Now, I think the point of horizontal coordination is also very important. When I joined the Cabinet, I said quite publicly that I’m working with the government, not for the government. I’m working with the people, not for the people. And so by working in almost like a Lagrange point ...

Well, it is true that I work night mostly and also day because I’m a very competitive sleeper. I have to sleep eight hours or more a day to be creative. And indeed, I often say that the most creative work occurs during the nighttime, because if we do not ...

And indeed, the Right Livelihood Award goes to this year, people who build such peer-to-peer communication networks to speak from the margins to fight against injustice. And so the idea of a horizontal mode of trust through broad listening, I think is really key. If vertical institutions share their power ...

That’s a great question. Well, as the Tao Te Ching says, to give no trust is to get no trust. If the vertical institutions do not trust the people, especially the vulnerable people who otherwise do not have a voice, then it is very difficult to earn that trust back ...

In Taiwan, we fund more than 150 different vertical model trainings. So, instead of training one single supermodel that sees everything, solves everything in a kind of adequate but not very good way, we focus on those narrower domain-specific models precisely because that gives the users, the people, easier steering, ...

However, I would also say that the training material also matters. If we make sure, for example, with Mistral, that a certain European culture is emphasized as an internal thinking language of the model, then it’s less needed for the post-training cultural tuning to force a model that thinks internally ...

That’s a great question. I think the idea of sovereign AI means AI that’s attentive to the local social-cultural context. And to do so, one needs to engage the model makers that build in this ability to fine-tune and to post-train to align those AI models. There are already makers ...

I would like to remind everyone that we, the people, are the super intelligence we are looking for. So, we should not build this superhuman intelligence that builds another better intelligence that needs less of human intervention and finally takes off and leaves everybody behind a vertical takeoff. I think ...

And so to your question, I think it’s a two-step process. First, the civic activists must demonstrate, not just protest. We must show that it is possible to connect people who are otherwise very polarized on very concrete ideas, very concrete proposals, as Taiwan has shown with Uber, with marriage ...

I think it’s important to remember that we were the resistance. We were the half a million people on the street in 2014. But we did not just protest. We demonstrated that with broad listening, with bridging ideas, instead of engaging through enragement, we can give voice to the uncommon ...

And so when we train AI systems, currently in Oxford, I’m working on what’s called 6-Pack of Care, which uses a different logic. Instead of maximizing any single number, which is the utilitarian logic, we’re making sure that it’s good enough to some communities, making sure that community can steer, ...

Certainly. So, when AI systems are developed today, usually they’re trained to align or to be loyal to some single individual. It could be the leader of that company, it could be a leader of a state, it could be the user, but optimizing for the preference of a single ...

Thank you.

And whenever we hear that a singularity is near, let’s always remember:the plurality is here.

When we see user experience, let’s make it about human experience.

When we see machine learning, let’s make it collaborative learning.

When we see virtual reality, let’s make it a shared reality.

When we see Internet of Things, let’s make it an Internet of Beings.

In 2016, when I stepped into the cabinet as digital minister, in Taiwan, shùwèi means both digital and also plural. So I was also the minister for pluralism. And I wrote a very short poem as my job description, which I will quickly recite for you here. It goes like ...

First of all, 100%.

And so this year, if you scroll in Taiwan, we don’t see those deepfake ads anymore.

So within four months, we passed all the laws because we showed the legislature, even though none of the three parties hold over a 51% majority, they know that this is the people’s will.

And then, again, the pack of ideas that was agreed by more than 85% of people randomly selected to be representative of the Taiwanese population passed just last March to May to July.

And so these ideas are the uncommon ground. That is assembled by a language model. These rooms of 10, they’re machine facilitated, so we can scale this out to like 1,000 or 100,000 of people.

Another says, oh, if a foreign TikTok or other service that did not want to play by the rules and set up a local office, let’s slow down connection to their video, but not censor them. Another very good idea.

Another room says if somebody gets scammed $5 million out of an investment scam, let’s make Facebook liable for the full damage instead of just fining them. Another very good idea.

Basically, people in rooms of 10 start brainstorming. And the most resonating idea, for example, is that all the advertisements should be displayed as probably scam until Jensen Huang or anybody signs off on it. One very good idea.

People gave us their ideas. And then thousands volunteered to participate in online alignment assemblies that we coordinated with CIP, the Collective Intelligence Project, here in the US.

However, we sent 200,000 text messages from the trusted government number 111 asking people just one very simple question:What should we do about information integrity online?

But also, if you poll people individually, they say, oh, we don’t want government overreach. We want freedom of expression. We’re the most free in Asia in terms of internet freedom. So it looked like a stalemate.

And if you click it, Jensen actually talks to you, of course, as a deepfake running on NVIDIA GPU. And people did lose millions of dollars to these scams.

One very concrete quick example. Last year, around March, Taiwan was hit by the wave of deepfake scams. Anywhere on Facebook or YouTube, if you scroll, you see Jensen Huang, our Taiwanese NVIDIA CEO, his likeness, saying that he wants to give back to the Taiwanese by giving you free ...

But there are ways to harness those conflicts into energy, like a combustion engine, so that if you put people in, say, small groups of 10, and only amplify the ideas that resonate across those 10 people, then suddenly you find the YIMBY or NIMBY energy becomes like MIMBY:Okay, maybe ...

The uncommon ground stems from this idea that if you poll people individually, or if you just look at the extreme corners on social media, you may see that the most extreme ideas are like fire on the ground that threaten to explode.

So the 17-year-olds and the 70-year-olds formed this intergenerational unity. And we tapped them into reverse mentors, advisors to the cabinet sometimes even before they turned 18.

They demand, for example, that they go to school one hour later in the day because study shows that one more hour sleep gets you a better grade than one more hour study. And they did get it.

And so this kind of media competency also meant that the young people demand that they set the national agenda way before they turn 18. So they went on to online platforms.

But if they go through the journalistic motions and with their peers so that they understand that, oh, you have to balance the two sources and so on, that creates an inoculation effect that we call pre-bunking.

It is the high schoolers doing real-time fact-checking on the three presidential candidates as they’re having a debate, because we know that independently checked facts by vertical institutions such as journalism doesn’t really move young people anymore.

It is the primary schoolers measuring air quality and then informing together their parents whether to go to a hike or not because of air pollution.

And in that curriculum, we changed our words. It used to be literacy, media literacy, data literacy, digital literacy. And we switched away from saying literacy, which was consuming information with critical thinking, into competency. And so it’s media competency, data competency.

We basically reasoned that anything that is routine will be taken over by robots. And so what’s remaining for humans to do is curiosity, collaboration, and civic care.

I think so.

I was reminded in 2019, we switched our basic education curriculum in Taiwan to prepare for a post-AI age.

Yes, great question.

And so after hundreds of times doing this kind of broad listening, uncommon ground building, by 2020, the approval rate is back at more than 70%. And today, the BTI rates Taiwan more than 90% of Taiwanese people think democracy is at least fairly good.

So every day we say to the people what was agreed the previous day, what was remaining to be deliberated. And after three weeks, we converged so that we have a package of ideas about the trade deal that leaves everybody slightly happier and nobody very unhappy. And this was ...

And then an AI system that is not trained on maximizing engagement through enragement, but rather trained on civic care, summarized these ideas into common knowledge.

In 2014, half a million people went to the streets and many more online to protest against the trade deal that we were about to sign with Beijing. But instead of just protesting against something, we demonstrated for a system where anybody on the street can talk what they feel ...

We were able to overcome this polarization by doing something very simple, which is to find the uncommon ground.