Thank you so much, Audrey. It has been fabulous to talk with you.

We have to have a discussion beforehand on what consciousness means.

That is an amazing answer. So, instead of defaulting to philosophy, you are saying that democracy should choose when something has reached a level of consciousness.

I want to ask one last question that I ask all of the guests on this show. What would it take for you to be convinced that AI is conscious or self-aware?

And she sent a letter and she said, “Just so you know, I have a new address. Please mail The Atlantic to my home in Taiwan.”

Yeah. I think the most cherished change of address form that we have at The Atlantic came from Madame Chiang Kai-Shek.

How do you think about the risk to Taiwan right now? You help make a country stronger, that is at great risk with China, and you work in an industry that is the linchpin of this conflict because so much of the decision-making around whether China will attack Taiwan is ...

So, you actually weren’t that impressed with DeepSeek?

When DeepSeek came out, how did you react emotionally? And by that I mean it was both a wonderful fulfillment of much of what you’ve been talking about, which is an open source model made widely available. But on the other hand, it was a huge technological achievement from a ...

Do you think it would be better if the US AI industry and the Chinese AI industry were more integrated or do you think it’s better that they’re going in separate paths?

That was a wonderful essay and he said that he wants more tinkerers at the big companies but he also wants the government to fund more tinkerers. He also said we need to make sure that we slow down China’s AI industry so we have time for the tinkerers to ...

So, now, this is dependent on two things. One it’s dependent on the horizontal race being able to keep up and then, two, it’s dependent on whoever controls the garage door behind which the dragon sits deciding that it’s time to call in the tinkerers as opposed to letting the ...

Okay. So, this metaphor is about to go insane, Audrey, but I hope you’ll hang with me. So, big companies, they’ve built this giant dragon. Small tinkerers, they’ve been working on these small dragons, they figured out interoperability, they figured out human verification, they figured out how to stop CSAM, ...

So, this is what you’ve called summoning the Buddha. So, your expectation is that the folks building this massively powerful super intelligence are going to say, “Oh, boy, we’ve built a dragon and the garage door is still down, maybe we shouldn’t open it, let’s go ask some people if ...

Yeah, that’s a terrifying document. But isn’t that why the horizontal race might not matter? If that happens, if the scenario in the AI-2027 doc where you have these all-powerful self-recursion agents that are not aligned with human interests, then who cares what’s going on in the horizontal race? It’s ...

But now, if the big AI companies are able to make recursively self-improving superintelligent models, if big company A creates a model that can then invent the next model and the model after that and then they’re so smart, why is it going to matter if we’ve got six tinkerers ...

All right. So, the vertical race is these big companies with zillions of dollars building these models going towards superintelligence. Horizontal race, smaller, simpler, open source models and a whole bunch of developers like these communities you have in Taiwan and elsewhere working to build good things into it.

It’s what we’re seeing right now and that’s Elon Musk versus Sam Altman versus Dario Amodei versus all the folks who run the biggest AI companies. It’s fascinating, it’s crucial, it’s important, and it’s terrifying. So, that’s the vertical race. Now what’s the horizontal race?

So, you’ve sometimes talked about the vertical AI race and the horizontal. Explain what those two things are and why this concept matters.

[Break]

More with Audrey Tang after the break.

So, the part where I am somehow able to get all of my opinions to weigh in in a complicated conversation without me having to take eight hours in the Zoom room, that’s appealing. Could we get there?

So, in order for AI to really be able to make decisions, we need smaller AI that people can run on their own laptops so they can test and verify that everything was done correctly in the big model. So, Bowling Green came out with this report and there are ...

Let me ask you about AI. So, AI is clearly helpful in this project. Would you ever trust an AI to actually make decisions? To take all the info from Bowling Green and then just say, “You know what? We’re going to do this.” Will we get there? Do you ...

Let me ask you then a bigger question about this. So, my perception is that all of this has made Taiwan’s democracy stronger. Across the world, all the evidence is that technology has made democracy weaker. It has helped authoritarians, it has broken apart democracies. Is the reason it’s worked ...

It sounds like what they did in Bowling Green is very much a stepping stone to what you did in Taiwan.

