All right, how should we proceed? Maybe a round of introductions?

Yeah.

Hm. We’ll start from the Aya?

Okay. I am Aya Miyaguchi, currently the Executive Director of the Ethereum Foundation.

I'm Barry. I do a lot of work with zero-knowledge proofs. And in general, like programmable cryptography. I think it's a very exciting way to make our world more secure and more trustworthy.

Okay, great. I'm Audrey, in charge of cybersecurity, platform economy, and digital participation, which is actually an impossible term. And I am currently doing a lot of work on democratizing AI. And so, I thought that this kind of zero-knowledge machine learning is very much one of the enabling technologies for this kind of paradigm to emerge.

I'm Hsiao-Wei, Wang Hsiao-Wei. I work at the Ethereum Foundation Research Team. I work at the consensus protocol mostly. Yeah, and I'm Taiwanese.

I'm Phini. I work closely with Barry. But based in Taiwan, doing more of the ZK community development. Yeah, so, yeah.

I'm Sheau-Tyng. You can call me S.T. I work with Audrey for a long time. Yeah, and I'm currently a speech writer for Audrey.

Yeah, I'm Mashbean. And I come from the Department of Democracy Network. Yeah, and I'm also several contributors of Taiwan's local DAO. So, yeah. Nice to meet you!

We met in Tokyo.

Yeah.

Ah, I see, I see.

When Joy was organizing this...

Plurality Tokyo, yeah. And as part of that, we're running alignment assemblies to use plural mechanisms to have entire communities help the AI self-align. Language models in particular, without relying on OpenAI or Anthropic.

And Wendy Hsueh here is in charge of the alignment assemblies in Ideathon, the Taiwan arm of Alignment Assemblies.

Great. Speaking of plurality, one of my personal, but also the interest from the Ethereum Foundation is that Ethereum started mainly in European communities, like in Berlin.

And then I don't know, you've been working with Vitalik and others. But also, between Vitalik and myself too, like we've been talking about how to make the community a little bit more diverse.

And we did a lot of support in Latin America, for example, the last couple of years. And I think like we’re interested in supporting more things also in Asia next. And that's one of the reasons why I came over here.

And also, we have some great talent. She (Hsiao-Wei) is, by the way, the top star. We don't have a lot of diversity in the core researchers and developers yet. So, having her has been very inspiring to other people. So that's one of... but also things in the US, like laboratory situations are, you know, not always positive.

There's... we're trying to have a deeper understanding about how this can benefit the public. So, looking for more examples to support. It's always better to be built by local teams and local people, so that is just high level. Yeah.

Yeah. So, I understand that Vitalik is coming here mid-August and we're trying to make him also Taiwanese by getting him a gold card in the digital field finally.

And we changed the rules of the gold card so that eligible individuals can get three years of Taiwanese residency, universal health care, and all that, as long as there's some public proof that they've been working in digital fields for eight years, regardless of diploma or employment or whatever.

And of course, you know, public, almost like a notary of digital contributions is one of the things that Ethereum can provide to local communities so that their local impact can be recognized globally. I think there's a lot of real synergy.

So, there are a couple of topics I'm interested to discuss with Audrey here. So, one of that is, as a foundation, right? So, we care about public goods. So, we are also trying to explore how we can, you know, work together on that or how the Ethereum Foundation can help with that, you know, to develop public goods in Taiwan, especially in Web3.

And we already collaborated.

We do.

We run a quadratic funding round.

Right, right, right.

With the presidential hackathon participants.

Right, exactly.

And that's really helped legitimizing the use of quadratic funding because we've been using quadratic voting for a very long time now. But quadratic funding is relatively new.

And so, thanks to that experiment, we've been scaling up -- and out, and deep -- the quadratic funding in terms of, for example, the public innovation is entirely structured with the three leading crowdfunding sites in Taiwan to participate in quadratic funding.

So, as part of that plan, we work with the three leading crowdfunding sites. And there are others. So that they can form their API to prevent civil attacks and so on. The basic infrastructure needed to get quadratic funding to work and eventually morph into plural funding, which takes care of intersectionality.

So, I think we've got the infrastructure here. So, if you're interested in building on top of that infrastructure, I'm very interested.

Yeah. Do you think there is any other popular thing we can do similar to this one? Especially probably for the government or for the local community, not necessarily tied to Web3. But we can use this type of funding or voting for that.

Yeah. There are a lot of social entrepreneurs here doing similar things, but in different names. So, instead of talking about impact certificates, they will talk about social impact bonds. Instead of retroactive funding and so on, they would call it pay for success. But although the terms differ, ultimately, it's the same thing. So, my main suggestion would be just to start partnering with local impact investment communities.

The SERT, for example, has been working to restructure the Taiwanese company laws and the application of such laws, so as to enable a capped profit or a non-profit owning a subsidiary for-profit, turn it into a with-profit. This makes the company governance align with the kind of governance required for retro funding to work.

So, I think this is important because for retro funding, it's important that people over a long term optimize for the same vision instead of getting a new round of funding. And then instead of exiting to communities, suddenly exits to something else, which would really impact the legitimacy of the whole process if the participants all do some pay switch. So, some kind of value lock-in is important and there's a lot of local people focusing over the past few years on that. And we have a dedicated minister for that kind of social innovation, Minister Lee Yung-Te. They are important resources.

What's your interest and view on things like decentralized ID? What level do you want to get as Taiwan? Because there are a lot of experiments happening in the world.

Yeah. In terms of decentralized ID, we have two very specific use cases.

One is that of gold card holders or people who want to start, I wouldn't say identifying as Taiwanese, but to stand with Taiwan, to express some sort of, not just symbolic, but identity-wise, right?

Understanding of the Taiwanese context. So, instead of traveling to Taiwan and staying here all the time, we would like to extend the Taiwanese residency as something not entirely geographically bound. And I think the IDs are an important part of this, especially if that person is already part of a federated identity, like the European wallet ecosystem or whatever other ecosystem. It's important for Taiwan to bridge that ecosystem. So, this is one use case.

Another use case is we would like also for what we call overseas compatriots, people who are Taiwanese by some sort of accounts, but not entirely connected to the household registration system or passports or any other formal system. But for those people to still work with Taiwan. So, instead of not having been in Taiwan or not having existing connection to Taiwan, and suddenly identifying they already have a social network, a community that connects them to Taiwan.

