Meta Llama Incubator デモデイにおけるテオ・ジョセフィーヌ大臣のオープニング演説

Good morning. Congratulations to all winners and participants on completing Meta’s Llama Incubator Programme. I also want to thank Meta for partnering with us to promote AI adoption in Singapore.

As I was going through the exhibition booths and listening to the presentations, I couldn't help but feel a sense of progress. AI, which used to be on the fringe in Singapore, is increasingly in the mainstream.

Let me explain why I think this is so important.

In my previous roles, be it in the Manpower or Finance Ministry, one of our concerns was how technology could be diffused into the whole spectrum of economic activities.

Unless technologies percolate into the corners of society, only a narrow group will benefit. Frontier companies can always be expected to make the most of new technologies, and it is always the far reaches of the economy that get left behind.

When that happens, the economy continues to chug along, but it does not realise its full potential; it does not uplift all companies the same way as the rising tide lifts all boats.

But diffusion of technology is easier said than done. For a general-purpose technology like AI to be diffused, many thoughtful interventions are required.

That was one of our priorities for my team and me looking after the National AI Strategy in Singapore. We identified a suite of enablers to drive activity in the industry, government and research communities because ultimately there needs to be demand. We can’t just do things on the supply side but no one is really using what’s available.

The first edition of the National AI Strategy was in 2019. But if we were to look at the kind of activities that took place after its publication, it is fair to say that AI adoption then was still quite niche. We saw pockets of adoption in healthcare, financial services, transport and logistics, but it was not widespread. It hadn't entered the mainstream.

Fast forward to today. it is hard to ignore the fact that AI has become very mainstream. There are very few gatherings that you would go to that the term “AI” is not mentioned at all.

Even in the community hackathon that I attended in a few weeks ago in the heartlands at Bedok, the grassroots volunteers were also experimenting with the use of AI to bring about better benefits for the residents and help them with their day-to-day lives.

It has become very much a part of the Singapore landscape, whether you are in the Central Business District or the heartlands.

The fact that we have come this far is thanks to developers like you, who have embraced the idea, and because of the partnerships with Meta and other companies that have grown a strategic presence in Singapore.

Beyond going mainstream, there are also many different kinds of AI applications. The examples we see today show the growing sophistication in how AI can be used - to uplift businesses, fulfil compliance requirements, and help others.

For example, Straits Interactive built a suite of offerings to empower non-tech professionals to benefit from AI technology, which is meant to be democratising and useful to the wider ecosystem.

Another example is how LTA built a tool on vulnerability assessment and penetration testing reporting - which complies with the Instruction Manual 8 (IM8)¹ - thus enabling information security staff to focus on more important work like proper threat hunting and to secure our systems better. The same tool can be used to help colleagues in other government agencies, because they all have to comply with IM8.

MyRepublic’s AI Co-worker also experimented with agentic systems to follow up on sales leads and broaden its business base, while at the same time while keeping the human in the loop, not replace them.

These are good examples of ideas that are being implemented to deal with real-world issues. They signal a very important point of inflection in our AI adoption journey, where AI adoption has gone from the fringe to mainstream.

It is also still important for us to uphold an environment where open source is available because not all organisations will have the same access to resources to experiment.

Experimentation takes courage, but very importantly, it takes resources. The Llama incubator programme makes available not just access to an advanced model, but also the engineering resources, knowledge and insight that Meta, together with your partners, have brought to the table.

The Government has to walk the talk when it comes to open source.

Our contribution includes AI Verify. It is a testing framework and software toolkit that is completely open source. The more it is downloaded, the more it is used, the more knowledge and understanding that we have, and the more it can be improved.

Recently, I also launched the Singapore Digital Gateway. This is a way of open sourcing Singapore’s experiences in building up a digital economy and digital society with our colleagues around the world, as we are often asked to share our experiences.

Finally, I'd just like to respond to readings that we may encounter from time to time. Every now and then, we would come across a report that pours cold water on AI adoption. For example, distinguished organisations talking about how AI experimentation is failing, and AI is not producing the returns that the companies would like to see. These are fair comments, but they also reflect just how difficult experiments could be.

Yet at the same time, we have to learn. We have to acquire capability. Even for a general-purpose technology, it takes time for its benefit to be realised.

It's been often said – electrification was available from the late 1800s, but factories did not really use electricity until the 1920s. Why did it take so long?

It's because factories had embedded systems using older technology; they need to work around these systems. They also needed to find new breakthroughs at each juncture. I believe this will be the same for AI.

There are also important lessons to be taken away when studies show that AI adoption seen as a fool's errand. We need to find the sweet spot in the way we experiment, to build long term capabilities, and a sense of trust that the experimentations are serious and should not be given up on easily.

Looking at the use cases that have emerged through incubators like this and through our AI Centres of Excellence, they fall broadly into five categories - what I call the three P's and two A's.

The three Ps where AI can be put to very good use are personalisation, planning and prediction.

The two A's are automation and anomaly detection, that is the ability for AI to trawl through tons of data to find items that seem suspicious.

If we could build a case to accept that the experimentation may not always produce the results that we want, but we will get better at understanding the problem and build capabilities that enable the organisations to achieve success in AI in the future, we will still have gained something.

I'm not pessimistic because of all the studies that suggest that AI adoption will not yield results. It means that we haven't gotten it right as many times as we would like. But it is not a reason to give up. It is a reason to double down.

Finally, in all of these experimentations, it is important that we do not journey alone. It is important to engage people who are in the community because you will learn so much from their experiences.

For example, we have Lorong AI which is a community of AI practitioners that gathers every Wednesday. It's possible that someone in the community has gotten a closer answer than you have. You can then shorten your learning curve by being plugged into an AI community.

As this AI community continues to strengthen, we have a much better chance of solidifying this whole experience and making AI adoption not just at the fringe but in the mainstream, at the front and centre of how we uplift our economy and keep Singapore competitive.

Thank you.

¹Instruction Manual 8 (IM8) refers to a set of government policies used by all government agencies to safeguard Infocomm Technology and Smart Systems (ICT and SS) assets.