How Singapore is Using AI in the Public Sector

Technology shifts are like giant waves. When one forms, it happens rapidly and is unstoppable. When the wave hits, it forces people to paddle much faster instead of settling into a routine, says Joseph Tan, a deputy director at GovTech Singapore.

Tan painted this vivid illustration at the STACKx Data & AI conference last month as he shared about Singapore’s plan to empower public officers and democratize AI.

It is not all gloom and hard work though. “Every new wave also brings about new opportunities,” he said.

Equipping public officers

With that in mind, Tan says his team aspires to help the average users “stay afloat” and empower them to make meaningful use of technology. This might entail removing or reducing the fear around new technologies, equipping them, and inspiring them to get started.

And in the case of data science and AI, the benefits are felt even when users do not dabble directly in them.

“What we want to do is to equip public officers with the knowledge, the awareness of what data science and AI can and cannot do,” Tan explained.

“We also want to increase their sensitivity towards data quality. Because even if they are not the ones who will be analyzing the data [on a day-by-day basis], they are very likely going to be the ones that will be collecting the data.”

Tan outlined several programs to upskill public officers and support government agencies on their data transformation journey. This includes a four-hour basic data and AI literacy primer packaged as an e-learning resource that public officers can take at their own time and pace.

Already, 90,000 public officers have gone through it and given very positive feedback. Tan shared that the first module has just been uploaded for public access from the GovTech Developer Portal. Additional modules will also be progressively rolled out.

Taking power users further

The initiatives have also allowed the GovTech team to train up a pool of power users.

“[Power users] who we identified become our data champions. They are able to use the technology better than the regular users, and can harness the technology much better to their work.”

“What is most rewarding is when we observe data and AI practitioners take the technology and go much, much further. What might have taken them weeks or months to do previously, is now an API call away.”

Even existing software engineers are benefitting from generative AI. “We have been working with a lot of software engineers, they are picking up skills and they are trying to see how generative AI can be applied to the systems and applications that they have been building for the past few years.”

While the secret sauce to successful technology adoption lies in the users, supporting them calls for building communities to spur and empower them.

“Building a community around the new technology is important so that they can discuss, network, learn, and share. Empowerment is equally important; no matter what you do, if the staff do not get their hands dirty, or if there's a lot of friction, there's a lot of obstacles, [then] it's not going to work.”

The Data Science Connect (DSC) is a meetup platform that GovTech created for public officers to come together. According to Tan, industry professionals and data leaders are invited to share their knowledge, expertise, and their transformation journey at DSC meetings.

New AI tools

One tool that the GovTech team is building revolves around the creation of a generative AI assistant to help public officers work quicker and more effectively.

Using a large language model (LLM) out of the box would be simpler, less expensive, and straightforward to use, Tan notes, yet “is nowhere near how we envision personal AI systems to be.”

On the other hand, training a new foundation model or even fine-tuning an existing LLM does take quite a bit of time and effort – and a lot of data.

The approach that GovTech has adopted is to target the “sweet spot” between using a vanilla LLM and fine-tuning a model. This gives users the ability to impart their knowledge to the AI through a retrieval-augmented generation (RAG) approach.

“We've tried this out with GPT-4. And it feels it's fairly reasonable… and it gives us a certain confidence that this can be developed into something useful.”

What’s next for the team?

“There's never a time when technologies have come to a standstill. But by successfully overcoming the earlier waves, we would have built up the foundations for us to then tackle the next big wave.”

“When that happens, hopefully, it won’t be just us having to face the wave alone. But it’ll be us alongside our AI assistants, Tan summed up.

Paul Mah is the editor of DSAITrends. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose. You can reach him at [email protected].​

Image credit: iStockphoto/Jui-Chi Chan