It was just that it's entirely contextual to that person, like family and so on, but there was no way to formalize and recognize this connection. So, we have an overseas compatriot, again, part of the cabinet, a minister dedicated for that. And we're interested in expanding that use case as well.

How about the privacy side of ID? Because all the different governments are having the same challenge now.

Yeah. So, we've been investing in privacy enhancing technologies.

So, for example, we invested in the translation of what used to be called IRMA, but it's now called Yivi. It's a Netherlands zero-knowledge proof system, with support for range proofs too.

So, it's an identity wallet with not just personal data portability, but also proof properties about yourself in a zero-knowledge way, including age range and things like that in the form of attestations.

So, we've translated that into Traditional Mandarin. We're thinking about the situations in which it makes more sense to use this kind of identity. A classic example, of course, was during the COVID, right? It's important to prove that I'm vaccinated without revealing anything else. But we're looking for new situations that are isomorphic, similar to that configuration. And that's part of why we want to get 100 ideas from public innovators through a quadratic funding round, because our imagination is limited, but the community is limitless.

Yeah. Well, I think, first of all, it's very amazing that I've talked to many different governments, and then they're all thinking about DIDs, but I think no one has started thinking about that level of refugees, issues, and also privacy.

They don't like the censorship resistance and stuff like that. So, that's where I think that research level of thing, we can also... I guess, Barry is leading a lot of research into this. Do you have more questions about that?

Yes. Did you hear about Zuzalu?

Yes, I was part of Zuzalu online, actually.

Oh, really? You were there?

I logged into the Zuzalu.network server and participated in quadratic voting to support Divya and Saffron for the CIP session. And it turns out that I could vote online, without physically being there.

So, I was able to simply connect my Ethereum wallet and started voting. It asked me which camp I identify with, and then just deducted my votes based on the overlaps with my camp.

Was this with Polis, or with...

I think it's an augmented discourse forum with a plugin that allows for plural voting.

Oh, right. Yeah. Okay, cool. Well, at Zuzalu, we have this ZK password that you use to authenticate at events. You can use it to vote online and participate in the vote and stuff. For me, this is a really interesting use case of this technology where you can have a national, it's kind of like the Zuzalu national identity.

Yes.

So, it's a pretty interesting kind of use case and application. The wallet that you translated already; do you know what the proof system is that it uses?

I think if you look for IRMA.

IRMA.

Yeah, I think it's using a protocol pioneered by IBM, so it's somewhat similar to verifiable credentials, but it has its own proof system. There was also a transcript in which I talked with IRMA researchers about the details like zero-knowledge proofs and so on. So, it's all in the transcript.

Okay. And do you see a world where there's a single issuer of credentials like the government, or do you see a lot of different credentials being issued by different people?

I was a contributor on the Zuzalu token paper.

Oh, okay.

When you saw that quote from Dao De Jing, that came from me.

So that's... So just this early morning, Pooja, one of the co-authors, actually the main author of the paper, was talking about decentralizing AI, alignment, and so on, based on this kind of technologies.

So of course, I think AI makes this quite urgent and apparent that it's important that we don't wait for all the governments in the world to settle on one single authentication system, which people mutually recognize.

Something like that's outlined in this Zuzalu token paper, in that there's a multitude, a plurality of people's affiliations experimenting with their own ways of issuing such credentials, and for them to interoperate in a way that is zero-knowledge, that is to say, doesn't compromise the privacy to fit the race into the bottom.

I think this is important, and it's quite clear that only some sovereign nations will be able to issue the verifiable credentials or other DID tokens on that standard. But we're interested in partnering with everyone who reaches that standard, instead of just other sovereign nations that recognize us as a sovereign state, which is exactly 14 of them.

So, I think it's important that we build a democracy network based on how democratic the players are, instead of a bilateral or multilateral sovereignty-only network, in which Taiwan starts from a position of weakness. But here, in a democracy network, we start with a position of strength.

That makes sense. So, in this world where we have many different credentials... I've been thinking about this too, and there's a couple of algorithms that seem really interesting in being able to surface the real richness of the data in such a network. Did you hear about the EigenTrust algorithm?

Yeah, I heard about that, I haven't read it.

It's really cool, it's like, with just a whole bunch of peer-to-peer connections, you're able to calculate people's global trust score. And this seems like really... like this is very researchy at the moment, but it seems like a very exciting idea to be able to, once we have this data set, it's really nice to be able to... you can have all these different ways of thinking about the data inside the set.

Yeah, and we're at a point where we can actually start operationalizing such experiments. For example, I'm a Lithuanian e-resident, so I can already sign contracts and things like that as a European wallet-compatible person.

So even just between me and myself, I can get different trust systems, assign different social connection recognitions and so on. And for these two, the Lithuanian e-resident, Audrey, and the Taiwanese citizen, Audrey, to count as one vote. These are kind of the minimally viable examples.

And we actually are pushing for the digital gold card and so on, precisely because we want to create a condition in which it’s normal for someone to be like three or four different citizens or residents in the world, in which case this kind of algorithm which I just described is not just for specific circumstances, but become an absolute necessity.

There are some ways that we can actually bootstrap these networks without having to have the issuer agree. So, the way the internet works is based on TLS, transport layer security, and this is inherently non-attributable.

It was designed to be the case that when I request a website, I can't make a proof that a website serves me certain data. So, we have this project called TLS Notary that turns this into an attributable thing.

So, I can make a proof that I logged into Google or logged into a bank, and you can use this to export all these different interesting credentials. And this seems like another interesting way to achieve the same thing that you're talking about.

Yeah, exactly. It's what the Taiwanese community calls Web 2.5. It's like the Kukai Wallet, reusing the Google authentication to serve as the entry point to the first wallet.

And you just described that as piggybacking on top of StartTLS or whatever. And I'm sure that it can also use not just a website connection, but also the TLS for SMTP or whatever, the mail connections, which is also opportunistically TLS-able.

So, every email actually has an RSA signature from Google, and it's part of this anti-spam mechanism. So, you can also make a proof that you received the email.

Right, exactly. And so DKIM and so on, these are, of course, just between the mail user agent, the mail transport agent, and the two senders and recipients. But if we want to make, for example, like I signed a legal contract that should be notarized, currently the easiest way is to upload a copy to Adobe, which Mashbean actually just did when we signed a Lithuanian Memorandum of Understanding.

