STACKx Data & AI 2023 (Opening Remarks)

Opening Remarks by SMS Janil Puthucheary for STACKx Data & AI 2023

18 July 2023

Introduction

Good morning everyone and welcome to STACKx 2023. This is our fourth year of this collective, coming together of a community of tech-minded individuals, who are passionate about technology within the Government.

Today, the focus of STACKx is Artificial Intelligence. My colleagues from GovTech and Smart Nation and Digital Government Office, they roll me out once in a while, to persuade you that the Government of Singapore gets technology. So, I have to do my job and I’m quite happy to do so. It’s always a little bit daunting when I have a room full of professionals to convince that the Singapore Government gets a part of technology or Artificial Intelligence.

So, there was a line in my speech about how we get it as much as the next guy. I promise I’ve seen the list of all the next guys so I’m not going to make that claim but I will try to explain what we’re thinking when it comes to Artificial Intelligence (AI). And as we do that, we also have to take into account the underlying data that drives the development of AI. From the 2 Government policy perspective, from a regulatory perspective, we get to deal with the boring stuff while you get to do all the exciting stuff.

Like thinking about Generative AI, the buzzword of the day. A little bit of buzzword bingo in many of these sessions, and one of the things about the buzzword bingos associated with Generative AI is whether or not the speech was written by humans or written by ChatGPT. I will make no claims one way or the other. I’ll see whether I passed the Turing tests today.

The excitement around Generative AI is entirely justified. It has changed the way that we think about automating complex tasks, including how we analyse data using images, sound, music and video.

And today, we’ll be talking a lot about Generative AI, and many other developments that affect our work. As we do that, I do want to call back to the name of this conference: the STACK. And this community is ultimately about the tech stack and the Government tech stack that we have here. As we think about a stack, we have an interdependent series of layers, technologies, and applications. We have to think about them as layers, but we also have to lay them together in the appropriate way. Collectively, we have an understanding that this is the process necessary in order to build a successful series of products. And, it is no less necessary, from the perspective of Government, in order to have the right approach to governing, developing, regulating and promoting the use of AI.

I’m going to briefly touch on what I think are four elements to a successful AI strategy for Government, and how these strategies need to be correctly stacked.

The four elements are the products, the people, the partnerships, and the governance.

Products

Let me start by telling you a little bit about some of our products.

Our team has moved quite fast in trying to harness AI for the public good. About a month after ChatGPT was first announced, the team from Open Government Products, or OGP, one of our GovTech teams, started developing Pair, spelled P-a-i-r, not the fruit pear. Pair is powered by Large Language Models (LLM) and aims to be a suite of productivity tools for the Singapore public service.

We’re just getting going. It has about 4,000 registered users and it plays an important role to allow the secure and efficient use of LLM as a writing assistant within the Government development space. As the officers are writing the documents and writing the submissions, and doing their research, that ChatGPT-powered LLM assistant is there. It’s not quite the paperclip from Microsoft Office; we’ve moved on a few generations. But it does provide a sparring partner for ideas, and it can even provide coding support. We’ve ensured that there are heightened security arrangements in collaboration with Azure OpenAI, and that means that our officers can 4 use relatively sensitive data without worrying about data retention issues. It seems to be working well with the early adopters within the public service and we hope to be able to make it available soon to all of our agencies.

There are other AI-powered tools that we’re exploring to further boost the productivity of our public sector. We want to use AI for what it is particularly good for today – the automation of repetitive and potentially mundane tasks.

But we do have some higher ambitions, beyond just improving our productivity and efficiency. We want to explore what these technologies can bring to the table when we serve our citizens digital services. We have put quite a lot of effort and it has become very much the normal way which Singaporeans and residents here in Singapore interact with the Government and public sector services.

One example is SupportGoWhere. Those of you who were here during the COVID-19 will know that SupportGoWhere was a platform that we put together to retrieve links to resources easily. Healthcare resources at that time, and now a variety of other support for citizens and residents. SupportGoWhere allows you to discover schemes, Government programmes, and COVID-19 support. SupportGoWhere was previously a standard interface – you can key in a postal code, you had a drop down list or some checkbox – it was the way in which we interface with all these types of products not so long ago. Now, it is possible to aggregate services and ideas from across the public sector and you have a natural language input, where someone describes a problem, and we tell you what is the 5 support and where you go to get them. We want to make the experience more citizen-centric but also more natural in its interaction.

If we get this approach right, though we’re still experimenting, a citizen will be able to express their personal story without necessarily knowing how the schemes and support of Government are structured, which is something you still have to know today. A natural story, a personal story, and then have our tools say “Well, these are the schemes and services that we think you need”.

Both these examples, Pair and the enhancements to our support schemes through SupportGoWhere, they are the first tranche of our signature AI use cases in Government. And taking this approach – narrow, specific, problem-solving use cases – allows us to coordinate our development efforts around very practical needs, with a view how we can scale and multiply with similar impact to our citizens across the public service. We need to improve our productivity and efficiency for us, but ultimately, we need to be able to deliver on our mission to our citizens.

People (Capability Development)

This approach that I have described, how we are building LLMs and getting it into the public service, is only possible because we have the right people and the right capabilities. Developing our people and our capabilities is one of the many reasons why we have STACK and STACKx – to integrate with and learn from the knowledge and wisdom gathered through the STACK community.

