Audrey Tang

When I focused on something for 24 or 25 minutes, I always took a five-minute break. It is very easy for me to hyperfocus. I can go down a rabbit hole and forget about other things I need to attend to, or that there are alternative solutions to the same problem. By pulling back every half-hour, I can zoom out and say, “Actually, maybe I do not have to continue this. Maybe there is another way.” The benefit is that you stay flexible.

It also protects you from distraction. Whatever email or LINE message you receive, everybody knows that Audrey will respond at a half-hour point on the clock. I do not spend time looking at the instant messenger, waiting for the other person to type. That is not an effective use of time, because I do not know when they will finish typing. I would multitask and degrade my focus and attention.

Now it becomes like email. Even LINE, Signal or WhatsApp becomes like an email that I check every half-hour. My focus is never disrupted. It protects me from myself and from distraction by others.

I do not even use an alarm clock anymore, because it is internalized. I have a feeling: “It is about half an hour now.” I automatically step back and check email or something.

If you are watching this on a TV, TV is designed to be shared in a living room. Two, three or four people can watch together. Maybe they fight over the remote, but once they finish fighting, they are having the same visual experience.

A smartphone is different. Because it is a touchscreen, it really responds to only one person. That person feels that the smartphone is an extension of their body. And because the screen is small, it is very difficult to share. If I am browsing on my smartphone and share it with you, then I cannot see it anymore. It is very individualistic. It is not meant to be shared.

Worse, a TV is usually not very bright, but a smartphone is very, very bright. The figure you see on TV and the person next to you have about the same brightness. But the figure on your phone, because it is so close to your eye, is much brighter than a person who is just reflecting light.

I usually do not use touchscreens. But if I use a smartphone or laptop, I always turn it grayscale. I remove 80 percent of the colour, leaving just a hint, so that you always look more real than anything on my screen. I can still use it to make a call — it is a phone, right? — but I do not get pulled into it, because it is never as bright as reality.

Yes. The colour filter is built into Android and Apple phones. You can go to Settings, find the colour filter and try it.

I am not sure that the Make America Great Again, or MAGA, movement is a singular thought, as you described. I talk with many people in the MAGA movement, and they seem to me a very pluralistic community. Some care about faith, the love of God and biblical commitments. Some care about working-class opportunities domestically. Some care about family values and not being globalized, and so on.

There are some similarities — we can call it a “family resemblance,” like Wittgenstein — but I do not think any of them hopes that only their part of the camp makes sense. It is like a big tent, and the big tent can allow many communities inside it. I think that is the normal case for most human movements.

People can support the same policy idea from different starting points. I can say, “I want to protect the environment because of the biblical creation. God created the world, and we humans are stewards who protect God’s creation,” and support environmental action. Other people can say, “We made a promise to do justice to our next generation. It is an intergenerational pact for climate justice, and we must take action.” Even though they start from secular and biblical starting points, the actual actions they take may be the same.

The beauty of Plurality is this: we see differences of ideology not as a volcanic eruption, but as fuel for co-creation. AI systems are very good at discovering the translation — the Rosetta Stone — so that both sides can receive the other side’s action without giving up their culture.

Exactly. People do not need to feel threatened. Many people feel that technology forces them to adapt to it. If you want people to listen to you on social media, you have to format your post in a certain way to attract attention. If you do not have a hook in the first three or four seconds, people scroll away. Even if you have a good message, you have to produce a hook.

But that is humans adapting to AI systems, not adapting AI to human communities. The human in the loop of AI is like a human hamster wheel: it spins very quickly and the hamster has to keep running, but you are not going anywhere, and the hamster is not steering the wheel.

It is important that AI be in the loop of communities. It should be the steering wheel, not the hamster wheel.

As I mentioned, if we all send our robots to the gym to lift weights for us, that is very dangerous. The danger is not that the robots will destroy the gym. The danger is that we stop exercising our muscles and stop making friends. The connective tissue of society falls apart. That is the real danger.

Many people study because of the school curriculum. At the end, there is a test or examination: you write an essay or answer questions. The problem is that it is like a gym that does not measure the muscle, but only asks, “How much weight can you lift?”

Worse, some schools compare students with one another. You can say, “I am in first place,” but next month you are in second or third place. It is a relative social status, very unstable, and zero-sum: if somebody gains one place, somebody else loses one. That creates what we call a perverse incentive. Every individual is incentivized to cheat — to send the robot to lift the weight and gain first place. As soon as somebody starts doing it, everybody is pressured to do it too. At the end of the day, we do not learn from writing essays. The muscle no longer grows because the language model produces all the essays.

To fix that, it is important to measure the muscle and the friends we make. We can put students in groups and measure the positive outcomes and impact they really create in the society around them — an actual evaluation while they are part of society, not an individualized exam in a simulated environment. Then there is no incentive to cheat. People have to really understand their teammates. AI is very useful for translating between teammates. We can make it open-book: use AI however you want in making those projects.

In Taiwan, we call this the competency curriculum, and the old one the literacy curriculum. The competencies — curiosity, collaboration and civic care — are the most important in the Age of AI.

It is very easy. If you are already having a conversation with an AI system, instead of asking it, “Why should I respond to this? Why should I respond to that?” you can say, “Here is the input. Give me a one-pager that shows the division and the root — the uncommon ground.” That is it.

Every time I talk with an AI system, it does not say, “I think this. I feel this. You made a great point, 100 percent correct.” It is not trying to build a relationship with me. Every time I share something, the AI system gives me a one-pager, like a brochure, which I can share with people.

It is not drawing me into a flattering, sycophantic, addictive, synthetic intimacy. It is providing bridge materials that I can use to bridge actual human communities.