We signed with our local certificates, but we had to use Adobe’s solutions to notarize that. But we would much prefer if Adobe doesn't become a choke point of all Memoranda of Understanding transported over email.

It would be much nicer if everyone agreed that they can choose a notarizing system in a permissionless way so that everyone can actually play the part of an EchoSign server. Because at the end of the day, what they do is not different from what you just described.

So, I see this as a really nice way to have interoperability between Web 2 and Web 3. If you're able to explore emails, specifically the proof of provenance of an email, because an email by itself doesn't really make much sense, but the provenance is what's important. This is the thing that we've been thinking of. On the research side, we've been thinking about that. This is a little bit like future plans.

I think that in the short term, digital identities and the Zuzalu Passport type things could be really interesting, because this is just a really nice API for developers. You can make any website and you can have all of the citizens of Taiwan sign in. And they just give a zero-knowledge proof.

Exactly.

I think that this is really nice.

Yes. And I think this is important as a kind of narrative layer change.

I'll use one example. Because it used to be that I was developing this computer language called Perl 6. But because it's called 6, everybody thought that it's replacing 5.

But what we're actually doing is we're adding a meta object layer, a language translation layer, a lot of adapters, a shared virtual machine, and things like that on top of the Perl 5 semantics. But because we call it Perl 6, it creates a schism of sorts in the community.

A very similar thing used to happen to Ethereum. So, what used to be called 1 versus 2 gets re-branded as the consensus layer on top of the existing execution... if I'm not misquoting the official websites. So instead of positioning Web 3 as a Web 2 upgrade, which is frankly speaking not possible, right, what we're basically describing now is that it's a layer, we can call it a lowercase Web 3 layer or a whatever layer on top of not just Web 2 actually, but also email, which is not Web 1.

It's not even web, it's Web 0. So, whether it's Web 0 like email or Web 1 or Web 2, they can benefit similarly with the research that you've just described by adding this, we can call it a provenance layer or whatever layer on top of it.

Panda.

Merge.

Yes. The memes are the best.

Yeah. It took a while for us to really convince everyone this is not like one, two, three, two, it's more like working together. I think sometimes memes are tricky. People's mindset was, oh, okay, so the first part is going to be eliminated was the image of the general public.

In the Ethereum 2.0 design, we said phase zero, and then there's no phase one. We don't use phase one anymore, but phase zero is still there.

That's very interesting. Commitment to continuous integration.

So, one of my questions is, I don’t really have experience working with the government, especially before this meeting, we discussed about how the Taiwanese government is in a different special position.

So, I guess my question would be, I wanted to know, your point of view, like how would you recommend people work with or help the government use tools or infrastructure like blockchain? What do you think?

So, from the government's point of view, you're a not-for-profit. For all the not-for-profits, we would ask, what are the public goods that you serve? And usually, the kind of public good that you serve determines the kind of primary ministerial contact that you have.

For example, the Ministry of Health and Welfare connects with many for-purpose organizations, and they become even like a natural extension of the ministry because they would have, for example, the Women's Rights Foundation and so on, that are actually partly governed or even governed in majority by the people in the ministry.

So, there is a spectrum in which the ministry is entirely public sector, and there are the non-departmental public bodies, like the National Institute of Cybersecurity, that are almost like public servants, except of course, they have a better salary and so on, like labor law applies. And then, like the Institute for Information Industry, the III, which is owned not in a majority, but still significantly by the government and with participations from the information industry sectors and so on.

And then, extending a little bit more, foundations that work regularly with governments and even like government officials, which participate in many of activities, but without being board members, and then just grassroots organizations that advocate and so on.

So, this is the spectrum. And then there's the other spectrum, where in addition to advocacy and working on your own grants and so on, some for-purpose organizations start making products or services, like the Children Are Us Foundation. It's actually very large and makes tasty cookies and so on.

And then the more services and products it provides, the more connected it's to the market ecosystem. So, the Ministry of Economic Affairs and so on, on this end. So, you can imagine like health and welfare on this end, and economic affairs on this end. And as more products and services you contribute to the market, the more interested that MOEA is in popularizing the idea that for every product and service purchase, there's now also some way to affect social good in a provable way.

So, the MOEA has policies like Buying Power and so on to encourage this kind of for-profit, but also for-purpose social innovators. So, depending on which product and service your strategy places on the spectrum, you have natural allies in this kind of ministerial symbiosis and this kind of ministerial market or I would say just ministry as incubators, ministry as accelerators and so on. So, it all depends on your strategy.

Yeah. But if we're talking about from the government, Taiwanese government, as an insider, right, you work in here. And then do you see there is some sort of angle or some stuff that groups working with these technologies can work with the government closely? I know the DID probably is one of the things we can use zero knowledge to improve the privacy stuff. And do you think there is some other?

Yeah. So of course, every year has different buzzwords.

Yeah.

If you're talking about all of government strategy, then it depends on buzzwords of the year, right? And this year it is definitely AI...

So, let's talk about AI.

Perfect segue.

I was really interested. Sorry.

Yeah, I know. I think we have to talk about privacy. It's the AI, definitely.

Okay. For sure.

Yeah. AI and privacy.

These are super interesting ideas here. So, we're doing a bunch of work with ZK machine learning. And basically, we're able to, I think we're almost able to make a proof that you did a single inference of ChatGPT 2, which with reasonable improvement time or whatever.

And so, okay, let me give you the vision for this. Okay, so one of the visions is that you can make a machine learning model that checks if you have a picture of a valid passport. And basically, you would give the model to people and people would make a zero-knowledge proof that like, oh, for this model, I gave this input and I got this output that says I have a passport.

And then you can go one step further and say that the picture in the passport matches this other picture of me. And you're able to build these more attestations about different ways that, yeah, you can have machine learning models that people are able to execute privately. And this is layer zero.

Because I've already got it running on my...

Yeah, exactly.

The GGML inventor, Georgi, just launched a new startup. And one of the early GitHub people arranged the initial round of investment. And one of their first work, which actually is running here, is in airplane mode. I was able to ask, why is Ethereum important in the future of zero-knowledge machine learning? And this answer here, as you can see, is almost GPT-4 level...

Oh, really? Cool!

Yes.

And you use LLaMA for chat?

Yeah, I use llama.cpp.

Okay.

