Brainstorm AI Singapore 2024: Singapore tightens AI regulations on political content
Josephine TEO, Minister for Digital Development and Information, Singapore Moderator: Clay CHANDLER, Executive Editor, Asia, FORTUNE; Co-chair, Fortune Brainstorm AI Singapore
Category
🤖
TechTranscript
00:00Good afternoon, everyone. And Mr. Teo, thank you so much for making time to join us.
00:05Good afternoon, Clay. Glad to be here.
00:07We're really thrilled to have you. And everyone in Singapore has been so helpful and supportive
00:13to us in helping us to understand how AI is being adopted here in Singapore and around the rest of
00:20the region and really appreciate that. Jeremy mentioned that Singapore was one of the first
00:25countries to embrace and unveil a national AI strategy. And I wonder if you could,
00:31for people who may not know about Singapore strategy, say a little bit about that and
00:35perhaps how Singapore's new national AI strategy 2.0 is different than the first one.
00:42Well, thank you for that opportunity. You know, back in 2019 that we put forward the
00:48national AI strategy, it wasn't as though no work had been done. In fact,
00:55about nine months even before the strategy was launched, we did put forward a set of principles
01:03for AI governance. So what was quite interesting for me is that even in those years,
01:09we thought about governance before we actually launched a strategy.
01:14Fast forward to 2023, when we put together our national AI strategy 2.0,
01:20the circumstances had changed quite a lot. By that time, generative AI had become
01:27something that was very accessible. And so a very important consideration is, along with
01:34democratising access to AI, how about the democratisation of risks as well? You will
01:41find it being more pervasive. There will be many more settings that AI may be used or misused.
01:47So how does our strategy take into account both the democratising access benefit as well as the
01:56risk democratisation of generative AI? So in NAIS 2.0, I would say a couple of features stand out
02:05in terms of the infrastructure that can support an AI ecosystem. How much compute capacity do you
02:11make available, particularly for segments of the community such as the research community
02:17that may find it harder to access compute? That was one important feature. And of course,
02:22we also know that there is no good AI without good data. The question is, how do we harvest
02:29data in a responsible way, taking into account people's concerns about privacy, taking into
02:35account companies' and enterprise concerns about data protection, and foster an environment that
02:43has responsible as well as active use of data? And I think another very important aspect to go
02:50along with it is that in any new field, the pool of talent is always going to be a concern. It's
02:58not broad enough, it's not deep enough. So what measures can we put in place to build up the AI
03:04talent to support the healthy growth of an ecosystem? And if you wrap all of these together,
03:10I would also say that it has challenged us to think differently about what it means to achieve
03:16excellence in AI governance, which is not just the guardrails that you put in place, but also
03:22the promotional factors that can help this ecosystem grow. I think one of the things that
03:26has impressed me, as I've learned about the Singapore AI strategy, is that it seems to be
03:31very collaborative. And there seems to be a pretty constructive relationship between, as you mentioned,
03:38academics, people in government, people in businesses, a lot of informal institutions.
03:43And as someone who is an American and kind of keeping an eye on the American system of
03:48adopting AI, but spends a lot of time in China, I see the sort of huge polarization between those
03:53two systems, where China has put the very dense regulatory framework in first,
03:58and the U.S. has almost an allergy to any suggestion of a regulation. And Singapore seems
04:04to have found this very interesting middle way, where it doesn't have explicit regulatory framework,
04:11but it does have rules and systems and norms and kind of understandings, which is interesting.
04:17Do you anticipate that Singapore down the road will have more explicit AI-specific regulations?
04:23Well, in the first place, there are certain kinds of AI-generated content, for example,
04:29that we already have guardrails against. So, for example, we have a law against fake news.
04:34How do you deal with the content if it is AI-generated? Well, it turns out the law is
04:39technology agnostic. It doesn't matter whether it was a person who wrote up that fake news or
04:46used AI to generate that fake news. And that already helps us to have some buffer.
04:54But there are also other areas of AI risks that need further effort. Specific again to AI-generated
05:01content, what if the AI-generated content was applied in the context of a political election?
05:08And this is something that already is observed in other countries where elections are taking place.
