Josephine Teo: evaluation

Transcript context

…Over-regulation. And nerfing is the Gen Z word for this, of capabilities in order to be safe. And actually that pushes what you call AI creators, some others might call LLM trainers, whatever. There are trade-offs. You cannot have it all. You cannot have safe and cutting edge sometimes, because sometimes cutting edge means unsafe. I don't know what the right answer is, but I will say that my perception is a lot of the Bay Area, San Francisco is on the, let everything be unregulated as possible. Let's explore the frontier. And Europe's approach is like, we're going to have government conferences on the safety of AI, even before creating frontier AI. And Singapore, I think is like in the middle of that. There's a risk. Maybe not. I saw you shake your head. It's a really interesting question. How do you approach AI development? Do you say that there are some ethical principles that should be adhered to? Do you say that there are certain guidelines that should inform the developer's thinking? And we don't have a law in place just yet. We've only introduced very recently a law that has yet to be passed. This is on AI generated content, other synthetic materials that could be used during an election. But that's very specific to an election. It's very specific to election. For the broader base of AI developers and AI model deployers, the way in which we've gone about it is to put in place the principles. We articulate what good AI governance should look like. And then we've decided to take it one step further. We have testing tools, we have frameworks, and we've also tried to say, well, if you go about AI development, what are some of the safety considerations that you should put in place? And then we suggest to AI model developers that they should be transparent. What are the things they ought to be transparent about? For example, your data. How is it sourced? You should also be transparent about the use cases. What do you intend for it to be used for? So there are some of these specific guidelines that we provide. They are, to a large extent, voluntary in nature. But on the other hand, we hope that through this process, there is enough education being done so that on the receiving end, those who are impacted by those models will learn to ask the right questions. And when they ask the right questions of the model developers and the deployers, then that generates a virtual cycle where good questions are being brought to the surface, and there is a certain sense of responsibility to address those questions. I take your point that until you are very clear about the outcomes you want to achieve, putting in place regulations could be counterproductive. And I think we see this in many different sectors. Well, since AI is often talked about as general purpose technology, yes, of course, in another general purpose technology, electricity, in its production, of course, there are regulations around that. You know, how to keep the workers safe in a power plant, for example. But many of the regulations do not attempt to stifle electricity usage to begin with. It says that, well, if you use electricity in this particular manner or in that particular manner, then here are the rules that you have to follow. I believe that that could be true of AI too. It depends on the use cases. If you use it for elections, then okay, we will have a set of rules. But if you're not using it for elections, then actually in Singapore today, go ahead. But of course, if you do harmful things, that's a different story altogether. How do you structure a ministry when the technology moves so quickly? Even if you think about the moratorium that Singapore had on data center build-out that was lifted recently, obviously, you know, that's a forward-looking thing. As you think about what you want to put in place for AI versus what you want to wait out and see, like, how do you make that decision? You know, CEOs have to make the same decision. Should I invest in AI now? Should I follow and see where it goes? What's the thought process and who do you work with?…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.