Yeah, initially it was CPU only, but now it's GPU accelerated. So, as you can see here, I'm not connected to the internet. And it's actually very performant now. As you can see, this is like five tokens per second.

Okay, nice.

Yes. So, like entirely locally, on par with ChatGPT, and can be tuned overnight. Let me just output this transcript to here, and it has a training mode. So, if you run this overnight, what you're saying basically is that people would like to infer in a private setting, but also share the model.

And currently, the plan, as I understand, is for people to train such models as incremental LoRAs, low-rank adapters, so that, for example, reusing a foundation model will be able to have this kind of passport recognition layers and so on, as a LoRA on top of a shared, it's like stable diffusion or whatever, model.

So, for language, for image, or for multi-modality, and for things like that, I think we need to make state-of-the-art, like reaching ChatGPT level models in open source that can be run in a reasonable speed on networks.

And so, I think the layer one prerequisites have already been made. So now, we need to talk about the ways to share such models in an interoperable way, because currently, the open-source community around huggingface and so on, like zero-knowledge or privacy-preserving sharing is not one of the priorities.

So, we need to talk about the strategy to get people excited about the privacy layer, now that the underlying capability now matches ChatGPT. Because if the capability doesn't match, then it doesn't help, right? Because you'll be comparing passport images, and it gives a lot of false positives or false negatives, and that wouldn't work. But now, that we're at a point where it's state-of-the-art, I think this is no longer research, and just begin to productionalize.

So, I think that we're going to live in a world where there's lots of different AIs, that I'm going to have a personal assistant AI that's going to tell my AI, oh, I want to be happier.

Exactly.

And the AI is going to make a plan for me to be happier. But that plan is going to involve me working with other people. Like the AI is going to maybe figure out that, oh, Barry's going to be happy if he's involved in an open-source project with people in the local community.

And then the AI is going to reach out to other people and find other people who have similar models or similar situations. But the AIs are going to be mutually distrusting, right? Because they don't trust each other, and they don't even trust the humans who are running them.

So, they're going to have all these ZK proofs. That's like, oh, Barry went to university for a couple of years, here's a proof. And he did this, and he's going to be a good contributor, I swear. And there's going to be all of these really interesting interactions to build trust between these models. And I think that's another place where we can have a lot of fun.

Yes. So, a social, interoperable provenance layer, not just for documents, but for contribution histories and everything.

Right. For that, but also for community relationships and things. I expect ZKPs to become this way of interacting, that they become this language that AIs speak.

The other thing or ways that I think AIs can communicate very effectively is that AIs think as probabilities. They're very good at saying, oh, this is probably going to happen. And I would really like to see some experiments with AIs and prediction markets.

I think that there's a lot of space here for training and improving AIs, and also for just competition between different people to be like, my AI is the best. You know this concept of a centaur?

With a lot of things, a centaur is a human working with an AI. They become super effective. I think that prediction markets and things are a natural way to develop these new centaurs.

Yeah, definitely. It gives the signals that can actually be just applied directly to AI without a central curator of sorts to interpret it. And that's the main insight behind Alignment Assemblies.

OK, so let's talk about... Sorry, I'm talking to so much.

You're getting excited? [laughter]

There's some other programmable cryptography that's going to be really interesting for AIs. The next one is multiparty computation. At the moment, I can train a model. You can only make proofs about things that you know. You can't make proofs about shared secrets.

So, we can use multiparty computation to train a model on both of our private data. So, I can contribute some of my data, and you can contribute some of yours. I don't get to see your data, but we end up with a model that's trained on both of those data. This is really cool, and it's interactive. The next step is to use fully homomorphic encryption. There are non-interactive ways of training and inferring them.

Yes, we've been working on federated learning for quite a while now. So, if you have a bounded, many clinics and hospitals agree on the same data formats, then federated learning, including split learning of sci-fi tuning and so on, is not research.

It's been implemented in Taiwan. But as you said, it requires pre-commitment and constant interaction between the parties to make it a reality. There's no way to join asynchronously or in a permissionless fashion. But fully homomorphic encryption might just be the ticket.

There was a presentation I received a couple of weeks ago about a hardware accelerator, an ASIC accelerator, from a startup called Chain Reaction. They promise to make FHE only 10 times slower instead of 10,000 or 100,000 times slower.

And so, like general purpose FHE for everyone is one of those holy grails, right? It's like a general-purpose GPU moment. And if they do go to production... Their proof is that they've made one of the fastest Bitcoin hashing chips, and that's how they fund themselves. Anyway...

It has some interesting moral repercussions. But anyway... the point I'm making is that I think the only bottleneck now is really just the degree of hardware investments that people have to do to get FHE working.

I don't think there are any remaining software or ecosystem bottlenecks. It's currently just really slow. And so, when it becomes fast, maybe in one year or two years down the road, I think we need to plan for that future now. Just assume that it will happen — FHE will become the norm. And then, what can we invest now to prepare the social layers and the public spaces for that future?

Like standardization of databases.

Exactly.

I think I was thinking about doing a lot of ZK bio problems. Basically, the problem is that researchers don't have access to data. Because it's all private and it should be private. It's like health data. But it turns out that standardizing databases is really hard.

But we could do something with your ZKPs. We could have ZK stats or whatever. And you send, like, I'm a researcher and you're a hospital. You have a lot of data and I have an algorithm I want. I send you the algorithm, you run it with a zero-knowledge proof and you return the proof to me and I have one.

Yeah, it's the open algorithm, like the reverse flow.

Yeah.

Many telecoms in Taiwan are very interested in that.

Cool.

Because that's important for, for example, signal data, right?

Yeah.

With signal data, we've actually used signal data during the pandemic to warn people of the areas where you cannot physically keep physical distance without going three-dimensional. So, the signal data would tell us which areas of tourism or things like that are our danger zones and so on.

And we will send, as part of national apps and alerts and so on, to people. But that needs to be done in an entirely privacy-resilient way because signal data, if it's fine-grained enough, is like surveillance, right? So of course, for research purposes, the telecoms already publish as open data, like way outdated, aggregated to not even villages, like districts, municipalities level. So, there's no way to re-identify anyone.

But this is, as you said, just to familiarize the researchers with the data formats. It's not actually useful for any real-time stuff, unless you're a historian, which is still very useful, right?

But what we would like, the world would like to see is for the telecoms to feel at ease that they're not giving away any private data of their customers, that they're not giving away their trade secrets, but can still attest so that in real-time, this tourism place is overcrowded, but without revealing anything about that tourism place and so on.

