MDDI 演讲稿 · 2025-11-07
杨莉明部长在 Workato AI 实验室启动仪式上的开幕致辞
Opening Remarks by Minister Josephine Teo at Workato's launch of AI Lab
要点
- • 新加坡 2025 年 AI 进展：50 多个 AI CoE（很多还有全球角色）；推出「企业算力倡议」（与 Google Cloud、微软合作）让 SME 也能获得云额度。
- • AI 治理工具栈：AI Verify（测试框架与工具包）+ Litmus / Sentinel（既能识别也能修复模型问题）+《全球 AI 保障试点》+《保护智能体 AI 系统指南》（在新加坡国际网络周边线发布）。
- • 采用率数据：企业 AI 采用从 2024 年约 50% 升至 2025 年 70%；劳动力使用率约 78%（同比 +9 个百分点）。但「在用」可能只是基础办公增效——别自我陶醉。
- • 落地三大障碍：识别真实业务用例（54%）、资金缺口（49%）、ROI 不确定（48%）。非使用者中 50% 说工具可及性差、28% 说不知道怎么用。
- • Workato 在新加坡建立「美国之外首个 AI 实验室」做定制智能体（custom agents）+ 协作智能体（collaborative agents）+ 智能体评估测试框架——这正是新加坡需要的。鼓励与 GovTech 互通——GovTech 今年引入「智能体风险与能力框架」（agentic risk and capability framework）。
完整译文（中文）
MDDI 英文原文译文 · 翻译日期：2026-05-02
Amlan、June、Ee Shan，
各位同事与朋友：
下午好——感谢邀请。我很高兴出席 Workato 在美国之外首个 AI 实验室的启动仪式。随着公司继续成长扩张——我相信你们会探索更多的领域与机会。
我们感到非常荣幸——成为这一系列「将来还会更多」的开端。年关将至，是反思的好时机。回顾新加坡 AI 发展这一年——2025 年是丰收的一年。我们《国家 AI 战略》中所识别的所有活动驱动力——政府、研究社区、产业——都取得了令人鼓舞的进展。
尤其值得一提——新加坡如今在各行各业有 50 多个 AI 卓越中心（CoE），其中不少不只服务在新加坡的团队、还承担全球角色。从我们最初只是「玩这个想法」，到今天 50 多个——我认为这是一个大跨越，也激励我们去争取更多。
作为「争取更多」的一部分，我们启动了「企业算力倡议」（Enterprise Compute Initiative）。原因很简单——我们不能假设每家公司都已经能轻松获得算力。如果你是大公司——也许这是常态；但如果你是较小的企业——即便你有非常好的想法，算力也可能成为「拦路虎」。我们不希望这样。
因此，通过与 Google Cloud、微软的合作——我们设计了一种机制——让有 AI 雄心的更多公司能拿到云额度、AI 工具，最重要的是——咨询。
怎么把这一切跑起来？我也想说——在所有这些活动向前推进时，我们要持续构建 AI 治理能力。我会再多讲一点为什么这件事对我们如此重要。
你们当中一些人可能知道我们过去为构建 AI Verify 所做的工作——这是一个测试框架与软件工具包。在它之上——我们也在看「南岸」（South Shore）。我们建了 Litmus 与 Sentinel——你要能识别 AI 模型出了什么问题，也要能修。所以 Litmus 与 Sentinel 是配对出现的。
我们也研究了诸如「AI 中使用个人数据」这样的领域，并就此发布了一些指南。今年我们启动了「全球 AI 保障试点」（Global AI Assurance Pilot），聚焦在真实场景里测试生成式 AI 应用。
更近一点——在「新加坡国际网络周」边线，我们发布了一套「保护智能体 AI 系统」（securing agentic AI systems）的指南。原因很简单——我们想跑在「智能体快速部署」的前面。所有这些努力都体现了我们打造「负责任 AI 枢纽」的承诺。
我们认为这件事重要的原因有两个：
第一——公民必须能信任「在新加坡开发与部署的 AI 是安全可用的」。
第二——全球 AI 保障标准还没有立起来。
在这些标准的制定过程中，新加坡希望有「一个座位」，并帮助塑造这些全球规范。这就是我们的目标——我们会继续聚焦在「负责任 AI」上。
在所有这些进展之中——我们也要思考还需要什么。坦率讲——我们看到 AI 生态在「企业采用」与「劳动力采用」上还有提升空间。在企业采用方面——如果你看一些报告，可能会感到一阵兴奋甚至过度自信。我们不希望那样。
微软最近发布的《AI 扩散报告》把新加坡列为顶尖国家之一。我们一直对自己说——别人那么说，但那不是我们想停留的位置。我们行业内最近的一项研究确实给了我们一种「在进步」的感觉。
比如——2024 年我们调研中约半数公司在采用，今年这一比例已上升到 7/10。我要打个补丁——他们说在「采用」时，可能只是基础的办公生产力类型的采用。我们不要拿这个数字让自己感到过度安慰。
调查里另外有用的——是公司识别出来的「他们没把 AI 推得更深」的原因。最常被引用的挑战是——「识别实际的业务用例」（54% 的受访者这样说）；其次是「缺资金」（49%）。
48% 的受访者谈到一个非常务实的问题——使用 AI 的「投资回报率」（ROI）不确定。如果 ROI 看不到——为什么要承担成本？
员工对 AI 的使用程度也在变高——今天 78% 的员工说自己在以某种方式使用 AI——比去年高出约 9 个百分点。
但若看「不使用者」——画面也很说明问题。50% 的非使用者说——AI 工具的可及性有限让他们用起来不容易；约 28% 说他们缺乏「如何使用 AI 工具」的知识。
无论是企业采用还是劳动力采用——我们都看到进展。但我们也看到必须填补的缺口。无论你看的是「半满还是半空」——都有更多事要做。所以我们正通过 IMDA 的 TechSkills Accelerator 等项目（以及行业层面的项目）积极扩大 AI 实务者人才池。
通过建立 AI 卓越中心，我们学到的另一件非常重要的事是——你的 AI 实务者（数据科学家与机器学习工程师）不能孤立工作。他们必须与领域专家、职能专家共事——那些懂生产线、懂这件事「应该怎么做」的人。我们也在鼓励非技术员工变得更「AI 流利」——把 AI 当队友，不是用来替代自己，而是帮自己更高效地工作。
我来说最后一段——Workato 与这个 AI 实验室如何融入。我看到有许多「重叠」之处——我们共享共同的兴趣。在「构建负责任企业 AI 采用」上——我们把你们视为生态非常重要的一部分。
我们知道你们在做「定制智能体」（custom agents）——这正是许多场景所需要的——不是泛化工具，而是为具体用例量身定制、契合你的语境的东西。除了缩短「价值工作」时间、提升准确性——潜在的成本节约会转化为更清晰的 ROI。ROI 正是企业想看到的。
