張莉媚部長在亞洲經濟峰會小組討論上的發言

Panel Title: “From Infrastructure to Intelligence: Turning Southeast Asia’s Digital Foundations into AI Scale”

Terence Lee (TIA): To start off, Minister Teo, you were talking about AI sovereignty, and how the word is often not well defined and needs to be reframed. You mentioned a few examples of how that could be done. Could you elaborate on how Singapore, or how any country, should go about reframing AI sovereignty and implementing this?

Minister Josephine Teo: Thank you very much, Terence. As I was listening to Minister Meutya’s presentation, an excellent presentation by the way, I thought that there were actually also points of alignment. Sovereignty is not something you disregard. I should say that at the outset. It is an important question.

There is a need for us to be able to use AI on our own terms in service of our people. Whether that is achieved solely through the building and ownership of everything along the AI stack is a question that I think we need to address and ask. Because very few countries will be able to build, own and maintain the entire AI stack. There are just too many different components involved.

So the way we choose to look at it is that there are three things you need to watch. First is the ability to use AI for the public good and for governance. So earlier when I discussed what capabilities are needed, I focused on organisational capabilities and workforce capabilities. Actually, the public sector's own use of AI is important. If you don't have the knowledge or the people that can use AI to deliver public services, there is a big gap. So that's something you have to build up.

The other aspect of it is that because you use it yourself, you are better informed as to how to govern and what rules to put in place; and if you have to rely on what other people tell you is needed in implementing AI, then your ability to govern AI can be questioned.

But the part that also needs a lot of effort to build up is the autonomy to make smart choices. I'll expand on what we mean by smart choices, whether it is the government, or whether it is a company. When we think about the choice of technology to be incorporated, there are a number of key questions. First, does it perform well? Second, is it affordable? Third, we must also consider whether the security and resilience aspects are well taken care of. Put the three together, and you want to make choices on the basis of how the technology product or supplier best meets your requirements. It should matter more how the overall value is achieved than whether this comes from country A or country B.

The choice is not so much between country A and country B. The choice is really whether, on balance, you are better able to achieve the combination of performance and affordability, security and resilience. So I think that’s what we want to be able to get to, to be able to say that I am not choosing on the basis of where it comes from, but on the basis of what it can do for us. That's what I mean by smart choices.

This flows quite consistently with how we want to build anchors. You need a healthy ecosystem, with many different suppliers and providers of technology solutions, and you don't want to be overly tied to or overly dependent on any one of them. This is aligned with what Minister Meutya was saying, that you can achieve this ability to set your own terms, if you achieve these two other goals. That’s how I think about it.

Terence Lee (TIA): Minister Teo, a follow up on your earlier point about buying AI technology and making smart choices around security and resilience. Could you elaborate on that and how that decision framework comes into play?

Minister Josephine Teo: A company that is digitalised has multiple systems. It doesn’t just have a single system. I’m sure in Indonesia, you have systems that keep track of inventory, stationary inventory, crown jewels, and even systems that track how much money you have. These systems are not the same, and you need different degrees of protection.

The security and resilience of a system depends on your requirements. If your requirements are very high, then, when you evaluate the vendor, you will ask questions on how they manage it and what rules are built in to protect your data from being accessed by people who have no business accessing it.

The way I see it is that companies need to be aware of which of their systems and datasets deserve a higher degree of security. They also need to know, if they are using a particular provider’s service and if that service is somehow no longer available to them, what options they have. That goes to the question of resilience, whether, if you have a disruption, you have the ability to switch quickly or to recover from the disruption?

I think these are important questions that deserve your attention, rather than just assuming that you will achieve security and resilience just because a vendor or provider claims to be from a particular country, or that purchasing from or partnering with an organisation from another country will compromise your security and resilience.

I think making smart choices means looking beyond just the façade but knowing how their services and products are built, from the perspective of those on the receiving end.

Terence Lee (TIA): A food-for-thought for AI vendors in the room and AI inventors.

