能用「会话级」就过得去——有些人需要能读、能写。
再小一群人，会成为两门语言的「大师」。
我们的 AI 生态需要不同精熟度的「AI 双语者」——我们也在系统性地寻找发展他们的机会。
所以我很兴奋——SIT 率先启动「应用 AI 博士培训中心」（Doctoral Training Centre for Applied AI），培育最高精熟度的「AI 双语者」。
这是计划中一系列博士培训中心的第一个——为已是领域或职能专家的新加坡人，提供同时成为 AI 应用专家的机会。
后续中心将聚焦在新加坡的优先行业——如海事与医疗——培育更多能为创新做贡献的「AI 双语者」。
围绕 AI 部署构建生态级合作
最后——让我谈谈另一种同样重要的合作——产业、研究社区与政府之间的合作。
SNAIC 是个好例子。
通过把 SIT 在应用 AI 与「转化研究」上的强项，与 NVIDIA 的前沿技术与专业能力结合——这个中心帮助大小企业理解、测试、开发并落地 AI 方案——覆盖广泛的行业。
在中心成立后的 1.5 年里——SNAIC 已与 70 家公司合作——在制造、医疗、金融、交通行业交付了 50 个产生真实业务影响的 AI 方案。
比如——这个中心与陈笃生医院合作开发协同 AI 工具，帮助临床医生交付更快、更准确的护理。
AI 工具帮助缩短了病人等待时间——并在每位病人康复的旅程中提供更好支持。
展望未来
在今天 SIT-NVIDIA AI 中心正式启用之际——我感谢我们的产业伙伴、研究者与教员对新加坡 AI 发展的承诺与贡献。
我也感到鼓舞——来自世界各地的伙伴愿与新加坡一起在 AI 上创新——欢迎各位继续与我们的 AI 生态合作。
携手同行——我相信我们能实现「AI 服务公共利益——为新加坡，也为世界」的愿景。
再次感谢。
英文原文
MDDI 官网原始记录 · 抓取日期：2026-05-02
Singapore Institute of Technology (SIT) Board of Trustees Chairman, Mr Bill Chang
SIT President, Professor Chua Kee Chaing,
SIT-NVIDIA AI Centre Co-Directors, Dr. Ng Aik Beng and Dr. Daniel Wang
Industry partners and Colleagues,
Good morning. I’m happy to be back at SIT to join you for the formal launch of SNAIC.
You may be sick of hearing about AI by now.
Is everything about AI?
Is there nothing else besides AI?
Let me start by saying we aren’t pursuing AI for AI’s sake.
We are far more interested in how this technology can be used to improve lives, whether for individuals or through organisations.
There is no use if we talk about how great AI is, if it isn’t bringing benefits in the real world.
You would have heard Prime Minister Wong share at the recent National Day Rally (NDR), some examples of how AI is being deployed in Singapore to best support our citizens and make the most of opportunities.
There are many more examples supported by the almost 50 AI Centres of Excellence (CoEs) that have been set up in the last 18 months since we launched the refreshed National AI Strategy (NAIS2.0).
I have been visiting these AI CoEs to understand the impact of AI within organisations, and I’m impressed by how it is being used to raise productivity and expand opportunities for our people.
Let me share some recent examples.
Razer’s AI-driven Quality Assurance (QA) Companion almost halves the time required for new games’ QA testing and bug detection. Other tools in its AI suite will also empower game developers and many others to design personalised gaming experiences.
Grab is using AI for code development. It is also going beyond internal productivity to help its platform merchants do business better, with intelligent assistants that provide business insights and operational support.
Clearly, there is a competitive dimension to AI. Companies want to use AI to give themselves an edge, and countries want the same too.
Both will drive the adoption of AI, but there should also be a collaborative dimension to AI.
Take AI governance for instance, where there are a lot of questions and no one has all the answers.
I just returned from the United Nations General Assembly (UNGA) in New York where the Global Dialogue on AI Governance was just launched.
And Singapore continues to be a consistent and steadfast in contributing to conversations like these.
Earlier this year, Singapore hosted the International Conference on Learning Representations (ICLR). It was the biggest that they have ever held outside of the United States and drew 11,000 delegates.
On the sidelines of ICLR, we also hosted the Singapore Conference on AI - International Scientific Exchange (SCAI-ISE) on Research Priorities for Safe and Responsible AI. This led to the formation of “The Singapore Consensus” on what the priorities should be. It is still talked about when I meet members of the community who care a lot about AI governance.
Next January, Association for the Advancement of AI (AAAI) will also have its meeting in Singapore.
All these collaborations seek to promote better ways to maximise the upsides and minimise the downsides of AI deployment.
Another important area for collaboration is in developing our people.
It is easy to focus on developing the technology, and not enough on the people who build and use AI.
It is important for us to carefully consider the relationship between AI and humans and how we want the two to work together.