我也知道——除了定制智能体，你们也在思考「协作智能体」（collaborative agents）——这是把边界往前推。
我也想点名——Workato 看到「加强 AI 治理」的必要——所以这个实验室计划开发新的测试框架。它们要做什么？评估智能体在实际场景中的表现。
我相信你们不只想让智能体做好事——你们一定也想让客户安心：智能体不会做坏事——「跑偏」（going rogue）的风险不存在，或者被压到最低。在这件事上——我鼓励你们与 GovTech 接触。
GovTech 也在试用智能体。今年我们引入了一套「智能体风险与能力框架」（agentic risk and capability framework）。这背后的原因是——在真正把「自主权」交给 AI 系统之前，我们要先理解风险。我们作为政府的目的——不只是把 AI 用来更好地交付公共服务，也是让我们能把经验教训分享给新加坡更广义的 AI 实务社群。
在这个意义上——我们欢迎 Workato 加入新加坡的 AI 生态。也很高兴知道你们与 6 所高等教育机构的合作——因为我们和你们一样，看到「持续投资于人才管线」的重要性。
我也希望——能有机会提升新加坡更广泛劳动力的能力。比如——如果你们把培训与对企业的部署捆绑起来，并通过 SkillsFuture 支持的项目分享洞察，机会出现时就可以做。
再次——祝贺整个 Workato 团队——期待继续合作。
非常感谢。
英文原文
MDDI 官网原始记录 · 抓取日期：2026-05-02
Amlan, June, Ee Shan,
Colleagues and friends,
Good afternoon, and thank you for having me. I am very pleased to join Workato as you launch your first AI lab outside of the US. As your company continues to grow and expand, I am sure you will continue to explore other areas and opportunities.
We feel very privileged to be the first of, perhaps, what's more to come. As we approach the end of the year, it's a good time for reflection. If we were to look back at all that has happened in AI development for Singapore, I would say that 2025 has been a good year. All the activity drivers that we had identified as part of our National AI Strategy – in government, research community and industry – have seen encouraging progress.
In particular, I just want to call out the fact that across all industries, Singapore now has over 50 AI centres of excellence. They don't necessarily only serve their teams in Singapore but also have a global mandate. From the time that we were just toying with this idea, to the fact that we now have over 50 – I think it is a big leap. It just gives us the motivation to try and get more.
As part of the idea of getting more, we did launch an Enterprise Compute Initiative. The reason is very simple – we do not assume that everyone has got good access to compute. If you are a big company, perhaps you know that is a given. But if you are a smaller enterprise, even if you had a very good idea, compute may well turn out to be a showstopper. We didn't want it to be so.
As a result, through partnerships with Google Cloud and Microsoft, we worked out the scheme that will enable more companies with AI ambitions to get cloud credits as well as AI tools and, most importantly, consultancy.
How do you get it all to work? I should also say that with all of these activities pushing ahead, we wanted to make sure that we continue to build capabilities in AI governance. I will say a little bit more about why it's so important to us.
Some of you may be aware of past efforts that we put in to build up AI Verify, this is a testing framework and software toolkit. On top of that, we are also looking at the South Shore. So, we built Litmus and Sentinel. To be able to identify what's wrong with an AI model, you need to be able to fix it too. That's why Litmus and Sentinel come as a pair.
