Feasibility of Establishing Body to Regulate Artificial Intelligence

Mdm Deputy Speaker, I thank Ms Tin for her question. Singapore supports the responsible development and deployment of artificial intelligence (AI) to unlock its transformative potential on our economy and society. Our governance approach is a practical and risk-based one, as explained at the Sittings of 21 April 2023 and 9 May 2023.

AI is fast becoming a general-purpose technology that is applied across many sectors and use cases. We cannot and should not adopt a one-size-fits-all approach to regulate it, nor is it realistic to address in advance every risk out there.

Nonetheless, the Government hopes to see AI deployed in a responsible and ethical way, aligned with international norms. The Infocomm Media Development Authority (IMDA) and the Personal Data Protection Commission (PDPC) have several initiatives to guide the development and deployment of AI.

In 2019, PDPC published the Model AI Governance Framework to turn AI ethical principles into implementable business practices for companies to safely deploy AI solutions.

IMDA has open-sourced AI Verify, an AI governance testing framework and software toolkit. It strengthens trust by enabling objective validation of a company’s implementation of responsible AI. It is, however, a minimum viable product that we hope will grow over time.

Later this year, PDPC will be issuing Advisory Guidelines on the Use of Personal Data in AI Systems to provide guidance on how the Personal Data Protection Act will apply to the collection and use of personal data in AI systems for decision-making, predictions or recommendations.

These initiatives have been noticed and commended by our international and industry partners as practical steps for thinking about how responsible AI developments can be implemented. We will continue to monitor developments and are prepared to adjust our implementation approach as knowledge and understanding evolves.

AI governance is also an area that benefits from more international cooperation. In this regard, Singapore is engaging widely with our international partners through multilateral platforms, such as the Global Partnership on Artificial Intelligence (GPAI). As the upcoming Chair of the Fourth ASEAN Digital Ministers Meeting (ADGMIN), Singapore hopes to work with fellow ASEAN member states to develop an ASEAN Guide on AI Governance and Ethics.

In addition, we are engaging other like-minded partners, such as Israel, the Republic of Korea and the UK, with whom we have signed Memoranda of Understanding (MOUs) to cooperate on AI and emerging technologies. With the US, we have been partnering their National Institute of Standards and Technology (NIST) to advance our alignment on AI governance.

I thank the Minister for the reply and also addressing my question on AI a few times. I would like to ask, given that AI has been progressing at a speed that was even faster than expected – now it is artificial general intelligence (AGI) and the next one, we do not know when, may be super intelligence – and the impact on ordinary citizens' lives have been far-reaching and wide-ranging – economic, social, legal and so on – can I ask how is the Ministry working with other agencies to look at the impact of AI on a more comprehensive approach?

I understand that it may be quite difficult to legislate AI altogether, especially as it might inevitably hamper innovation. But I am wondering whether there would be continuous effort to look at more upstream factors so that we can better control and mitigate any impact there may be in generations to come.

Mdm Deputy Speaker, I think Ms Tin hit the nail on the head when she talked about working with other agencies to understand how AI is being deployed in their respective sectors and also the industries that they oversee and then to identify what should be the right approach in regulating them.

If I can give an analogy, when I am asked this question about AI governance, there is no doubt it is happening very quickly and we will see accelerated deployments potentially across many different sectors, but it still takes time to understand fully how the risks present themselves and what to do about moderating these risks.

The automobile industry is a very good example. When cars were first made available, their reach was not very wide. People could not afford to buy automotives. Those came a little later. It would probably be the same for AI deployment, too. AI, for example, takes a lot of compute power, so, it is not cheap to implement AI systems. As such, the early implementations will still take time to present themselves.

But going back to the example of automotives, it was not immediately obvious to people that what helps to prevent deaths and injuries in a motor vehicle accident would be something like a seat belt. Seat belts came much later. And subsequent to seat belts, then people figured out that having an air bag is quite useful, too. But these kinds of measures, they do not present themselves at the outset. We do not really know what will work. It takes research, it takes a lot of exchange of knowledge and understanding. And then, after a while, we figured out that maybe, seat belts should be made a requirement; maybe air bags should be required as well.

I would also say that using automotives as an example, it is not just the thing itself, the device itself, the equipment itself, that can have safety features built in. In order to promote road safety, actually, human society figured out that traffic lights are important. We figured out that speed limits are important. But you do not need the same speed limits on highways versus country roads. And then, we figured out that, okay, if you had an ageing population, vulnerable segments of the population, you can do a Silver Zone – well, that is what we do.

So, my own sense of it is that, in AI, the equivalent of the seat belts, the air bags, the equivalent of the traffic lights, the speed limits, all these things will have to be built up.

And back to the Member's question then about whether there will be a central agency that can be effective in regulating AI, it remains to be seen because we have to understand what the useful regulatory measures are in the first place. Keep in mind that we cannot operate only on the basis of domestic interests. All of these regulations have to inter-operate with our counterparts overseas. Otherwise, our businesses will find it very difficult to meet the requirements when they operate in different jurisdictions. So, international standards, what will be useful, these things will be developed. We will see the emergence of new standards, new certifications and new regulatory frameworks.

The approach that we take in Singapore is to try our best to be plugged into as many of these conversations as possible. We will not be the only ones thinking about governance. We are not the only ones. The more we share with our international counterparts, the more we are able to look over the horizon and take steps to raise our own regulatory measures in accordance.

The way in which we have introduced regulations for online media, for example, how do we look after safety issues, how do we tackle criminal harms that are being carried out online – and yesterday's new legislation is a good example of this. When it has become clearer what is necessary and useful, I think Singapore's advantage is in being able to move quite fast and that is the advantage that we should try and sustain over time.