I believe our first instinct should not be “how to make the AI so good it replaces humans”. Instead, it should be: “how can we make AI so good, that it enhances all humans?” There is just a fine distinction between the two.
Researchers at MIT found something that supports my belief.
When radiologists were allowed to use an AI diagnostic tool called CheXpert, the accuracy of their diagnoses actually declined, even though the AI tool alone performed better in diagnoses than two-thirds of radiologists.
Why did this superior AI tool produce inferior results when used by professionals?
The problem wasn’t with the AI’s technical ability.
The problem was likely because the tool was not designed for collaboration with human experts.
It offered little transparency about its reasoning, leading radiologists to suspend their better judgement even when they had suspicions about CheXpert’s diagnosis.
More fundamentally, it reduced expert opinion to image scanning, failing to incorporate other essential considerations such as medical histories, doctor-patient conversations, and exchanges with other experts.
In other words, we should be careful when we think about how to use AI.
AI will sometimes be very effective in automating tasks performed by humans, with little or no loss of value. Automation makes sense in such cases.
But this is not a given. In some instances, the AI is not good enough and pursuing the last bit of performance improvement just to replace a human makes no sense.
In such cases, we should instead discipline ourselves to think of using AI to collaborate with humans and let the two work together as one.
This is the idea of AI as a teammate which I spoke about recently. Where AI tools enhance human performance, and AI assistants perform tasks that complement humans.
How we can we go about this?
Developing our pipeline of AI practitioners
First , we should continue to grow the pool of AI practitioners. These are people steeped in technical skills like data science and machine learning.
In the era of Generative AI, they will also have the skills to build or fine-tuning large language models (LLMs), or Retrieval-Augmented Generation (RAG) to orchestrate teams of AI Agents.
Having more AI practitioners will expand our capacity for deploying AI as teammates.
Today, I am pleased to announce a new SNAIC AI Programme that will contribute directly to this goal.
Supported by the Infocomm Media Development Authority’s (IMDA) TechSkills Accelerator (TeSA) initiative, the programme will train more than 200 fresh graduates and mid-career professionals over the next three years in advanced AI systems and building AI applications.
Trainees will begin with two months of intensive upskilling based on the NVIDIA Deep Learning Institute’s curriculum.
They will then undertake four months of guided hands-on industry projects, and emerge from the programme ready to contribute meaningfully to industry.
This will add to our growing pool of AI practitioners that we are nurturing from other existing programmes.
Nurturing interdisciplinary “AI Bilingualists”
The AI “dream team” in Singapore will not just have practitioners who are experts in AI. There will also need to be what I have called “AI bilingualists”.
These are domain experts like radiologists, accountants, technicians, lawyers, and creators, who are already knowledgeable in their fields – or their “mother tongues”.
They can thus bring valuable knowledge to help the team make good use of AI. They are able to provide the context and provide insights that the data scientists and machine-learning engineers do not yet possess.
But for this team to work well together, it helps that these domain experts to learn a common language – that is the “language” of AI.
Singaporeans know from experience that learning two languages is not easy, but it is not impossible.
It is more fun to learn together and practise with one another. Many of us will get on well enough with conversational level ability, but some of us will need to read or write.
And a smaller group may become masters of both languages.
Our AI ecosystem will need “AI bilingualists” at different levels of mastery, and we are systematically identifying opportunities to develop them.
So I am excited that SIT is taking the lead to nurture “AI bilingualists” at the highest level of mastery, with a new Doctoral Training Centre for Applied AI.
This is the first of several planned Doctoral Training Centres that will provide Singaporeans who are already domain or function experts, the chance to be experts in applying AI as well.
Subsequent centres will focus on developing more “AI bilingualists” who can contribute to AI innovation in Singapore’s priority sectors, such as maritime and healthcare.
Building ecosystem-wide partnerships for AI deployment
Finally, let me say something about a different but equally important kind of collaboration – between industry, research community, and government.
The SNAIC is a good example of this.
By combining SIT’s strength in applied AI and translational research with NVIDIA’s cutting-edge technology and technical expertise, the centre helps businesses of all sizes understand, testbed, develop, and implement AI solutions across a wide range of sectors.
In the one and a half years since it was established, the SNAIC has worked with 70 companies to deliver 50 AI solutions that create real business impact across the manufacturing, healthcare, finance, and transport sectors.
For instance, the Centre has worked with Tan Tock Seng Hospital to develop collaborative AI tools that support clinicians in delivering faster and accurate care.
AI tools have helped reduce patient waiting times and provide better support for each patient in the journey of recovery.
Looking forward
As we mark the official launch of the SIT-NVIDIA AI Centre today, I want to thank our industry partners, researchers, and faculty members for their commitment and contributions to Singapore’s AI development.
I am encouraged by the partners from around the world who want to work with Singapore on innovating in AI, and invite you to continue collaborating with our AI ecosystem.
Together, I believe that we can realise the vision of AI for the Public Good, for Singapore and the World.
Thank you once again.