So, this is one of those pandemic-resilient things that I've been constantly thinking, like when the next Greek letter comes, right? Where we need to have a much better privacy-preserving layer to do whatever we did during the pandemic, but in a zero-knowledge way. So, I think that's also one of the interesting cases, it is just telecom signal data.

So, the algorithm you want to run is basically like how many people are inside, like one kilometer or whatever.

Exactly, exactly. And that can be actually very privacy-infringing, right? And we want it to be like, even with quantum computers, you cannot re-identify anyone.

And I think ZK is one of the main routes going there. Like I'm heading in this direction, I want to see, if I keep going in this direction, would I hit a point where I cannot physically keep physical distance?

Cool. I would like to talk more with the telecom people about this.

Yeah, that would be interesting, right? So, we can provide some solution or...

Yeah. Yes, we have a team of people in charge of talking to telecoms about this. They're on the same floor, very close to me. It is the Department of Plural Innovations.

Cool.

Yeah, and the Plurality section is like an embedding of plurality, in the Department of Democracy Network. Yeah.

So, I want to tap on the quadratic voting stuff again.

Sure, of course.

Yeah, that's just like my personal interest. So, I don’t know, is there a way that researchers or members of our community could be helpful with the government exploring things like that as well?

We do QV all the time.

I know, I know. Yeah, but I don't know, do you think it's a protocol to involve other departments, government as well? Or like how, I don't know, for the bidding process, right? Like we talk about in Taiwan, and I think it's a system we use for ages, but doesn't mean it's efficient or powerful or useful. So, I'm just wondering, is there a pilot project?

What kind of bidding process?

Oh, like for the government, like a procurement contract.

I see, government procurement. So, what you're saying is that the people who evaluate procurements, they vote on a spreadsheet with quadratic formula?

Something like that. Or I don't know how far we can do, but maybe we can give the power to a certain group, or I don't know, or even, I wouldn't say public, public, public, everyone, but I'm just like, I don't know, is there...

Well, if the participants are bounded, then it's exactly the same as the Presidential Hackathon or the RadicalxChange board meetings, right? And for that, we have spreadsheets. It's good technology. And collaborative spreadsheets that tell you automatically based on quadratic formula, that's already a thing, right?

So, zero-knowledge or Ethereum only enters the picture when it becomes isomorphic to what we just talked about, when permissionless participants want to get into the action.

Right, so I think that when you try and do on a larger scale, the issue of bribery becomes a much bigger problem. And the tool that we're working on is specifically designed to make it impossible to bribe someone, and the way that it does it is roughly, it just makes it impossible to prove how you voted, so then you can't do the trustless voting.

A secret ballot, yeah.

Yeah, yeah. It's a receipt-free ballot. So not only is it secret, it's also that you can, if you want to, show someone how you voted.

Really? Even if they're standing next to you?

Yeah. They can never trust that you're actually talking, because you're always able to simulate, oh yeah, I voted the way you wanted, but you can never really believe that. So, we think that this is something that can, if you want to scale quadratic voting, that this tool is going to be important.

Because with regular voting, bribery isn't really super, it's not really a huge problem, right? Because the trade-off between how many people you need to bribe and the actual effect in the result is quite difficult to coordinate. But with quadratic voting, it's different, right? It makes a lot of sense.

Yeah, it encourages collusion, actually.

Yeah.

Yeah. So, but just to check my understanding, you're saying that if I'm somewhat eligible for quadratic voting, and even if I, in my computer, run a program of this briber's design that does my vote for me, still, I can somehow cancel that to run it only in a simulation?

Yes. So basically, what you do... the way that it works is that you're able to do two things. You're able to vote, or you're able to do two things. You're able to vote, you're able to update your vote, and you're able to update your public key. Right?

And the state, whether you did any of these actions, is a given. Right?

So, if I take the briber's program and I run it, the briber will never know if I actually updated my public key before that. And if I sell my key to someone, it's like, oh, here's my public key, here's my private key, give me some money, they don't know if I've updated it or not.

But if there's a voting period, certainly the briber can specify that with a trusted timestamp service, you can only run over the last couple milliseconds of the voting period.

Right, but they don't know if you've updated your public key before that.

No, no, it's before that.

So, you can update your public key, which means that that vote will be invalid.

I see.

Yeah.

But what if that program also updates my public key?

Yeah, but they don't know if that update is going to be valid because you could have updated your public key before that.

I see. So, the upload public key part is like, it can happen in a different point in time?

Yeah.

Okay. A nice design.

Yeah, yeah, it was the thought that I had.

Is it already used for on-chain governance in any form?

So yes, we do some quadratic voting for hackathons. We've only done very small scale until now.

Okay, so it's not like... because when I participate in the Zuzalu online pods, I don't notice anything like that happening. I don't have a public key that I can update.

Oh yeah, that's a different tool.

Okay. So, well then, this challenge becomes just to get everybody updatable public key pairs.

Like, you have to have updatable public key pairs, but it has to be like, you can't prove to someone that you updated or didn't update your public key. Actually, you can't prove that you didn't update your public key. Like, this is the thing we have to hide.

Hmm, okay. So, if I want to replicate this for the next presidential hackathon or whatever, what should I do?

I think we did it for the last presidential hackathon.

We didn't. We just, you know, like we do, you know, like aside from the presidential hackathon, because like we cannot, you know, like fund it. So, like how we run it is like...

But you had a separate funding round?

Yeah, we had a funding round, so we can fund it.

Was it using this technology? Updatable public key?

Yeah, we used that in the different rounds. Yes, the round we run aside from the presidential hackathon, because we cannot work with the presidential hackathon directly. So, we pull out another round for that and we use that.

Okay, great.

Yeah.

Okay, what was the name of the technology we're talking about?

MACI.

MACI.

Uh-huh, okay.

Yeah. M-A-C-I, minimum anti-collusion infrastructure.

Okay, I see. So maybe it would be good to just fold it into part of the presidential hackathon proper, because that's what we already do. However, I think the quadratic voting round is currently ongoing.

Oh, okay.

So how do we actually migrate a currently ongoing quadratic voting into the system? This is a very interesting problem and we've only got, what, six days? Maybe not, right? Maybe next time.

Yeah, six days. Too short to make things happen.