Terence Lee (TIA): I want to touch on data flows. I think that was an interesting topic that was brought up in both speeches. Minister Meutya, you talked about facilitating data flows across countries, and Minister Teo, you mentioned the DEFA. So, how does that work? How do you see rolling out, and what can companies expect?

Minister Josephine Teo: If I could just expand on what Minister Meutya said, with an analogy from the civil aviation domain.

We’ve all built our own airports, but these airports are not very useful unless the airplanes can fly from one city to another. Whether you can fly from one city to another is a matter of air services agreements between countries, and DEFA should be thought of as the equivalent of an air services agreement.

Even if you have the air services agreement, consider this scenario: the airplane comes into Singapore, you say: “Sorry, you are using a different door to your aircraft, it doesn't fit into my aerobridge”. Or you say: “Your landing system uses a format that is different from what my air traffic control uses, so you can't fly.” Then what is the point of having the air services agreement?

DEFA is this top line agreement. It says that we agree on how there should be interactions between our digital economies. But the actual mechanisms, the agreements on the mechanisms for data flows, still have to be worked out.

Fortunately, long before DEFA was even thought about and discussed, within ASEAN, we already had a data management framework. We agreed on contractual clauses between data-exchanging entities in different jurisdictions because we must accept that our data laws are not identical. But we have a way of saying that even if they are not identical, they can connect and interoperate with each other. The next steps are to really flesh out all these detailed workings, so that when the equivalent of an aircraft flies from one airport to another, it doesn’t encounter the problem of incompatible systems. This is the hard work that goes on.

The fact that every single day there are, I'm guessing, maybe 30 flights between Singapore and Jakarta, operated by Garuda, Singapore Airlines and many other airlines; the fact that we can all fly comfortably from one city to the next without problems, is not because there weren't any interoperability issues. It's because the interoperability issues were settled. So, between now and the point where all the data movement rules are settled, that is where the hard work lies.

Terence Lee (TIA): With all these investments coming in, where do you see the areas in AI in which more work needs to be done, more developments need to happen? What are the urgent areas you see in the respective countries?

Minister Josephine Teo: For me, there are two. We are building a lot of compute capacity through data centres, but these data centres need to run on power, and one potential bottleneck is that we don’t have enough power. Of course, there is also the constraint of water, but on power alone, apart from each of us continuing to build up our grids to support the development of digital infrastructure, there is also something very exciting going on within ASEAN.

The ASEAN power grid that is being discussed amongst the ASEAN member states could potentially help us to achieve better energy security, which we will all need. And at the same time, it can potentially support our energy needs in a more effective way over the longer term. I think that’s one area.

The other part that I should say is that as AI applications are being implemented across many different domains, particularly in consumer interactions with AI, the question of trust becomes really important.

We come into this building, trusting that it will be safe. That is because somebody checks the sprinkler system and makes sure that fire safety and building safety, have been taken care of. At home, we use many different appliances. We get into a car, not expecting something bad to happen. These are simple things. The only reason we are able to have this sense of security and trust is because these products have been tested; but today AI is reaching the hands, the heads, and the hearts of millions of people, and we are not really sure whether it has been properly tested.

So, the question for me is: what efforts are we making to ensure that this technology can truly be trusted? Apart from some of the efforts that I have talked about, the next step we are looking at is the equivalent of nutrition labels. If an AI application is being used by you, should the application developer be responsible for telling you what it is good for and what it is not, what it is meant for, and what it is not? Just like when you buy a loaf of bread in a shop, it tells you exactly what is in the bread. If you buy an ointment or medication off the shelf, it also tells you the dosage you are supposed to take for your age group – not more, not less. I think we will need something similar for AI applications that reach millions of people and are used by multiple entities.

Terence Lee (TIA): Could you elaborate on the nutrition labels, and what that will look like?

