First of all, would you please introduce yourself?

Hello, I'm Audrey Tang, Taiwan's digital minister.

Tell us about your visit to the UK. What are you hoping to achieve here?

Sure. So we had a great talk with OneWeb, the satellite company. As you know, Taiwan is a country with many islands. And just this year, there were these subsea cables being cut between our islands. So everybody's focusing on low Earth orbit satellites as part of our communication resilience.

I'm also here for the London Tech Week and AI Summit to talk about what we call the Taiwan model, the Alignment Assemblies. We take these AI models and align them using community practices, such as town halls, so that our local people with 20 national languages can align AI to the liking of those language communities, instead of relying on a monoculture.

When you talk about the subsea data cables being cut, that clearly raises the specter of the tensions between China and Taiwan, and I want to come back to them later on in our conversation, if we may.

But on artificial intelligence, it has become a reality largely because of the computer chips developed in Taiwan.

Yes, exactly. Whether it's Nvidia or AMD, it's produced by TSMC.

What do you think AI is going to do to the way we live our lives? How big an impact is it going to have?

I think first and foremost, it enables what I call assistive intelligence, the personalized assistants akin to personal computing, but for all the different activities we do around language.

For example, on Twitter, I posted a picture as I was traveling here, in airplane mode on this MacBook, running a language model — without an internet connection — that is trained on my public speeches and my emails. It can draft emails entirely in my style.

So this kind of hyper personalized assistants, trained and used by not just individuals, but also communities, I think that's going to make it very effective for people to empower the people closest to the pain and convey their viewpoints, to effect more effective communication.

That's fascinating. You're a minister. You're also very techie, as our listeners would have just gathered from that answer. You trust artificial intelligence to write emails on your behalf?

That assistive intelligence is entirely trained by me on my MacBook, not on somebody else's computer. It's using my personal data, granted, but it's not transferred elsewhere.

So what I mean is just like personal computers, the devices are an extension of our daily lives, not something that people push from a monoculture to me.

I often liken it to assistive technology such as eyeglasses — I wear this to enhance my idea of what you're asking about, to enhance communication, but it's not pushing advertisement to my retina.

What are the dangers of artificial intelligence? On the email point, you're a minister, and your AI system surely could go haywire and could give out a directive which your staff would follow, but which you may not actually want your AI system to issue?

Exactly, exactly. So it's AI — the assistant — in the loop. It's not the one hitting the send button.

Do you read it before it goes out?

Of course. So just as a human assistant would do, the transparency, the accountability and so on; it's exactly the same way as we would expect of an assistant or staff.

What do you think the dangers are of artificial intelligence?

You recently put your name to a letter, as did many people in the world of technology, saying that there need to be safeguards in place to ensure that artificial intelligence bluntly doesn't kill off humanity. How likely is that, no matter how small a risk it may be?

I think erosion of trust is the main societal risk scenario I'm worried about. Currently, in order for people to collaborate across jurisdictions, we rely on video calls, we rely on emails, we rely on all sorts of things that are intermediaries between people.

Nowadays, it's easy for me on my MacBook to synthesize my writing style, my voice, my likeness, and so on. But because I'm a public figure, everybody else can do that as well, to me, and so interactive deepfakes become a thing. And we already see in Taiwan, scammers use voice clones to gain confidence from friends and families of a person, because it sounds the same. If you ask questions, it actually acts in the style of that person.

So this kind of thing will make people mistrust those intermediate systems. And if we don't mitigate that erosion of trust, the end result will be that we only trust people in the same room. If that's the case, then international coordination become very difficult. The next pandemic, the climate crisis — all those global issues cannot be solved if we lose the ability to coordinate across jurisdictions.

Well, I should reassure our listener that we are sitting next to each other in a room. I'm not talking to a synthesized voice of Audrey Tang.

You check your sources.

What measures need to be put in place then, to ensure that people do trust artificial intelligence products? Because at the moment, if you use one of the many AI chatbots out there, they do regularly spout complete rubbish. They make things up.

