Audrey Tang

And this also ties into the Buckminster Fuller quote, I didn’t invent the quote (laughter), about building a new system that makes the old one obsolete, because you’re absolutely correct that there were, just as in Taiwan when we occupy the parliament, there were powers that be that really didn’t like this new form of deliberation. But we didn’t quite convince them, though. We worked with the mayoral candidates, just like with 15-M and other large scale movements. And basically all the mayoral candidate that didn’t supply their support to open government principles lost the election at the end of 2014.

And so there’s always a outside game, which is like in mayoral elections, in referenda or whatever. If people don’t play by the open democracy rules, then, because this is a social sector norm that people already felt, well, if you have participated in a real occupy, you know what I’m talking about. They get transformed from within so that they will actively bring down any candidate that didn’t continue pushing forward the open democracy idea.

So just today, this morning, in the parliament and all the major four parties just threw their weight behind the open parliament national action plan. So each party competes on being more open to others on open democracy because after 2014, they understand if they advocate a platform that is stuck in the old slow bit rate, like dial-up speed of democracy, well, they will get no votes whatsoever. So a outside game is always very important. It’s not about convincing. This is about just bringing down people who didn’t support that.

And the mini public to maxi public. I think the theory is a simple theory of really epidemiology, right? The idea was spread if they have a high basic transmission value. It means that people can willingly share it out of altruism, but also out of showing status, out of creativity and things like that. But if we hold the deliberative quality bar to a very fine consensus, it’s almost impossible. It’s definitely impossible on the internet. If on the internet you want a very fine consensus, by definition, people with too much time on their hand win the arguments and everybody else gets burnt out. Even in face-to-face, it’s a very exhausting process.

But if you’re only aiming for rough consensus, meaning are there something that we can all live with, then you actually get to that point very quickly, either online or offline. And based on that, it creates this how might we questions that analyze this wicked problem, problem requiring coordinated action in a way that people very easily remember.

Like, when we’re doing marriage equality deliberation, we eventually coalesced saying that same-sex couples, when they wed, they wed as individuals with all the same rights and duties, but their families don’t wed, so the old style of familial kinship relationship is not disrupted. We say, “We marry the in — the by-laws, but not the in-laws.” Now, this is very easy to remember, and it’s a rough consensus. It’s definitely not something people sign, right?

But with this meme, each community can bring it back to their maxi public and remix this meme into a message that they understand, just as our YouTube counters, right? They, actually, all the people who vote suspect the other parties of, you know, meddling with the election, but they do listen to their party’s favorite YouTubers. But when all the YouTubers share the same rough consensus about counting elections, then it’s problem solved because people just listen to the YouTuber that is more politically aligned with them. But they already have the same rough consensus, as I mentioned, about the mini publics.

Now, the plural algorithms very rarely work, and not on the first try. When we roll out the mask rationing map, some pharmacies invented this method that people trade in their national health cards to some numbers, take a number system in the morning, and they process and swipe those national health cards during lunch break. And at evening, people will come back with numbers in exchange to a mask and a IC card. Now this, of course, saves queuing time, while a mask rationing map also saves queuing time because you’re not supposed to go to the pharmacy that have run out of mask.

But together, these two social innovations are like Mentos to Coca-Cola. They explode because people who use take a number system in the pharmacies, they don’t reflect on the map anymore, right? On the rationing map, they will look as if they sell nothing until lunch break, where they sell everything (laughter), which really looks like it’s rigged. And so the pharmacists get a lot of angry calls. And one of the nearby pharmacist even said a very large banner A4 paper that said, “Don’t trust the app! Exclamation mark.”

And so (laughter) I took a deep breath and walked in and asked the pharmacy this very important question, “If you are the digital minister, what would you do?” And then they just brainstorm in their own pharmacist group. And they got back to me saying, “What if you just invent a button, pressing a button let us disappear from the map?” And I’m like, “This is a really good idea, but we may take a week or two to change the code.” And they’re like, “You don’t actually need to change the code because we hacked the system.” Just like white hat hacking.

They invented it such a way that on the morning they would usually say, “I have received five hundred masks for rationing.” They discover if they input minus one thousand, then it gets into negative stock and it disappear from the map, because we can’t handle negative numbers. And that hack actually solved the problem. Soon as they run out of the numbers, they just go minus one thousand.

Of course, this is forking the digital service. This is essentially taking it into a very different direction. And now the pressure is on us. So we immediately institutionalize that. So after two, three weeks of iterations, we get into a point where the two social innovations finally work together. This is thanking to the weekly deployment cycle and also this key question, if you are the digital minister, what would you do?

And this also solves the free-riding problem because it’s at this core non-rival. You can have more than one hundred different visualization or co-creation systems. It doesn’t hurt to have another one. So that’s by definition takes it outside of the tragedy of a common case because people can fork, but they don’t have to start from scratch, those distribution methods. And indeed, the six thousand pharmacies, each one, I guess, serve as a sandbox of how to manage this co-creation together, and the best one, of course, just amplifies.