I know, I know. But as I said, we use quadratic funding and voting semi-regularly now. Voting quite regularly and funding experimentally. And so, I'm quite interested actually in improving the legitimacy of the QV portion. Currently, we only allocate something like 12% of weight to the QV portion in the presidential hackathon, because everybody expects collusions to happen, right?

But ideally, we would like to increase the percentage of the weight vis-a-vis the board, the jury, right? The professional evaluators. I think it's just like a participatory budget.

When it first got invented in Brazil, in Porto Alegre, the city council and so on allocated I think something like 5% or so of the city's budget to this participation mechanism, because they did not fully trust that it would produce useful outcomes.

But as it produced more and more useful outcomes and proved to be quite resistant to bribery, although not resistant to grassroots organization, but this is something we want to have anyway, right? Then it increased gradually, the percentage, until it reaches almost half.

So, I think something needs to happen also on quadratic voting with this kind of technology to improve resistance to cyber-attacks and collusions and so on. And then we can see year after year, the percentage become from 12% or 15%, 16% and so on, until it reaches something like 50%, in which case we're at a point where we can popularize this and say, just use it, right?

To account for half of the score in the public procurement and so on. So, I'm happy to experiment more with this technology.

Yeah, I think before the quadratic voting, I think the identity, digital identity or decentralized identity is a big issue as well, because before you can make your voting publicly or whatever, so there is another challenge we have to solve as well.

But as DID is sort of an ongoing topic right now, so I wouldn't think that would be a problem, but we're happy to sort of...

But it's not a blocking thing, because we just talked about all sorts of Web 2.5 ways to kind of grandparent in your existing, like as my main email identities, client certificates, Google auth or Facebook auth identities or whatever identities for them to produce attestations that are compatible with a full...

So, it's kind of a bootstrapping process. But for the citizen, for the resident, it really doesn't matter. For them, this is exactly like social signing. It's the same workflow. It’s just like signing on my Kukai wallet. It looks like a Google sign window, because it's reusing the Google sign window.

So, I think these are important to understand that we're not saying that everyone needs to migrate to anything. If people already have TW FidO, or if they already have a citizen digital certificate, or even if they have a Lithuanian e-residency card that happens to be European, blockchain compatible, all these are good, because they can produce interoperable attestations.

I think that's the route we're going. So, maybe we will forever be in phase zero... Like there will be no global DID interoperable layer. It would just be a patchwork of compatible layers, and that's still fine.

That's good. Exciting as well.

I'm concerned about the civil resistance. If we do a quadratic funding round and you're able to sign in with a Gmail or something, that could be... You can buy a Gmail account.

Right. You can have several...

Well, I guess that we could use ZK email to be like, oh, if you're a student at the following universities, you can prove to me that you got an email that's like, welcome to Freshers Week.

Right, exactly.

Yeah. And this is a nightmare for UX. But it's possible to do. Basically, the UX is you have to go search your emails, find that email, click the extra details, copy the RSA signage, or copy the whole email digest from one window to another, and then be like, make a...

That's something language models can do. It's well within the level of GPTs.

That's true. That's true.

Yeah. Because the thing that was blocking this kind of integration from happening is because of the asymmetricity between data and information. It used to be that if you have good quality data, it produces information like visualization, analysis, summary, and so on, almost for free, which is why data quality is so important.

But now with language models, the reverse is easily achieved. So, even if you have a bunch of unstructured, scattering websites, whether your student registration data and so on, as long as it's available in some form of information, and you can clearly specify the destination data formats, some combination of language models are going to make that work. So, the info data asymmetry is now becoming symmetric. And that changes a lot of UX assumptions.

That's true. That's true. I have to make an AI ZK wallet.

Something like that... which is why it needs to run on laptops. Because if it runs in the cloud, somebody else's computer, we don't even... [laughter] It's impossible to prove in somebody else's computer situation that it's acting on even the best interest of not just the user, but the operator. It cannot actually prove anything like that.

But if it's an extension of your personal computer, then it just becomes like a normal audit problem, a normal cybersecurity problem. And we know how to solve those problems.

That's really interesting.

As we talk about cybersecurity, as Taiwanese, we notice our government recently focused on this kind of data leaking problem. So, do you think there is a sort of angle like ZK can come in to...

Yeah, of course.

Or like blockchain as well?

You cannot have personal data breaches if you don't collect personal data in the first place.

Exactly.

Or if you collect personal data, but you never send it to anyone else for processing, then that's also safe. So, I think we need to get the ministries and so on holding national level personal data into a mindset that it's never okay to send entire data sets to anyone, even other ministries, even other units in the same ministry for the analysis and treat it like radioactive waste or something, right? Something that's very dangerous.

But when people want to make statistics or analysis or whatever on top of it, then do it the reverse way, the open algorithm way, sending algorithms and the models to the data storage, to the aggregator, and run the results, either sending back results or even just an attestation of such results. So, I think this is the basic configuration.

And we are at a point where the ministries, like the Ministry of Interior, is now taking this stance on household data. So, I think we're at an opportunity window. Because later this year, there will be a body in charge of setting up the personal data protection unit in Taiwan.

It will be a full commission, an independent commission, just like Europe. And then that unit will re-evaluate all the data flows between the ministries and so on, and then come up with a revised personal data protection act for the legislation to consider. Well, for the next legislation to consider next January.

It's unlikely they'll do that this year but at least we'll have a bootstrapped unit. And for that unit to fully understand zero-knowledge is very important, because otherwise, a dilemma will be posed. Like, are we sacrificing ease of access, usability, public good, scientific research, history, crime prevention, contact tracing?

Are we sacrificing all that just to satisfy the new commissioners? Of course not. But if we don't have to sacrifice that, but still satisfy the new commissioners, that will be the best world. So, I think this year in particular is the opportunity window to make public advocacy demos and so on.

So, that the new commissioners come to their work fully understanding the problem of the legacy model as established by our constitutional court in the constitutional court ruling number 13, and also the future. You don't have to make a compromise to say no.

How much these different parts of the government, different ministries are knowledgeable about things like zero-knowledge at this moment?

And I know you are quite a researcher and an understander.

I'm just an engineer here. I guess I do applied research too.

But yeah, I wonder what level of education is already being done?

Yeah, I talk about zero-knowledge in the cabinet meeting, so at least the ministers have heard of this new innovation. So, zero-knowledge, zero trust, privacy-preserving computation. I think all these are cabinet meeting level topics.

