Transcript of Minister Josephine Teo's Fireside Chat at Family Business Summit

Moderator: Minister, you are the face and the person behind many of the AI policies that Singapore has been unveiling over the last few years. Singapore’s economy is expanding, fuelled by many AI developments, to such an extent that the digital economy made up about 19.3 percent of the economy last year, and is set to grow even more. Do you think this growth is sustainable, given Singapore’s land and manpower constraints? 

Minister: Zuraidah, thank you for having me at the SCMP Family Business Summit. I also want to say a big thank you to Joe and Sergio for your very warm welcome. It's a great pleasure to be back in Hong Kong, and thank you for the opportunity.

To your question – is there a limit to what we can do with AI? I would put it this way: we want to keep the growth story going as much as we can. Singapore is a place where, whether you look at it from the angle of breadth or depth, the foundation is certainly very strong. However, when you are operating from a certain base and the cost environment presents challenges to businesses, we have to ask ourselves: what's the next thing that can happen?

Ultimately, if you want to grow, you have to raise productivity. When we  looked at what we could do with AI, it presented so many opportunities that were hard to miss. Rather than ask, "Is there a limit to how much we can do with AI, given the land and manpower constraints?", I think we should flip the question around: wouldn't we be even more constrained if we didn't learn to master this technology? That was our perspective. The question then becomes: how do we make the most of the resources available to us? That was the approach we took.

Moderator: For the sake of the audience here tonight, could you walk us through, in very broad strokes, what are some of the key principles that are guiding Singapore's AI policy?

Minister: There isn't a single way to describe it, but it builds on what we've been doing in digital governance, and how we think about being a Smart Nation. There are five things that you will always need.

First, you need infrastructure. In the AI era, this includes ensuring access to computing power and cloud infrastructure, as well as putting in place the right enabling environment. For example, just this evening, there was a debate in the Singapore Parliament on national digital identity infrastructure for AI agents. The fact that Parliament is able to debate a question like that says something about where we are.

Second, you need capabilities, which, to my mind, are threefold. These include workforce capabilities, because people must be able to use the technology; enterprise capabilities, because most of the value is not created by individuals but by organisations, which must be able to make full use of the technology; and governance capabilities. This is where we challenge ourselves. If we are not using the technology deeply, we won't really understand how it works. And if we don't understand it, how do we put in place the right policies to govern it?

The third piece is that it would be too narrow to think that the government has all the answers, resources and means of reaching every single citizen. You are going to have to rally the whole of society.

Fourth, you need rules and regulations, but you must always be careful about introducing them prematurely. Think about safety in cars and aviation. It took us a long time to  figure out the importance of seat belts and airbags. Until you understand what works, it can be premature to start putting rules in place.

Finally, the fifth thing you must always do in digital development is to look for partners internationally. So those are the things that I remind myself of.

Moderator: Out of these five imperatives, what is your biggest concern right now? Is it about the possible displacement of jobs, cybersecurity, cross-border attacks, or biological warfare?

Minister: All of the above. I would say that you worry about these things, but worry only gets you so far. The next question is: what can you do about them? We try, as best as we can, to systematically understand the real risks.

For example, when it comes to how AI can go rogue and potentially break into systems, even without intending to do so, the fundamental issue is not so much that AI has developed new ways of attacking systems. The approaches are not particularly novel. They are known ways of infiltrating systems. The real problem is that systems have not been very well guarded. Vulnerabilities exist, and addressing them gives you a better chance of preventing such attacks.

Moderator: So you're not too troubled by the naysayers who say that AI poses an existential risk to humanity? Could there come a time when superintelligent AI systems take over, and we find ourselves at their mercy?

Minister: A friend of mine is a physicist and professor at MIT, Professor Max Tegmark. Some years ago, Max wrote a book called Life 3.0, in which he very eloquently describes a world where AI becomes superintelligent. I think he posed a very interesting question: if AI can run human society better than humans can, why don't we allow it to do so?

So, to your question about existential risk, I think that, as a matter of interest, human beings should try to understand what that world might look like. But on a day-to-day basis, my role doesn't allow me to dwell on these things. As a matter of personal interest, yes. But in my role, I spend more time on  cybersecurity, misinformation, and concerns about jobs.