05:16I've said this before, that we will have to look at what are the measures and what new
05:22laws need to be introduced. We are actively studying the Koreans, for example, put in place
05:29a law to ban political content that is generated using AI in the 90 days that is the run-up to
05:36their election. But in our case, we don't have a clear window of a 90 days. So, to what extent
05:42can we replicate a law like that? But I should just add that AI-generated content is just one
05:48part of it. There are other aspects of risk that we also make available a set of principles that
05:54we'd like people to adhere to. So, just as in 2019, before we had the strategy, we had the
06:01Model AI Governance Framework, this time round, very soon after we published our updated National
06:07AI Strategy, we also updated our Model AI Governance Framework for generative AI. We've
06:13taken it one step further. We say, here are the principles, but how do you actually demonstrate
06:19that the principles have been adhered to? So, we have a software toolkit and testing framework,
06:25which we call AI Verify, and that was very applicable in the context of classical AI,
06:31traditional AI. With generative AI, we created another project, we call it Moonshot. The idea is
06:37how do you test for the safety of AI systems with generative AI? So, those are the practical ways
06:42in which we try to move forward. So, it's an interesting combination of, as you say, guardrails
06:48and things to kind of encourage and promote innovation. You mentioned data, which is a really
06:54interesting component of this equation in the case of Singapore. So, as I mentioned when we
07:02chatted the other day, I spent some time trying to understand how a country like Singapore,
07:09with 6 million people less, very small, limited land, what it does to get access to lots of data.
07:19You will be familiar with Kai-Fu Lee, who speaks often at our events. He's got this book that
07:24argues that we're heading into this age of AI superpowers, that the two big superpowers will
07:31be the US and China, and they will be superpowers because they have access to the new oil of the
07:35digital age, which is data. So, how does a small city-state like Singapore compete in the age of
07:42AI superpowers and big data? So, this point about the size of our population and the availability
07:49of data has been talked about before, but I think you can look at it from a different
07:54perspective. If you consider the size of our GDP and use it as an indication of the breadth and
08:01depth of activities that are taking place in Singapore, and that every single one of those
08:08activities generates a data point, then actually maybe the data is not as small as we think it is.
08:15In fact, there are also ways in which the training of AI models need not necessarily depend on a very
08:21large amount of data or the data being directly available to you. There are techniques in privacy
08:28enhancing technologies that will allow, for example, federated learning to take place. You
08:34may not be in possession of the data, but it doesn't prevent you from being able to extract
08:40the learning from that data. So, investments in privacy enhancing technologies is one of the ways
08:48in which we can complement the need for data that also requires us to push ahead on things
08:55like cross-border data flows, making it easy for companies that may have a multinational presence
09:02to be able to, within their own organizations, be able to move data around. So, those are the
09:08parallel efforts that we are doing in order to try and deal with the question of data.
09:13So, my article is in the next issue of Fortune magazine, which will come out soon. It just went
09:17on the website today. But one of the things that I mentioned in there is that Singapore has been
09:23very proactive in thinking about how to use some of the assets that are unique that it does have
09:29to compensate for the fact that it doesn't have a huge population. And there's the airport,
09:35which is one of the busiest airports in the world. The container port, now the second busiest
09:41container port in the world. You've got some of the world's biggest banks right here. And then,
09:47as you mentioned, there are all these multinational companies that are coming in. So, it makes for a
09:51very interesting ecosystem where there's a lot of technological sophistication and data, even
09:56though the population might be limited. The concept is very similar, whether it is for
10:01the aviation hub or the maritime center or for financial services. We've sort of positioned
10:09Singapore in the flow. You are in the flow of capital. You are in the flow of goods. You are
10:15in the flow of ideas. You are in the flow of people. And similarly, you can also be in the
10:21flow of data. That's how we're thinking about it. One of the bottlenecks on that that's been a bit
10:26of a challenge for Singapore has, of course, been data centers. And I wanted to ask you about that
10:30very quickly. Singapore has 70 plus data centers. That's an enormous number for its size. And yet,
10:38this is a place that is resource constrained in many ways. Land is scarce. Energy is expensive.