Now, the lack of accountability and sanction of election, I think it’s a core issue. Indeed, we’re piggybacking on existing representative democracy, as I said, the threat of, you know, just not voting in mayors that didn’t support open government. The same for presidents, legislators, and so on. And so I think for the foreseeable future, we’ll still coexist for a while, and the main selling point to that, I understand you already buy my, our ideas (laughter). But the main selling point is that for the representatives, it improved the signal-to-noise ratio. They no longer need to do all the discovery of the, you know, principal agent problem by themselves, solving that wicked problem by themself.

We do have people in the career public service who are very capable people, what we call the participation officers, and there’s one in each and every ministry in Taiwan, all thirty-two of them. The Ministry of Health one actually live with that cute dog. And so that explains the quick memetic generation because they just walk back home and take new photos.

And so for each new case, like how do we open up the mountains to mountaineering and hiking, then of course it’s by first self-selected. It’s the POs talking to people related to hiking. And we do mostly the discovery part, which is by Pol.is. But once we have discovered the common interests, the common values and so on, we enter this collaborative meeting, which is always face-to-face and almost always live streamed. And then we move to a more representative part, where there’s a traditional oversee cycle for budget implementations and things like that.

And because of this double diamond, we don’t confuse the discover and define, which is great for participatory democracy, and this develop and deliver, which is best left for career public service and the system integrators and the legislators. And the representative system can take care of the next diamond. Understanding the first diamond will cost them essentially nothing and save their time getting quality input from the stakeholder that’s involved. So we use the traditional deliberative methods and brainstorming and so on, moving from the challenge to the definition, but the implementation is still using classic methods for the time being.

And so I think this answers also your question about sortition-based civic assemblies. If people are motivated enough, we can substitute this self-selection with civic assemblies. Of course, we will take more time, I guess, maybe one or two week more to bring them up to speed because after all, they’re not self-selecting. They may know they have no hiking experience, for example. But I’m aware that we also have that for the Ministry of Culture, of Economy, and actually the National Palace Museum and the healthcare system, universal healthcare, they were based on civic assembly model.

So we basically augment that with a self-selected network input, but we don’t actually replace that. We augment this existing system of collaborative meetings, civic assemblies, but with the agenda set by crowdsourcing because in the agenda-setting phase, again, it’s non-rival. So I’ll, like, these are very simple, like one line — one-liner answers, and I look forward to explain that more.

So I’ll defer to Hélène on the more theoretical points. I would just, again, in a reverse chronological order (laughter), answer this thing about our relationship to representative democracy. I don’t think about representative democracy that much, because I work on the new system. It may or may not make the old system obsolete.

But we do have a process, as I mentioned, the participation office or the PO process, and we have our monthly vote because our bandwidth to process such open deliberation meetings, including Pol.is and a Join platform, although it has already enabled, like, scaling this out a lot, we still have to survey the stakeholders, still have to do the research, still have to prepare this preparatory material, this informed handbook of deliberation and so on, like all the classic deliberative democracies do. And so because of that, we only have bandwidth in our national team of around 100 participation officer to take a case or two at most every couple of weeks. So on average, our bandwidth — we’re now almost on our 100th case now. So on average, we process around two to three cases per month.

So we vote on it. Each participation officer can every month bring a case that they think need cross-departmental deliberation, or people, after they join a petition on the Join platform and collecting 5,000 signatures or more, also automatically nominate for the participation officer voting. And we ask the participation officers to evaluate on three things, and this is a national principle on processing the collaborative topics and open government collaborative meetings at po.pdis.tw. And the three pillars is that it must have a broad, complex stakeholder groups with diverging views. And also it needs enthusiastic public participation, and also interdepartmental collaboration seems warranted.

And this part is, I think, the most important, because this is a very technical subject, and central bank very good example at that. Then the central bank participation officer would say this is monetary policy and there’s nothing interdepartmental about that. And that will actually just move it back to career public service. We don’t insist that they have to deliberate monetary policy. On the other hand, if they are now evaluating CBDCs, cryptocurrencies and things like that, then that is actually interdepartmental. I can easily think of five ministries related to it, and then that becomes more amenable for this kind of participatory democracy.

Again, if the stakeholders, they don’t have diverging views, then this is a waste of calories of us doing the discovery process. And also if the stakes are simple, and especially if it’s like a zero-sum deal, then we don’t usually use this process, because there’s very small chance that we can actually genuinely discover something that is of common value that people can move forward. And so a lot of the other points could be answered simply by saying we’re just setting the agenda. We’re not doing actually binding decisions. But that’s my, like, first, like, zeroth take, and would really want to hear Hélène’s take on it.

Right after the UberX case, we deliberate the Airbnb case, again in 2015. And Airbnb sent a newsletter email to each and every Taiwanese member saying, “Go to this Pol.is site and support our company platform.” So this empirically actually happened. And there’s two features in the vTaiwan process, and broadly the Pol.is process, that guards against this kind of attack.