Moderator: Talking about jobs, I want to cite a speech you made in 2017, before you took over your current portfolio. The speech was titled "Jobs of 2037”, in which you said, and I quote: "If the past is any guide to the future, technology has often resulted in a net creation of jobs.” Almost 10 years on, with AI in the picture, do you believe that this will still be the case? And if so, do you think the challenge is even greater now, given the mismatch between the skills of the workforce and those required for new jobs?

Minister: Let me put it this way: it took about five or six decades for electricity to go from early technology developments to widespread use in factories. In contrast, it has only been a few years since ChatGPT came onto the scene, and we are already talking about AI agents.

I think our biggest challenge today is that the timeline is very compressed, whereas human society previously had much more time to adapt to the impact of technological change. AI hasn't given us much time to pause and reflect more deeply. Everyone would like to benefit from this technology, so we push adoption and encourage individuals, companies, and whole sectors of the economy to think about how they can use AI intelligently. But we also have to create room for our people to adjust to these changes, and it is worth unpacking just how big the challenge is.

In the immediate future, we want to help members of our workforce be as skilled in AI as possible, because that is potentially key to remaining relevant to their employers. It is not just about using AI for personal productivity, but also helping their employers reimagine how their businesses should operate. The more we equip our workforce, the better their chances of helping their employers succeed, giving employers a reason to retain them.

However, as employers change the way they operate, there are bound to be jobs that AI is going to do better than humans. The question is: what do we do with the people who used to occupy those jobs? We need another layer of support in society to provide an adequate cushion and give people time to adjust, when technology may not necessarily afford them that time. This comes down to whether we have developed the infrastructure for continuing education and training, set aside the resources, and built a system to deliver this support to citizens.

There is another layer that we have to think about. Today, AI is very capable of performing junior-level and entry-level jobs. However, in time to come, organisations will need people with stronger domain knowledge to take on greater responsibilities. So, we potentially have a broken rung on a career ladder. Is there a way for us to overcome this? Can the government intervene to ensure that the career ladder remains intact and find ways to fill in the gap? I would say that addressing the impact of AI on jobs requires a multi-pronged approach.

If I can add one more consideration for policymakers, it is the broader structure through which economic benefits are delivered to people. Wages are a very big part of this. If organisations transition into an environment where more of their expenditure goes towards tokens and less towards wages, we have to ask what happens to the revenue model of the state. What will sustain the way governments deliver public services? So that's another layer to think about. It's a very heavy topic for dinner, I'm afraid.  

Moderator: As we all know, Singapore was one of the founding members of Pax Silica, and we now have World Artificial Intelligence Cooperation Organization (WAICO), a Chinese-led initiative. You've said that Singapore is considering whether to join and that it is not meaningful to compare membership of the two initiatives. Do you see any potential contradictions in Singapore being a member of both?

Minister: At any one time, we participate in international organisations of varied nature, and we are not the only country that does so. It is very common. Our participation in these organisations takes place at different junctures in our development, serves different objectives, and aligns with our evolving priorities.

With that as a backdrop, it's almost like managing an investment portfolio when we receive an invitation. We look at it and ask ourselves: how does this fit into our priorities in AI and digital development?

Some years ago, because Singapore was the convener of the Forum of Small States (FOSS) at the United Nations, we created a digital pillar. Many members of FOSS are also members of WAICO, so Singapore already has many opportunities to interact with them. In fact, we run capacity-building programmes and share some of the things that we've done with them.

There are also other platforms where we have opportunities to contribute. I am one of the commissioners of the International Telecommunication Union’s Global Commission on AI for Good, which brings together companies from the US and China, so we have yet another platform for engagement. When we receive an invitation to join an initiative such as WAICO, we look at it seriously and consider the obligations and commitments involved. WAICO is a serious organisation with an established structure, so we're assessing how it fits into our priorities.