We have also looked into areas, such as what happens when you use personal data in AI, and we have some guidelines on that too. This year, we launched a Global AI Assurance Pilot focusing on the testing of Gen AI applications in the real-world context.
More recently, on the sidelines of the Singapore International Cyber Week, we decided to launch a set of guidelines on securing authentic AI systems. The reasons are very simple. We want to try and get ahead of the rapid deployment of agents. All of these efforts really reflect our commitment to being a hub for responsible AI.
The reasons why we believe this to be important are twofold:
Firstly, citizens need to be able to trust that the AI that is being developed and deployed in Singapore is safe to use.
Secondly, is that global AI assurance standards haven't been stood up.
In the process of developing these standards, Singapore would like to have a seat at the table and help shape these global norms. So those are our objectives and we continue to focus on responsible AI.
With all of these happening, we also want to reflect on what more is needed. And I think it's fair to say that we see opportunities for the AI ecosystem to better support enterprise as well as workforce adoption, where enterprise adoption is concerned. If you read some reports, you may be feeling a sense of euphoria and over-confidence, and we don't want to do that.
If you look at Microsoft's recent report on AI diffusion, they had put Singapore as among the top countries. We always tell ourselves – that's what they say, but it's not where we want to be. Our own recent study within the industry does give us a sense that there is progress.
For example, adoption seems to have expanded in 2024 – it was about half of the companies that we surveyed. This year, it has gone up to seven in 10. I want to qualify by saying that, when they say they're adopting, it may be quite basic workplace productivity types of adoption. We don't want to kid ourselves into thinking that the number should fuel you or make you feel very comforted.
The other thing that came out through the survey that is useful, are the challenges that companies identified as reasons why they have not gone further with AI adoption. Amongst the most commonly-cited challenge is identifying practical business use cases. 54% of the respondents said so, and 49% cited the lack of funds.
48% of respondents talked about a very practical issue – uncertainties, in terms of the return of investment for the use of AI. So, if there is no ROI, why incur the cost?
Workers also are using AI to a larger extent. Today, 78% of them say they use AI in one way or another. This would be about a 9% increase from the last year.
However, if we look at the non-users, it's also quite telling. They say that a limited access to AI tools has not made it easy for them. 50% of them said so. About 28% of respondents say that they lack the knowledge on how to use AI tools.
Whether it is in terms of enterprise adoption or workforce adoption, we see progress. However, we also see gaps that ought to be plugged. Whether we look at the glass as half full or half empty, there is more to be done. Thus, we are actively growing the pool of AI practitioners through programs like IMDA’s Tech Skills Accelerator programme, and others at the sectoral level.
Another very important thing that we have learned through setting up the AI Centres of Excellence, is that your AI practitioners – who are your data scientist and machine learning engineers – cannot work in isolation. They need to work with domain and functional experts – people who know the production line and how the function is supposed to be carried out. We are also encouraging non-tech workers to be more AI-fluent and to use AI as a teammate – not to replace themselves, but to help themselves work even more effectively.
Let me just come to the last part of my comments, which is how Workato and the AI Lab fits in. I see many areas of overlap, where we share a common interest. We see you as a very important part of our ecosystem, in building up responsible enterprise AI adoption.
We know that you're working on custom agents, and we know that that is precisely what is needed in many instances to be able to address specific use cases, not generalised tools. You need to have things that are bespoke that works to your context. Apart from reducing time for value work and improving accuracy, the potential cost savings would translate into clearer returns on investment. The ROI is exactly what businesses want to see.
I know that apart from custom agents, you're also thinking of collaborative agents, and this is pushing the boundaries.
I would also just like to call out that Workato sees the need to strengthen AI governance too, and so the lab plans to develop new test frameworks. What will they do? They will evaluate the agent performance in practical scenarios.
I believe that you are not only looking for agents being able to do good work. I am quite sure that you also want to assure your clients that the agents aren't going to be up to no good – that the risk of them going rogue is not going to be present, or maybe the risk is reduced to the minimum. In this work, I would encourage you to engage with GovTech.
GovTech is also experimenting with the use of agents. This year, we introduced an agentic risk and capability framework. The reason behind this was because we want to understand the risk before actually delegating autonomy to AI systems. Our purpose as a government is not just to use it to deliver public services better, but also to be able to share lessons with Singapore's broader community of AI practitioners.
On that note, I would like to say that we welcome the addition of Workato to the AI ecosystem in Singapore. We are very happy to know of the partnership that you have with the six Institutes of Higher Learning, because we, like you, recognise the importance of continuing to invest in growing the talent pipeline.
I hope that there will also be opportunities to upskill the broader Singapore workforce. This could be done, for example, if you bundle training with deployment to enterprises, and offer insights through the SkillsFuture supported programs where the opportunity presents itself.
Once again, congratulations to the whole Workato team, and we look forward to continuing to work with you.
Thank you so much.