First of all, it is true that the Taiwan-produced chips powers pretty much everything from advanced military to scientific to artificial general intelligence, to everything really. It’s one of the most general-purpose technologies imaginable. And because of that, I think people trust Taiwan.

Or the spice, you can call it that.

And once people have that peak experience of building a bridge together, they can afford then to build longer, bigger bridges.

So there must be some venues in your society that enjoys this non-partisan, all-partisan, credible neutrality. So build from there. And starting there, it could be very small, very local level, you then get the kind of space that promotes the bridging experience.

So we — in the g0v movement — intentionally sought support from primary school teachers, from educators of all levels, from the national academy, which is beyond the control of parties or indeed the cabinet, the open-source bulletin board system built by Taiwan University students, so on and so forth.

That’s a great question. Just 10 years ago, in 2014, not only the administration was enjoying only 9.2 percent approval rate, but there was great polarization and discord in Taiwan. The way we overcame it is to treat polarization as a symptom, not a cause. The symptom is the lack ...

The existing institutions are already there on the internet, so just talk to your local open-source civic tech community.

So simply investing in increasing that bit rate — so the citizen can speak their mind and build bridges together — does wonders to make sure that your polity move on from those isolated void vacuum of online meaning, so that they do not get captured by those addictive, persuasive ...

And really, just trust the citizens. The citizens mostly have already figured out the right values, the right steering wheel points of direction, for AIs and our technologies and our investment to go to. It was just the citizens have very small amount of bit rate, essentially just a few ...

We have, for example, Vitalik Buterin, who builds one of the most resilient distributed ledgers in the world, to work with us to secure the security, resilience, and so on. And so this is an international community. Again, you can find many experts in this particular domain in Plurality Institute.

Because in Taiwan, all the tools that we develop are developed in conjunction with international experts. We offer this digital gold card that anyone who have contributed for eight years or more on open source, on internet commons, Wikipedia, and so on, can become Taiwanese for three years and become ...

First of all, build connections to your local open source, civic tech communities. They already have all the tools. It’s just that they need a platform to connect them to the pre-bunking, to the counter information manipulation work, and so on.

Sure. Yeah, if you’re a policymaker…

Yeah, this is a pre-bunking and defense argument. And of course there are efficiency sacrifices, but these are minor.

Live streaming a paper ballot.

Yes, because that’s how people can observe.

And still, we get a result within four hours or so, so it’s not particularly lagging.

And to pre-bunk that, so far there’s no other better technology than ask each of our citizens, if you want to bring your high definition camera, which may just be your phone, and to contribute your part in witnessing the public counting of a paper-only ballot in your nearby station.

Because the information manipulation attack does not seek to support a platform or counter a platform, what it seeks is for people to no longer trust in democratic institutions. So it is the fairness and the secrecy of the election that they most target.

Whatever the accusation was, you can find in that particular counting station three different YouTubers belonging to three different parties, that did have an accurate record of the count.

So by using cutting edge technology, broadband, high-definition video, and things like that, only on the defensive part — that is to say, to guard against election fraud — we entirely pre-bunked the election fraud deepfakes that did appear right after the election. There is no room for it to ...

There’s no electronic tallying, there’s no electronic voting, and YouTubers from all three major parties are practically in every station, with a high definition video cam recording each and every count in their own device.

We in Taiwan have a long tradition of each counter in each counting station always raise above their head each and every paper ballot.

The one take away that we would like to share, and this is controversial, from our January election, is to only use paper ballot.

I’m going to say something quite controversial to a US audience.

Like the one takeaway?

Yes, exactly. I don’t mean it in a left wing, right wing thing. It takes both wings to fly.

So we now live in a world where we really have to decentralize the defensive capabilities just because the existing rephrasers, remixers, are already there in forms of open source, and it’s almost impossible to ban them now, which is why we’re shifting left to the actor layer, which is ...

What Tristan just alluded to, is that you can always take a watermarked output from DALL-E 3 or whatever, and then use a small open source model to rephrase it. And once it’s rephrased, remixed, the watermark is gone and the original maker, OpenAI, doesn’t recognize the remixed picture anymore. ...

But we do not live in such a world.

Or not even watermarks, they just keep a record of all their output it has generated. And then you can always ask the maker, “Is this sentence generated by you?” And if there’s only two or three suppliers, this is a bulletproof way against deepfake attacks because you can always ...

Then the defensive technologies will not be distributed. We will rely on those two or three model makers to implement, say, watermarks. The sentences they generate, the pictures they generate will have those invisible watermarks that they can tell.

Yeah. Suppose that the world has no open source AI models, and there’s only two or three choices, not really choices, of generative AI models to choose from.

So the more a technology looks like surgical-grade masks, the more we should make it open source. And the more a technology looks like something a P3 lab would produce and that have a very high lethality rate, the more we should insist on regulations and also drills, really, audits ...

And this is because medical-grade masks are fundamentally a defensive technology. It protects not only against coronavirus, but also the common flu. It is quite general purpose. And there’s no military use I can think of of that technology.

But if there is a way to make N95 and medical-grade masks at scale, very quickly, with everyday materials in everybody’s homes, that is the kind of technology we should open source. Everybody should set up a mask making plant at home.

In 2020, Taiwan rationed out surgical masks. We have this communication meme that says “don’t touch your face with your unwashed hands, wear a mask.” This meme really worked, because obviously wearing a mask protects your face against your own hand, and it sidesteps all the conversations about N95 or ...

When in doubt, fall back to pandemic analogies.

I can try.

In fact, the open source community is doing just fine with smaller models at just 7 billion or 13 billion parameters, and that’s because the compute power of people’s AI PC or MacBooks are about that level. And so, I do not think having more open source activity there would ...

So I don’t think the two concerns, one is about frontier model capability, and one is about this open research where people get to scratch their own itch, people get to discover vulnerabilities and so on, are at odds.

So when I fine-tune it toward my liking, to be more like me, basically, I’m not really increasing its general capability. I’m just making sure that it can save my time, and to be biased toward where I am, instead of biased against me. It feels less like a cultural ...

I can tune those experts, depending on the private data that I have. Maybe one expert is great at answering emails, maybe one expert is great at doing podcast recordings, and so on, based on the materials that I have that I don’t necessarily share.

Now, in AI development, I think what really works is to take a model that we know is small enough that it will not cause apocalypse, and then just tune it toward some person’s likings. So I’m working with, say, the Mixtral models, which is some quite small models called ...

Actually, the deliberate polling is a kind of horizon scanning, because you can ask everyday citizens what are the threats, the scenarios that they’re anticipating.

So there’s this idea of buffering, of when we hit around one-sixth of the capability threshold that we think will cause harm, we just stop there and do a horizon scanning.

And so comes the idea of red teaming, of finding the vulnerabilities before it’s released, instead of just publishing everything by default. And also comes the culture of setting a buffer, like if it hits this capability, even though we know it’s probably not empowering this kind of agentic or ...

Now, AI, and generative AI by extension, has always been developed in this tradition, in open research. It is not until around GPT-3 do people start to think, “Actually, if we apply this tradition, as the models become more and more powerful, maybe it will first hack democracy, or hack ...

It means that the companies, or any group that produces software, is willing to let outside contributors in, and also contribute out to its ecosystem, so that people share the burden of maintaining the source code together, and also look for security vulnerabilities, look for portability improvements, many small things, ...

Sure. So, open source, when that term first started in the late 90s, means very specific things.