10:44There's no renewable energy to speak of. And it's hot. And so, running a data center in Singapore
10:50can be a daunting proposition. Singapore had a moratorium that was imposed in 2019 because
10:57of concerns about data centers just consuming too much energy. That's being lifted. There's a new
11:05blueprint in place. And where do you see that going? Can Singapore continue to add data centers,
11:11or will it have to start parking things next door in Johor? I'm really glad you brought up the fact
11:16that we're very hot. We can only ever give you a warm welcome in Singapore. You have to remember
11:20that. But let's go back to the question of data centers. It's not just the number of data centers
11:25that matter. It's the combined capacity. And if you look at our data center capacity relative to
11:32the size of our GDP, and if you compare it to, say, Japan, or if you compare it to China,
11:38actually, we have way more. If you take Japan, for example, it has got a population that is
11:45maybe 20 times the size of Singapore. GDP may be 10 times the size of Singapore. But actually,
11:50if you look at the data center capacity, I don't think it is more than one and a half times
11:54more than Singapore. So you do that comparison across many other jurisdictions, and you come to
11:59the conclusion that actually, in Singapore, you already have one of the densest data center
12:03capacity. But it doesn't mean that we don't want to accommodate more growth. Because we do accept
12:07the fact that with AI being a more prominent technology undergirding our entire economy,
12:15you will certainly have additional demands being placed on data centers. More workloads need to be
12:22hosted in Singapore, just so that the inferences can be done in the time that they need to be done.
12:27So some growth is going to be helpful. The question is, how do we do it whilst also being
12:32able to fulfill our commitments to the net zero pathway, as well as being responsible in
12:40sustainability? Tropical DC standards is one of them. So we introduce a new set of standards. If
12:49you're operating a data center in a tropical climate, do you need to operate it at the
12:53temperature that is so low? Or actually, even with one degree higher in temperature, you're okay. How
13:00do you do that? And then a more sustainable way to the future is to move towards greener DCs.
13:06This is either the operations becoming more energy efficient, or the source of energy that is used to
13:12power the data centers will have to become greener. So those two pathways, we think, will give us some
13:18room as to how much. It also depends on how quickly the technology evolves.
13:24I'll ask you maybe very quickly, if I could, about the prospect for regional collaboration,
13:30as you think about how to project your weight in terms of data and AI. Singapore is one of the
13:37most advanced AI adopters in this region, if not the most. But yet there's some data that has to
13:43be sovereign, has to stay on shore. How do you see the scope of possibilities for collaboration
13:49with the rest of Southeast Asia, and particularly Malaysia? Very much so, because ASEAN was one of
13:56the first to have a data management framework. It's been around for a few years. And on top of that
14:02data management framework, we've been able to put together a set of legal requirements that we can
14:10meet on both sides in order to facilitate the movement of data across the two jurisdictions.
14:18And this is not just between Singapore and Malaysia, it is pan-ASEAN. I would also say that
14:25there is a lot of interest and commitment to ensure that progressively we move towards more
14:33adherence to international standards in cross-border data flows. And so not just Singapore,
14:40several other countries are members of the US-led global cross-border privacy rules
14:46that was introduced some years ago. So it's not just ASEAN, it's not just regional.
14:52Not just. It could be really global and have a foot in multiple camps.
14:56Very much so. I think my colleagues and I, amongst the ASEAN digital ministers,
15:01all recognize the importance of water. I would just, importance of data, what am I saying?
15:06Water is, of course, also important. I would just like to say something in response to what
15:10Kai-Fu has said about data being the new oil. One small problem with describing data as a new oil
15:18is that it gives the impression that data depletes. Because once oil is used,
15:25you know, you can't reuse it. Data is not like that. Data can be reused many times over,
15:30it can be recombined. And data is actually more useful when it is being reused in many different
15:37contexts. So when we think of data as a new oil, it is very valuable, and that's what makes it
15:43look like the new oil, but it doesn't deplete like oil will deplete. So we have to think of data in
15:49a different way. We do want to protect it. There are legitimate occasions where the data may be
15:56security-sensitive, no country would want to share them. There will be data that will be
16:01enterprise, business-sensitive, and no right-thinking business owner would want to share it.
16:07And you need to find ways to protect them. But beyond that, there are a lot of benefits to being
16:12able to share the data, and there are also increasingly more secure ways of doing so.
16:18Well, it's fascinating to see all the different things that Singapore is doing, as I sort of
16:22mentioned in the article. Singapore is a great case study for other small and medium-sized
16:27countries that are looking to kind of refine their strategy and to maximize the resources
16:32available to them. Minister Cho, thank you so much for joining us today.
16:37Thank you, Clay.
16:38Appreciate it.
16:38Really appreciate this opportunity. Thank you very much.