And that is why the ministers now, as I mentioned, including the Ministry of Interior, understand that there are other options, which is why they can afford to take the stance that we're not sharing those aggregated data anymore, personal data anymore.

I think one push that we're making this year is to make a clear delineation between what we call 無個資數據, like non-personal data. It's hard to translate back to English. Really not non-personal data. Personal data without a trace of personal data, like truly non-personal data, let's call it this 數據, versus data in general.

So basically, it's taking a different stance. Like data in general, we assume they would be re-identifiable in some way, but only through privacy-enhancing technology like ZK can we produce truly non-personal data that is then free for everybody to use. So instead of confusing these two, personal data and non-personal data, because in the EU, as well as many places, there's this reasonable threshold, right?

If you cross this threshold, it's somewhat non-personal, somewhat pseudonymized, somewhat anonymized, but we want truly anonymized in the sense that no matter how much computation you throw at it, no matter how much context you have about that person, still you will not be able to derive anything from it. So, it's more than the differential privacy epsilon stuff. It's zero as epsilon, basically.

So, we're making this kind of PR push this year, 無個資數據, and then hopefully everybody will just default to this. And when people default to expect this, everything that doesn't deliver this level is seen as, I don't know, scammers.

Sorry. You said the PR push is public, like the news?

Yeah, so you will see our MODA Facebook posts talking about 無個資數據 all the time. You will see 數據公益, 數據 whatever, that takes the new default, that this must not be re-identifiable in any way. And so that's the main PR push.

And when we establish firmly 無個資數據 in everyone's minds, and completely replace the previous terms like 去識別化 de-identification, which is a kind of wiggle word in itself. So, instead of saying de-identification, we just keep insisting on truly non-personal data. And I think that will establish the norms on top of which, that this kind of technology has become essential. It's like one of those zero emission things. That's the general strategy.

So, since you already have used something like this, in conjunction with presidential hackathon project funding as a side event, I think it's important that we also amplify each other's messages. That's something we can concretely do.

Is the name that you, sorry I can't pronounce that non-personal data name, that's created by the government?

無個資數據? I kind of invented it. [laughter]

Yeah, I think that's amazing. I only have the image because you said you can't really translate it into English, but that doesn't really exist in other places as far as I know. And then that is educating people about privacy, data...

You can also call it zero-knowledge data.

I've heard a similar concept called synthetic data where you have a bunch of data and you use that to make more data that's like not any person's data, it's just like kind of fake data, but it's very similar thematically to the original.

Yeah, I guess we can call synthetic data "impossible data" or "beyond data". Tastes the same.

I kind of believe that. Exactly.

Beyond personal data.

I love the naming process. You can name new stuff. Wow, yeah, it seems like we have so many things to do.

Oh, so one last question...

Sure, sure, that's fine.

What is a good proof of concept for this? Because it seems like this year is the year, let's prepare some proof of concept that we can share and show people. Is it like the idea that we have so far is to do statistics and just like given this, what's the probability of this?

And the thought of this is for health data. Given this genetic mutation, what is the probability that I have this disease? Is that enough or is there something more specific that we should do?

What was your target audience?

So, the target audience is the people in the new organization that need to like have as part of their...

Oh, our privacy commissioners.

Yeah, yeah, yeah.

Okay. Well, my two main suggestions. One is to go back to Constitutional Court ruling number 13 and to look at the specific case which was in an enclave, a physical enclave. The researchers doing research, I think it was on medical interventions for some chronic disease.

I forgot the details, but basically, they aggregated all sorts of personal data in an enclave. They cannot bring their USB in and did the research and published papers and so on. And the people who sued them were basically saying, I want to opt out of that research.

But the problem, of course, is that if a lot of people opt out, it affects the research. And then the main contest was that how much exactly would it affect? Basically, people who made the case basically said that it will only do so negligibly, but the researchers said that it may go viral. And if this opt-out goes viral, there's a point beyond which that research is no longer possible.

So, I think it makes sense instead of just doing the gene therapy prediction, the 23andMe case, which is personal, also make a case to the universal healthcare use case because it's very aggregated already. And people already derive a lot of value by this being aggregated. But people trust the National Health Insurance people.

The NHI may trust those researchers, but it's not transitive and it's not transferable. I mean, if I don't trust those researchers, I'm not asking to opt out of NHI itself, but I'm asking to opt out of the research. Right?

But when this happens, is there a way for zero-knowledge technologies to ensure that even if I opt out of revealing my personal data to the researchers, I reveal instead synthetic data or whatever, truly non-personal data, so that their research continues to carry on even if 51% of citizens opt out of sharing?

What we would try and do is make it so that the NHI doesn't share the data with the researchers, but instead the researchers send requests and get back zero-knowledge proofs.

Yeah, certainly, yes. But that requires them to have a workbench of synthetic data.

Oh, yeah. So, we would generate some synthetic data, and we would use this as the testing set. And then for the actual paper or whatever, they would run the same computation on the real data. And they would send the request to the ministry of the NHI, and they would do the actual computation.

Yes. And this is the end result. But you were talking about the MVP, the minimally viable demo, in which the NHI is not going to grant you access to the actual data.

Yeah, that's a challenge.

Yes, not the same level of abstraction, right? So, to zoom out a little bit, you need something that is isomorphic to NHI data, but that someone wants to work with you. Because the NHI currently doesn't want to work with any solution providers anymore, right? They closed off that enclave research venue.

With the constitutional court ruling, they're at a point where they don't want to make any new movements until the new privacy commissioners, as mandated by the constitutional court ruling, tell them that, oh, you can do this thing.

They don’t want to take risks?

Right. Of course, that’s a direct result of constitutional court ruling. So, it's not an incentive thing. It's that if they keep doing that, they will be breaking the law, it would be illegal.

So, the isomorphic case, which is my second point, we already prepared, is the sports data. So, it's also health data, but it's not as sensitive as the medical records and so on. And the III, as I mentioned, the Institute for Information Industry, already have aggregated the data from the sports centers and people in gyms, people running marathons and things like that.

And they're isomorphic to medical data. And there are obviously public goods. For example, people with seeing difficulties, if they want to run, somebody else has to accompany them. And their performance will be affected by the elevation in a way that is not the same versus people who can see, right? And so, with this data, you can improve the elevation, the configuration, or whatever at the road to make it more friendly for people with seeing difficulties to run.