Whether you call it hallucination or confabulation, basically people are abusing the language models that were trained just to model language. This means that you can ask for translation, for rhyming, like translate a rap into a poem or vice versa. But people are abusing it to be like a search engine or an oracle or whatever, which is not what these language models were trained to do.

As I mentioned, when we use them in a narrow AI context, meaning as my assistant in writing emails in translation, we need to recontextualize those general-purpose AIs into narrow AI usage cases. We also need to invest in assurance systems that's just like fireproofing a building — there needs to be inspections, audits, and so on to ensure that for the specific purposes, they don't suffer from the issue of making things up.

So, what needs to happen? You're saying various measures need to be put in place. But artificial intelligence is here today, and is presumably out there doing real harm to some people, albeit on a different, low-level, local scale; there is the risk that it could do harm at a much greater level.

What is the reality? What do you think is the likelihood of a global deal being done, where countries agree on what the limits are, so that technology companies, frankly, don't put us all at risk?

In Taiwan, we already passed laws that put a much stricter penalty for people doing those interactive deep fakes, whether for fraud, for synthetic porn, for meddling with elections, and so on. There's existing laws in place, and we've just amended it.

I think there is a consensus really, among all jurisdictions that there is no pro-social uses of interactive fraud. So it's much more likely that we will see those immediate misuses and abuses, and come to an international standard — not just to detect and report them, but also put penalties on such abuses.

Do you think that Taiwan, as a massive maker of computer chips which have enabled Artificial intelligence, has a particular responsibility to ensure that it sets standards so that Taiwanese companies don't put us at risk?

Definitely. And I think part of Taiwan's main vision is that of personal computing. So just as the Taiwan-made PCs, which were around the time I was born in the '80s, democratized not just access to apps such as spreadsheets and word processing, but also democratized its governance; because people could then just put together those PC-compatibles and experiment with different designs of computers and different interface patterns.

We're seeing something very similar now, what we call edge AI, which means that you get to do the training and the inference of AI models closer to where people are, instead of centralized data centers.

Once it becomes part of every phone and every personal computer, it's much more likely that a society will be able to steer the governance into something that fits what people actually expect of it, instead of the current situation, which is just a few of the largest labs currently being able to fine-tune the AI models at scale, affecting millions.

The global agreements on artificial intelligence would clearly have to involve China; do you think it is likely that there would be a big overarching global agreement between democratic nations and the likes of China? And also Russia, which, of course, at the moment is not on very good terms with much of the West because of the invasion of Ukraine.

If you are taking the open multistakeholder approach, which is more like fireproofing of buildings — the ISO standards, electric shielding and things like that — they tend to be more technical in nature.

One of the good things about these open multistakeholder standards is that they're not exclusive. Once we publish the guidelines, the guardrails if you will, that will align the systems to enable, like audits and joint data collaboratives to train such models and so on, they can be adopted by people who care more about safety than permissionless innovation in all jurisdictions.

So I think, yes, it does make sense for all jurisdictions to invest together in pooling together the resource towards safe and responsible technologies that can align these AI models, and I think this transcends jurisdictions.

Is Taiwan being targeted in any way by China using artificial intelligence, whether it be deep fakes or whatever else?

Of course, in Taiwan, according to V-Dem and other academic sources, we are the topmost — for many years now — jurisdiction for foreign information manipulation and interference (FIMI) meddling.

And so we, of course, have in our basic education curriculum, instead of just media literacy, or data literacy, digital literacy, we swapped the word "literacy" for "competence," so that we understand that against such FIMI, it's much easier to counter this virus of the mind, if people can inoculate themselves by doing fact checking.

It's not the checked facts that inoculate people against polarization; it is the act of going through fact-checking and working like a journalist that inoculates the mind against this kind of polarized, viral outrage. So we've been investing a lot into what we call "competence" over "literacy" in our basic education curriculum.