My favorite suggestion was one person who said Bowling Green should mandate that, in every drive-thru, ketchup is given to every person. Didn’t get a lot of support but it shows just there are pain points that you may not know about, Audrey.

We’ve actually had an experiment along these lines, as far as I know, the first one in America—in Bowling Green, Kentucky with your friend Yasmin Green and Google’s Jigsaw project, where they did a version of this and they went out to Bowling Green and they said, “Here are a ...

So, you can solve an issue that is fairly easy to grok which is, “Hey, they’re deepfakes of Jensen, he’s trying to give you crypto.” What about a really complicated one? Could you set tariff strategy from scratch using this system?

That’s insane. Thomas Jefferson would be very proud. It seems like it might possibly be better than a system where you just have a bunch of highly paid lobbyists pay a bunch of congresspeople and then they decide what they want to do. Maybe?

Okay. So, you do this, you have this great conversation, everybody talks, they get along, they get a bunch of ideas, everybody votes on them, you then see … The ideas are sent back to who? They’re sent back to the government?

Wait, wait, wait. Okay, you have these insane Zoom groups of people randomly selected, you have AI that’s monitoring for quiet people so to make sure that all the men aren’t talking over the women or all the women aren’t over talking the men but whatever it is, more likely ...

Okay. So, that’s bananas. All right, so these 450 randomly selected Taiwanese people, some of whom have gotten home from work, they’re taking care of their kids but they’ve got ideas on how to stop deepfakes. So, they come up with a bunch of ideas, presumably you have AI synthesize ...

This is insane. So, you basically have a 450-person focus group chosen randomly by the Taiwanese government to advise the Taiwanese government through a bunch of Zoom rooms?

Shouldn’t you know that’s fake because Jensen’s already made $100 billion and the guy doesn’t need to sell crypto?

And is the Taiwanese government still using the system to get feedback and to understand what the people want and how to think through decisions?

How did the state implement that? It wasn’t as though you got control of all the private companies and said, “Hey, private companies, do this bridging.” You’re running a state social media system and, normally, when the state runs a product, the product is terrible. Usually, that’s why the private ...

But Audrey, can you explain what you mean by bridging?

But actually it’s a profound algorithmic idea because I think what most people don’t understand is that underneath whatever the system is that makes things go viral is some logic. Does it stir a certain kind of emotion or what is the outcome the AI is pointing at that controls ...

Intentional and a little more human, right? So, through the action, you’re reinforcing the humanity. But that’s you. And it’s not just that you’ve done a good job with social media, it’s that your whole country has done a good job.

No notifications, grayscale, what else?

So, I assume also, while we’re at it, I’ve got to imagine you have no notifications on your phone. Is that correct?

Audrey, even as you say that, I feel like I want to check Twitter. I’m just kidding.

Explain the differences in how social media works in Taiwan versus the way social media works in the United States. We have Twitter, we have Bluesky, we have Threads, we have Instagram, we have all these big platforms controlled by large corporations. And on those corporations, people, when they talk ...

And this is part of the genius of what you’ve done which you’ve used, you and your colleagues, have used all these methods to somehow make people in Taiwan, as I can tell, trust each other more, trust the government more, and fight about stupid stuff less. One of the ...

Had it been a more serious attack, had they actually taken down the electric grid, you couldn’t have used humor to defeat rumor, you would’ve had to have been very severe in your response. Tell me about how you calibrated the decision to use humor as opposed to, say, citizens ...

It’s an interesting situation. The Chinese government hacks, they have some effect, they create this evil Nancy Pelosi that’s taking over Taiwan and your response is humor to defeat rumor.

A lot of interviews with you start in 2014 when you come to power and you help lead a revolution in Taiwan but I want to start in 2022. And so, you’re there, you’re the digital minister, you’ve come to power through this people’s revolution, and you’re helping Taiwan’s democracy ...

I think maybe a good rule of thumb is that, if you make a viral video of two minutes in which you explain that you’ll get AI to summon a Buddha, you might get follow up. So, that’s what we’re going to do here today, Audrey. By the end of ...

We had a good time in Paris and we made a cool little video that, weirdly, a lot of people saw.