And this is not hypothetical because we've already done that, right? So, there are already ways for privacy-preserving computation to affect the public good with researchers willing to work with this new regime, which does take some adjustments, right? You have to accept a low-quality synthetic data. You have to submit your algorithm and things like that. So, it's not entirely intuitive. But there's already API for that now in the III.

In the III right now? So, they work with a model or like something?

Yeah, of course.

Oh, okay.

If you look for 運動數據公益平台, they are the flagship of 無個資數據. It's called a sports data altruism service.

So, data altruism is also, I think, a good branding for this kind of arrangement, right? We're donating not my personal data, but synthetic personal data for the public good. So, I encourage you to look into this sports data altruism service. And there's a lot of available data already here, standardized API, like everything.

And if you can say that, oh, your technology makes the data quality even better or to make the researchers' life easier or their research more legitimate because there's attestation, notary or whatever, or for people who want to join in this kind of data collaborative, spend less time setting up their environment or whatever, there are many touch points in which that zero-knowledge technologists can help.

And people who participate in this already are volunteering anyway, unlike the NHI, which is a universal service. And so, this is like the testing ground. If you can prove here that this really results in less risk for everybody involved in the risky move, then chances are, privacy commissioners would take note of that and then take whatever we learned this year to the NHI case.

Okay, thank you.

Well, she's mainly... because she's part of this team and representing this project. But also, our interest is also supporting education about it, like promoting, supporting these projects that we let people understand the problems of the existing systems, which the general public don't really understand yet. So, not just promoting new technologies, but why do we have to care about privacy?

Yeah, definitely. In the cybersecurity front, I think even just one year ago, almost no MPs know Zero Trust. But now everybody knows Zero Trust. You see the MPs regularly post about Zero Trust architecture. And so, this is the kind of work we can help. It's just to normalize to mainstream technologies.

So, if we do the same to zero-knowledge, what we did for Zero Trust in the service of data altruism, truly non-personal data and things like that, then chances are a year from now, just like Zero Trust, zero-knowledge will roll the tongue of MPs in their interpolations.

And that becomes the kind of environment with the new privacy commission that these technologies can just become daily life, and they’ll be like ‘Of course, it's normal to expect that.’

Thank you for the talk about many topics today, AI, ZKML, and the privacy and everything. And what would you or the government prioritize everything, every topic? What's the mindset for you in the next five years?

Our priority is to survive for the next five years.

Earlier this year, a fishing vessel flying the PRC flag "accidentally" destroyed the subsea cable between Matsu and Taiwan. A week later, another cargo vessel flying the PRC flag also "accidentally" dropped anchor and kept moving and destroyed the other line. And then, Matsu was without internet. So, we set up new routing. We increased bandwidth for the microwave transmitter.

The Telecom Technology Center did set up non-geostationary satellites in one of the Matsu islands to receive internet from the sky. Of course, eventually the subsea cable is repaired, but it's kind of a dry run of what would happen when another earthquake accidentally destroyed the subsea cables around Taiwan.

It may not be a physical earthquake... It could be a logical earthquake. So, something like that, we expect to be the most likely scenario for any sort of foreign information manipulation and interference.

Because if we continue to have high bandwidth internet connection to the world, currently the FIMI, the Foreign Information Manipulation and Interference, could not reach the critical virality that is required for a crumbling down of democratic societies, even though it's now very easy to generate interactive deepfakes. Apple just made a product based on interactive deepfakes, the Apple Vision Pro, which asks me to synthesize my face, right? Although in a privacy-preserving way.

So, these technologies, although they are now in production, they do not have the critical virality to mutate into a lethal pandemic. But if our internet connection to the world is reduced to some point, at that point, the FIMIs will become endemic to the Taiwanese population.

So, our main work now, in addition to increasing communication resilience, working with satellites and so on, is just to have drill runs of simulated red team attacks and so on for that sort of situation.

So, safety and security for that situation require people becoming aware that everything can be synthesized, interactively deepfaked and so on. And that requires normalizing provenance technology, which we didn't quite talk about this time.

So, in addition to the privacy-preserving arm, which we've talked about in this conversation, there's also this provenance arm; such technologies can be adopted by the people. And if we make the provenance arm strong enough, we can be resilient even when the main connections start to break down.

We also frequently ask ourselves: "Can the Ethereum P2P network survive after World War III?" (when designing the Proof-of-Stake protocol).

Exactly. The internet was designed for that scenario, right? We're just continuing the work of the internet pioneers. So that simulation itself is important. And the Taiwan open-source communities, like in the COSCUP this year, they actually have a workshop to simulate exactly that situation, called "The Design We Open - 網路中斷黑客松".

So, to participate more in these resilience-minded open-source communities, I think it's also important just to get people on the same page. And national defense and surviving for another five years are the kind of public good that doesn’t need convincing.

Every other public good that we talk about, including better healthcare and everything, needs some convincing for people to accept it as a public good, to be able to retroactively fund it, to recognize it as an impact certificate, and so on.

But for a nation to survive another five years, you don't have to convince anyone. Anyone of any party affiliation will tell you, oh, it's a public good. So, this also solves one other problem of spreading the public good idea, is just to anchor on something that is absolutely a public good.

That's probably the reason that is making Taiwan a very unique place. I'm always interested in working with developing countries because they have necessity and that creates more creativity and urgency and creativity. I don't see that in most of the developed countries. But you have these similar... sorry, unique challenges with smart people in the government. That's why...

We are happy to...

Yes. Like she said, we don't normally go out to the government and then ask, hey, how do we work together thing. So, it's that she was more representing her team. But I am personally, as much as I think of Vitalik, in the same queue that admire what you are doing here.

Would you like to get a Taiwan gold card too?

So as to have a stake in the continued success.

Then I will visit with Vitalik.

Excellent.

But no, no, no. Like this is my sincere personal... like especially very... because I'm as one of the Asian people and it's very inspiring to see this happening here. And yeah, and I hope that if there's anything we can support, let us know.

And like for me, it's my interest to see more great examples like how you're working, like educating about open source and privacy. We like to see that more. I mean, I don't have to categorize this as public sector or private sector, but it's still like it's for the public interest.

Yeah, definitely.

Yeah. Thank you. Thank you also for your time.

Great. Thank you.