Are you also kind of battling with a kind of island-wide firewall trying to stop manipulated videos or whatever else coming in onto people's phones or their computers?

In our public sector, we draw a line by saying that these harmful product and services cannot be deployed in public sector. We've had that cybersecurity regulation for four years now.

Just last year, at the height of the cyber attacks, last August, following the US speaker Nancy Pelosi's visit, we discovered that it's not sufficient to just put this line on the public sector communication systems, but also, it's important to safeguard like the advertisement billboards on the front walls of the Taiwan Rail stations, or the convenience stores and so on.

That is to say, even though it is not connected to sensitive network information, its nature as a broadcasting medium makes it a target for FIMI attacks. And so we have adjusted that regulation. That's actually the first official document I signed last August as the Minister of Digital Affairs, to put them into the same cybersecurity protection.

We were talking about Taiwan's computer chip industry earlier on. How is it faring in the face of the tensions between Beijing and Taipei?

As I mentioned in the very beginning, NVIDIA is, of course, driving a renewed interest in the TSMC factories — as is AMD recently. So this AI-related boon is doing very well to the TSMC and its supply chains.

As the digital minister, part of my job is to work with the semiconductor supply chain to ensure cybersecurity, like the zero trust network architecture around international standards such as SEMI E187, designed by Taiwanese cross-sectoral governance, and to apply it to all the supply chain around semiconductor, not just TSMC.

TSMC, the biggest computer chip maker in Taiwan has opened a manufacturing plant in the United States; others are planned there and in other parts of the world. And that is because governments want to ensure that they have a steady, reliable supply of computer chips.

They're bluntly concerned, aren't they, that if China does invade Taiwan, they won't get their computer chips from Taiwan. Surely this attempt to reduce risk in their supply of computer chips undermines, undercuts Taiwan, the island's competitiveness, economically and internationally?

The throughput of the plants currently set up, as you mentioned, across different jurisdictions, may guarantee the core needs of top-end scientific research or military use. Which, of course, as you mentioned, is a concern.

But for the chips we're talking about throughout this interview, general purpose AI chips, the throughput relies absolutely on the Taiwanese plants. So we're not talking about the same things here.

This is more like confidential computing for top military or scientific research uses. And here we're talking about something general-purpose that powers the entire digital transformation across the world.

You're saying these Taiwanese chip factories in other parts of the world don't in any way really hit the business model of Taiwan?

Exactly.

Because there has been what some people have called the silicon shield. This idea that Taiwan is so vital to the modern global economy because of its computer chips, that democracies would be compelled to come to Taiwan's aid, if China did invade.

Is that silicon shield still intact, is it still your kind of trump card, if you like, should Beijing decide to try to take the island by force?

I think Taiwan is indispensable. Not just because of the chips of the TSMC, but also because of the important geopolitical place that we are in, which most of the ships, the vessels and cargo ships pass through.

But more than that, we have seen people around the world, especially after last year, coming to Taiwan's defense. When we were cyber attacked last August, we suffered in the same day 23 times more volume of denial of service attack compared to the previous peak, so absolutely a very large volume. And we have seen people contributing to our defense.

Because our ministry uses a web3 technology called Interplanetary File System (IPFS), it means anyone across the world can donate — a little bit of your hard drive, a little bit of your connectivity — to help back our websites up and help keep us afloat. And we've got the IPFS network, with 200,000 computers across the world and many contributors.

I got a lot of emails from people not just contributing a resource to support, but also saying that they work in this cloud company or that cloud company, and they will do whatever it takes to help defend Taiwan.

So, I think this alliance of like-minded jurisdictions and the people-to-people ties that we have, not just with technology and scientific allies, but also industrial ones, is more than enough to shield Taiwan.

That brings us back to where we started this conversation.

When you visit capitals, like London and elsewhere, are you reassured by the words that you hear? From other ministers, from senior civil servants, about should you need the West, should there be some kind of conflict?

Yes, 100%.

Thank you very much indeed for your time.