In terms of our bilateral relationships with the US and China, we engage through different platforms. With China, we have a Digital Policy Dialogue, where we meet regularly to discuss issues of mutual concern, including AI safety, AI governance, and how we can promote AI adoption. We will continue to look at these and assess what would be of interest and aligned with our priorities at any point in time.

Moderator: Before I open the floor to questions, I think we do need to touch on the subject of sovereign AI. The Singapore government has a very enlightened policy in focusing on targeted sovereignty. Could you walk us through how this looks like, particularly in ring-fencing mission-critical spaces, such as public services and government agencies? How have you embarked on that journey, and what does it look like now?

Minister: When I interact with ASEAN digital ministers or anyone at international fora, this topic comes up frequently. AI sovereignty is often thought of in ownership terms – that you have to build and own the entire AI technology stack in order to achieve sovereignty. But if you define sovereignty in those terms, I think very few countries in the world can achieve it.

The AI supply chain is not easy to contain within one country, and building an entire technology stack would be very expensive. Many countries also have different fiscal priorities. In Singapore’s case, we have an aging population. From that perspective, I think defining AI sovereignty in terms of ownership is unrealistic. What I think countries or companies want, is to be able to use technology on their own terms. To do that, they need their own capabilities. If they are unable to use technologies from different stacks, then even if they have access, that access is of very little use.

These capabilities do not simply appear overnight, you have to invest in the effort to growing them. That means making a very deliberate effort to be plugged in to different innovation ecosystems,  engaging with those organisations, and inviting them to be part of your innovation ecosystem. That is what we have been deliberately trying to do.

If you define AI sovereignty as being able to use the technology on your own terms, by developing your own capabilities and plugging in to different innovation ecosystem, you stand a better chance and that's the way we look at it.

Audience: With advances in technology and AI, I think there's a changing thought in how we educate our children. What is your thought process on the introduction of AI within the educational system, and how do you see your country continuing to give your children the best advantage within that system?

Minister: I must say that the approach we try to take in Singapore is that we help students learn about AI, learn to use AI, and learn beyond AI. This means that AI cannot be the first and last of how they experience and learn about the world. It is a skill that we want them to have, but it is not the only skill they need. In many of my interactions with organisations who have implemented AI, what stands out is how essential human empathy and decision-making will continue to be. The worry I think we have is that if we spend all our time just learning AI skills, we may neglect the human skills that still need to be learned and taught.

When I go to universities and polytechnics and meet with young people, my appeal to them is this: if you want to stand up and ask a question, try not to ask ChatGPT to come up with one. Ask me something based on your own understanding of what is happening in the world, what worries you, or what your aspirations lead you to think about. When you depend on AI to manage every aspect of your life, you are taking a risk. So, every chance we have, we try to remind people of the importance of learning beyond AI.

Audience: How do we get the older generation to embrace AI and not just say, "Oh, I'm not paying attention to any of that." I would like to know what to say to my parents and to the older generation.

Minister: Let me tell you a funny story. I bought my elderly parents a new smart TV, and I was very excited because I wanted them to be able to use an AI chatbot through the television, where the words would be bigger. I thought I would now be able to help them figure out how AI can work for them.

But what upset them most was the remote control. With the old remote control, they simply pressed the buttons up and down to change channels. What we don't realise is how much our habits have evolved with the devices we use. The idea that pressing and holding a button, as opposed to pressing and releasing it, would produce a different response and was something my parents were not able to figure out.

I also realised that they had left the workforce well before the concept of scrolling became commonplace. They  know how to swipe, but they don't know how to scroll. My takeaway is that once you step away from technology, it becomes very difficult to get back into the fold because technology develops and evolves so quickly.

But we shouldn't give up. In Singapore, we have tried to encourage seniors to find the fun in using AI. A couple of weeks ago, I was at a community club where we were teaching seniors to use AI to generate greeting cards for their friends. It's a small step, but it gives them the confidence to see that technology is not all that difficult and that they can learn bits of it. Will they become proficient at building AI agents or vibe coding? Not so easily, but it is important to help them take the next step and gain a little bit of confidence.

In Singapore, we keep track of how our seniors are learning digital skills, and year by year, they get better. They are interested in things like finding good deals online and improving their health, which motivates them to learn. So, if you use those as entry points, I'd say we don't need to give up.