Minister Josephine Teo : We are engaging with application developers to ask them what they can tell consumers that does not give away proprietary information, and what would look like a responsible way for them to declare what the user should and should not use the technology for.

The way we approach things in Singapore is, when we don’t know the answer, or don’t know exactly what the label should look like, we engage with the industry to ask them what they can show that will help the user. And I think the conversation is going well, we have gotten some very useful feedback. We are taking this feedback on board and would love to discuss with our colleagues in ASEAN and other parts of the world whether they would find this kind of label useful as well. These attempts are baby steps, but ultimately, when ready, it will likely start as a voluntary framework. We can then assess whether it is useful and effective, before we consider next steps.

Terence Lee (TIA): With AI investments coming in, how do you ensure that AI benefits not just the large companies, the SMEs, but also the micro-SMEs, the Gojek riders, the Grab riders, in both countries?

Minister Josephine Teo: That’s a challenge with successive waves of technology, whether it's the Internet, mobile, or even previous waves. It's always the frontier companies that benefit most. They are quick to adopt, make full use of what the technology can offer, and then become even more competitive. They grow in size and scale, then disrupt other companies’ businesses, and you have a long tail of companies, generally the smaller ones, who can't keep up. That is a worry with AI too.

On the other hand, there is one particular feature of AI that is potentially quite game changing. Take the fact that you can use AI to code. It's not perfect, but you get somewhere. Yesterday, when I met with some Indonesian companies, they shared that they are Software-as-a-Service (SaaS) companies. But they have their own engineers. So they are thinking to themselves, I'm paying a SaaS provider a lot of money every month to do simple things for me, but my software engineer, with the help of Claude code or some other tool, can create an app to take care of administrative functions so we no longer have to buy services from other service providers.

So that's one way in which even smaller companies can benefit from this technology. If they learn how to use AI effectively, they do not need to be so dependent on other service providers, which would be very costly for them to access.

But I really like what Minister Meutya pointed out, that there is a need to ensure that underserved communities benefit from this technology. Some of the examples I shared earlier were remarkably simple because creating the application was not the most difficult part. It is about understanding their needs and constraints.

At the showcase we attended, where the ASEAN youths, I remember some of the youths telling me that the app that they created uses AI, but it is AI that is very lean. It doesn't need a big server to run. It is AI that is so lean it can sit within the phone, processing is done at the edge, and only 2G or 3G is required. So these innovators were able to overcome these constraints with little investment and not a lot of difficulty.

That's what I think AI potentially can do that previous waves of technology could not. You don’t need very heavy investments to benefit small communities and their very bespoke needs. It is what the World Bank president, Mr Ajay Banga, calls “small AI”. A bank wanting to do complete transformations, or manufacturing companies wanting to use embodied AI to make their production lines super-efficient, that’s “big AI”.

He says there are many examples of “small AI” that we can use to benefit the many communities in our region and enable people to access AI in an affordable way. And that's why it links back to what we are thinking of, which is that we would like to own and leave room for people to access very affordable AI, so that they do not need to build their own.

Terence Lee (TIA): To wrap up, just a quick question for the both of you. If you could use AI to give each citizen a superpower in your countries, what would that be?

Minister Josephine Teo: I have more than one. My Ministry looks after the National Library Board, and we have 26 physical libraries in Singapore. We know that people are flooded with information and content.

Fortunately, our libraries have continued to reinvent themselves. People still visit our libraries, and they can also borrow books through an application. What I would like is if the application, or our library, can introduce one book to each citizen every day with a summary of the book. That way, everyone learns something valuable from a book every day. If we could do this 365 days a year, everyone would become a lot smarter.

I have another wish, and it concerns the SPH Media Trust. We are flooded with news, and there is a lot of heartbreaking news around the world that can be rather depressing. Can SPH Media Trust personalise content to ensure each reader receives one piece of good news every day?

Something that makes me happy may be different from what makes Mr Chan or Hui Fen happy, so let’s have a personalised “good news of the day”. Maybe they can do that!