Our people, however, need more than just technical expertise. They need the tools and the means, and the intent to empower the end-user of our digital services. And it is increasingly relevant at this time when you do not need to have a lot of resources to be able to access cutting-edge capabilities. AI is available to everyone through very simple web interfaces. Once you have access to the internet, you have access to AI tools.

We have to pay equal attention to building broad-based digital literacy amongst our public officers, and also amongst the citizens, so that use of expert digital capabilities is maximised.

We are taking steps to train our public officers. We have a data literacy programme which has helped over 90,000 public officers with an entrylevel understanding of data and AI. We are refreshing that to include Generative AI and LLMs. We need to train our public sector officers on how to do prompt engineering. All of these is now a normal course of business for the Government. We made this programme, ePrimer, publicly accessible for all via GovTech’s Developer Portal.

What we have done within our public sector, the education and training that we think are necessary for staff to be able to maximally extract value out of the technologies, we also think it is useful across a much wider group, and we are making this publicly accessible. But we also need to reach out to the public to drive familiarity and instil confidence in technology. We ran a programme called Data Science Connect, a regular 7 event for public officers to discuss these things, learning about data, LLMs and public sector transformation.

I provided you a few examples of the many efforts that we have in place to try to upskill our public service to match the immediate demands of our time. We need to do more, to go above and beyond, in order to keep pace with the rapid rate of an innovation in AI. If we think we have this settled and put aside, we will be good for a few days but we will rapidly fall behind. We constantly need to think how to keep up to date. And that means finding ways for our teams, our officers across public sector to engage in self-directed learning, to go for training courses, to be incentivised and rewarded to do so, and to work with start-ups and research institutions to be integrated with the community and developments in the private sector. We need to do all these and more, deepen our capabilities, our ability to deploy and develop the right governance for AI, to deliver on our mission.

Partnerships

The Products, the People, are part of our Government AI strategy. They are all important but clearly not sufficient. The third component is Partnerships.

We have a growing list of experts and leaders in this space. We do not have all the answers. We need partners who are keen to journey with us to develop AI and technology for the public.

We have the privilege of early access to some of OpenAI’s latest capabilities. For example, we have access to OpenAI’s multimodal capabilities, enabling the use of images in a chatbot. We have also been able to experiment building ChatGPT plug-ins to extend the functionalities of our products. We have not shied away from that access, and the use of that very cutting-edge technologies, being deployed to some very mundane and pedestrian things which irritate our citizens, and where Digital Government services can change their lives. For example, where to find a parking space, a very day-to-day pedestrian use case, but the use of this type of technology can help.

Our partnerships also include the recent launch of the Artificial Intelligence Government Cloud Cluster (AGCC) with Google Cloud. This partnership has allowed our agencies to have access to Google Cloud’s AI technology stack to accelerate the deployment within the public sector.

We will also work with Microsoft to advance data and AI capabilities, training, and capacity throughout.

Each of these partnerships, and they are a small subset of a long list, they demonstrate a high-level of trust and recognition that others have of us here in Singapore, for which we are extremely grateful and which we do not take for granted. We will build on this confidence. We are grateful for the confidence. We will build on it. We will build these partnerships together. We will accelerate Singapore’s development of AI for the public good.

I would like to take this opportunity to invite all of you to join us on this. If you are from a start-up, a research institution, a large firm, and you are looking to apply your talent, your work, your technical expertise, or your resources to the public good, please contact us, connect with us. Let’s see what we can do. My staff will be very happy to, but my email and my WhatsApp are all public, so you can contact me directly as well. We can see what we can do together.

Governance

The final, and from my perspective, perhaps the most important component of our strategy, is Governance.

We know that we have to balance opportunities, and the risks involved with experimenting and deploying the latest AI technologies.

As we do so, as we progress along this journey, trying to find that right balance between opportunity and risk, our guides, our waypoints are the standards and guidelines on the use of these models, LLMs, and AI technologies within the public sector. We have developed them over a number of years. We have published them, some of them are in their second iteration, some of them well-used outside of Singapore. For us, our guidelines make it very clear that our public officers remain accountable, that the humans remain accountable for the work. AI tools are there to assist, but officers are responsible to ensure that the content is accurate and free from plagiarism.

This is a cautious approach that causes cautious consideration towards these technologies. I think that is our role. We should be taking this cautious, risk-based use-case specific approach but well-informed by the partnerships, by the community, keeping an eye out on where we should going or following fast.

We mitigate these risks by imposing a stringent testing process on the accuracy. I hope I have described how much effort we are putting into the user education before deployment for these LLMs frontends. And the truth is that once we have an application that is public facing, we will have to have even higher standards that will need to be met before even beta-testing is done.

We will have to continue to update these guidelines in a fairly agile manner to match the pace of the development of these technologies, so that we can experiment safely, and yet develop innovative products.

Conclusion

I hope I leave you with the understanding that we get AI, maybe not quite exactly like the next guy, but fairly close. We believe that AI is the next big shift, and that includes how we engineer the Digital Government and use the technology for the public. But we know technology alone is not enough. All of the various elements – the products, the people, the partnership and the governance – must come together in the right relationship. Not exist, they must come together in the right relationship. As we do that, we have a conceptual stack of AI governance, which we build on, we iterate, and 11 we improve year-on-year every time we get this community together. Only then will we be able to realise the potential of AI as a technology for public good.

Thank you all very much, and I hope you have a wonderful conference today. Thank you.

Dr Janil Puthucheary
Senior Minister of State,
Ministry of Communications and Information & Ministry of Health
Minister-in-charge of GovTech Singapore