Audience: You're responsible for cybersecurity in Singapore. A few months ago, Anthropic came out with Mythos but they did not make it available to everyone. They had a programme called Glasswing, it gave early access to around 40 companies, including banks and critical infrastructure companies. My question has two parts. First, given Singapore’s participation in Pax Silica and being an allied country to the United States, if Singapore were to ask to be part of the Mythos programme, do you think it would have been granted access to the programme? Second, in the same scenario, if China had developed the same sort of cybersecurity capabilities, would Singapore seek China’s help?

Minister: It is helpful to be on Glasswing but do you need to be on it to improve your cybersecurity? The answer is not necessarily. Many of the capabilities are already available in open models, and they are quite performant. If you use those models well enough to find your own vulnerabilities and build sufficient capabilities to address them, you will be much better protected, even if you are not a member of Glasswing.

That is what we keep in mind. It is good to have access, and we should always try to have access to the most sophisticated technology. But it would be a mistake to think that that's the only way you can strengthen your own capabilities and cyber defences. We are constantly talking to our colleagues in China, and they have made many of the models available as open-weight models. We are strong proponents of this approach, for that as we believe it strengthens the environment for AI adoption, but also because we believe it is the only path to ensure that most of humanity can benefit from this technology.

So, there is no reason for us not to be able to use various models. In Singapore, you have access to all the models. The way we think about it is that all organisations have to assess them based on their performance, including their cost. You then have to consider whether parts of your system require a higher degree of security and resilience than a particular model can provide, and make your own assessment accordingly. We don’t start with an attitude that certain models can or cannot be used. As much as possible, we try not to limit our own choices.

Audience: How can AI and technology benefit SMEs, and what policies does Singapore have in place to support them? In Hong Kong, SMEs make up almost 90% of all the companies so I would like to get your insights on this.

Minister: SMEs comprising the bulk of the enterprises is true of most economies, and Singapore is no different. The way we think about it is that the diffusion of technology always runs the risk of companies at the frontier running ahead, while leaving a long tail of companies that remain unfamiliar with the technology for quite a long time.

In Singapore, Bee Cheng Hiang’s customers received an email that exposed the email addresses of many other customers when they weren't supposed to see them. It turned out that they had an employee who built an AI agent to handle the emails. Unfortunately, it was not built very well so the AI agent ended up making this mistake.

What this story illustrates to me is that one of the challenges and opportunities for SMEs today is that AI tools are very accessible and not that difficult for individuals to use. Sometimes, it is the SMEs themselves that haven’t quite caught up. If we want them to benefit from using AI, they need to allow their employees to use these tools more. But if they don’t improve their own understanding of how the technology works, then they won't be able to put enough safeguards in place and may eventually have to stop using it.

So what have we done in Singapore? We decided that helping SMEs develop a better understanding of AI is important. They won't be able to build all the safeguards themselves and that's where intermediaries, such as cloud service providers, can play an important role. We have a programme to help SME leaders build their capabilities. At the same time, we also have a programme that encourages them to build tools that many SMEs can use. Through a subscription model, SMEs can access tools that have already been tested, giving them greater confidence and familiarity with the technology.

The analogy I use is IKEA – its products have been tested and they are not expensive. Anyone can assemble an IKEA table. If you want technology to be widely adopted and diffused, you need to think of the IKEA model.

Moderator: How do you use AI in your work and your personal life? Do you have any secret AI indulgences that you can share with us tonight?

Minister: I use it all the time. I vibe coded my own AI agent that scans the world for news articles and it reads about 16,000 articles every single day – that's a lot more than I can manage. It then produces a summary for me based on my interests, which gets delivered to me every morning through my WhatsApp.

I'm now very interested in building another AI agent. I've been thinking about how to use it because I have not visited Hong Kong for a while. My colleague at the Mission told me about the very exciting and vibrant concert scene here and tickets are apparently very hard to get. So I'm trying to find a way to use my agent to keep track of when the ticket sales are launched and to help me get the tickets. So, wish me luck, Zuraidah!