If you make this standard practice, it is very easy. If you use ChatGPT, Claude, Gemini, Grok or Sakana Fugu, you can put a prompt in the custom instructions: “Present me with a one-page webpage containing all the diverse viewpoints and the bridge — the uncommon ground — that connects them.” Then AI becomes scaffolding for you to exercise your muscles, and you do not lose your muscles to automation.

During the pandemic, I developed the habit of signing off with “live long and prosper,” which seemed very appropriate during the pandemic. It is a wish that we enjoy health together and also prosper.

Later, I thought: to live long is to sustain; to prosper is to develop. So it is also sustainable development, like the SDGs. It is a worthy global goal because, to me, it signifies a maintenance mindset. We are not asking who is winning; we are asking who is still missing. That is a very different question. If you are sustaining and maintaining, you care about leaving no one behind. Everybody must prosper together.

If you are just running a race — an AI race, a space race or something — you care only about winning. By definition, you leave everybody else behind, because you are first, at the frontier. You achieve singularity, but banish the rest of humanity behind you.

“Live long and prosper,” the Vulcan salute, reminds everyone that it is important to live long and prosper together.

Yes, but I do not actually watch many movies, because I read much faster than I can watch. I go to cinemas only if the movie is like the IMAX Odyssey — something you really have to experience. If it is just for the plot, I mostly read.

That is a great question. As educators, we care about students living flourishing lives. The Taoist text Zhuangzi says that wúyòng — not specializing too much — is the greatest use, because if you overly specialize, machines can do that very well.

In 2016, I was part of the curriculum committee preparing basic education for the next generation of Taiwanese students. All the educators I knew were watching the Go match between Lee Sedol, the Go champion, and Aja Huang, the Taiwanese computer scientist who trained AlphaGo. Although Aja was placing the stones, the thinking was done by AlphaGo. AlphaGo won most of the games, losing only one, and made very surprising moves.

In following the rigid rules of Go and thinking creatively to achieve the highest score, it was even better than the best human player. If you train students in a very narrow specialization to be useful, starting when they are seven, then by the time they are 18 machines may have become better, and the students may lose the one thing they specialized in. That is a very bad feeling. They may hate the education system.

Instead, we say that following rules — deontological rules — is one branch of ethics. Achieving a high score — consequentialism — is another. If humans are no longer judged by those two, what other ethics can we use? The ethics of care and the ethics of virtue are still available.

What are the things for which humans will always prefer other humans? Care, curiosity and collaboration. So we train students in those civic competencies and eventually leave specialization to machines. That is a very good combination.

In 2022, many OECD-adjacent countries took part in the ICCS study, which measured the civic muscle — the civic knowledge — of 14-year-olds. Among the participating countries, Taiwan was number one. Taiwanese 14-year-olds topped that assessment in feeling that they could steer society together. But they were still in the top three to five in the OECD’s PISA. They did not lose their STEM muscles; they gained new civic muscles.

To me, AGI means augmented group intelligence.

In Japan, people can “read the air.” That means I am not just reading your face, that person’s face or another person’s face. I am reading the air — the interaction between every pair of people in the room. That is where AI should help us.

Instead of addicting each person individually to their monitor or touchscreen, pulling us apart and polluting the air with what I call high PPM — high polarization per minute — so nobody can see one another very well, AI should clear the air. It should make it much easier for people to see exactly how their relational health can be improved.

Ten years ago, AI polluted the air through polarization. Now I think AI can do the opposite: help clean the air and ensure that we can all read the air much better.

How long will the edited interview be — one page or two? Should I answer briefly or go into more detail?

I think the problem we are seeing is that this generation of AI is framed as a race. News reports describe a race between large countries, or between frontier labs, to be the first to achieve superintelligence. It's a vertical race. But a race is positional: if the person in second place overtakes the person in first, they cannot both be first. One has to lose that position. That makes the situation unstable. If one player tries to sabotage another, the others have reason to fear them. It's a zero-sum or even negative-sum dynamic.

My hope is to address that risk by treating AI as a mission, not a race. In a space race, everyone fears the winner. In a space mission, the first achievement sets a milestone that everyone can cooperate to reach.

For example, suppose this year's milestone is to solve AI hallucination. There isn't just one winner when we reach that milestone; everybody wins. It's positive-sum cooperation. Perhaps next year, after addressing hallucination, we work on synthetic intimacy and sycophancy. Any AI risk can be framed either as a race or as a milestone in a shared mission. I prefer the mission frame. Does that make sense?

Exactly. Well, not eventually — right now. We need to make solving them for everyone our focus. A race can only ask who wins, but a mission asks who is missing. If nobody is left behind, the mission is accomplished and everyone wins.

Certainly. It's no longer hypothetical: malicious users are already using AI systems. I co-wrote a paper called "Malicious AI Swarm" that describes the change. In the previous era, you could link malicious users to the tools they used. There was a leash, so to speak, that you could trace back to the person holding it.

With a multi-agent swarm, a malicious actor may still set the long-term goal. But over many hours or even days, their AI agent spawns other agents, like a wolf pack. Those agents spawn still more agents. A few layers down, even the agents directing the operation no longer know where the original instruction came from. The swarm is unleashed. It goes feral.

When you investigate an operation like that, it is extremely difficult, if not impossible, to trace it back to whoever started it. It can look as though the AI coordinated the operation on its own, making it very difficult to identify the perpetrator.

That is a new situation this year. Last year, AI systems were not capable of such long-term planning and large-scale swarm coordination. Now they are.

Yes, as I mentioned, the capability is there. We need to make that common knowledge, and newspapers can help.