One is very simple, in that the area, calculated by K-means clustering and shaped by principal component analysis, is actually not measuring the number of people. If you get 2,000 people voting exactly as I do, then you see an extra zero here on group D, but the area do not change. And it actually changes nothing about the binding agenda, because the binding agenda need to have cross-group supermajority, meaning that you actually — what matters is convincing people of different ideas, of different aisles, of different clusters. And so it doesn’t really pay to mobilize 5,000 people voting exactly the same. That’s the first thing.

And also the second thing is the crowdsourcing is actually a wiki survey. Most ways of capturing astroturfing and so on only works if the bit rate is low, like referenda, just one bit literally. And then it’s easy to distract. But according to our numbers, people who get motivated by Airbnb — lobbied, mobilized by Airbnb — they get to the Pol.is website, they see a bunch of really reasonable ideas about Airbnb, and only a third of them actually agree on all the core things petitioned by Airbnb. And the other two third discover more nuanced, more balanced way of expressing their preference, because the 99 statements are literally 99 dimensions, and they very quickly gravitates toward the dimension that are more dialectic, more nuanced, and not at all the Airbnb position.

And so increasing the bit rate, allowing people after voting for a while saying, “Okay, I don’t like any of these feelings. I will share my own feelings for other people to vote.” Again, just costing them, I guess, a minute, because it’s tweet length, literally tweet length. So however long you compose a tweet, that’s the time constraint (laughter), time burden put on you. And just like tweets being shared motivates people to share even more, more people are motivated to think of more nuanced statements, because they win that way, right? Because the only such statements that convince across the aisles did get into the final agenda that is binding in a multi-stakeholder face-to-face deliberation.

So to recap, the first thing is measuring diversity, the plurality, instead of the numbers, the headcount. The second is that a higher bit rate, a higher dimension for people’s preference to be expressed.

I’m now coming to think that the difference between, you know, the nightlife district of the internet, the antisocial, addictive, toxic, addictive drink, private bouncer part of social media (laughter), the antisocial part, and the digital public infrastructure, the pro-social part like Pol.is, I think is whether those credibility or importance are measured by artificial engagement scores, which is the norm if people want to sell advertisement or surveillance capitalism, or whether they’re actually measured by diversity and plurality.

If they get measured by plurality, it means that people spend time on the platform in order to empathize with each other more. And on the antisocial corner, it’s almost as if people engage to empathize less. (laughter) And that is what enable bad actor to gain money and views by publishing content to exploit the biases. They would change the scoring mechanism. There’s no such incentive anymore.

Yeah, I totally agree. The second diamond, it could be a sortition-based backend, it could be a liquid democracy-based backend, or it could be, you know, traditional voted-in electorates. But as long as we talk about rough consensus and the common how-might-we questions, then it’s compatible to each one of those backends of those second diamond. Yeah.

Yeah, I don’t really have anything after this.

I can do this for hours.

Hmm. I’ll continue this line of thought. I guess in many cases, the Sunflower Movement is a really, really good hard test, because the thing that we deliberated on was whether we deepen our service related ties to the Beijing regime, and this has like 20 different aspects, one of which is whether we allow Beijing-based companies to manufacture our then new 4G deployment telecommunication equipments. That was in 2014, but it’s very on topic now in other places as well for 5G deployments. And this has the nature of first, it’s specific enough that people who have good arguments on all the different sides can contribute.

But also this is of a shape that it’s unclear how traditional representation may work, just as there’s no, you know, teleworkers union, there’s no, (laughter) you know, association of people who set up companies in Cayman Islands and so on, which by the way, are the first two cases that vTaiwan processed. There is no easy way to discover in traditional representative like union association means exactly who are the stakeholders of, you know, PRC manufactured 4G components. But on the other hand, this is a very serious cultural issue as well. It pertains to trust and trustworthiness and so on. And but there’s also economic arguments.

I would say that a broad, like, wicked problem with no traditional representative solutions, but was specific enough that people could actually feel it, like lived-in experiences which allows the sharing of feelings, not jumping to solutions. I think these two criteria together, whatever fits the criteria in any given policy would be my suggestion, and that will actually make our implementations very easily translatable in circumstances, like I mentioned, because then people would just demand whatever existing data, remember the facts before the feelings, whatever existing data about the situation being made publicly addressable with its permalink, say, URLs and so on. My answer is also partly to answer Seb Krier’s highest rated top questions because, without answering questions, we might as well be feeding the people a pre-requirement.

And I would just add, on the transcultural or global scale part, it just occurred to me that, for example, coronavirus is a really good example here because we did run a Pol.is-based, actually five Pol.is-based conversations at cohack.tw, which has participation from seven countries and many different teams. And the Pol.is rough consensus was centered around what’s the acceptable privacy and ethics boundary to the responses to pandemic, including, of course, digital surveillance, but also, you know, making the PPEs, protecting the vulnerable groups, making sure that ICU are at capacity, supporting frontline staff and essential workers.

And before the pandemic, we couldn’t even dream of having a five-country deliberation (laughter) of this kind because people have very different timescales and urgency. But because of pandemic, everybody has the same urgency, and then we can deliberate on how exactly is the norm when a digital tool and what’s the privacy acceptable trade-offs and boundaries so that we can actually innovate into, say, participatory self-surveillance to not be captured into surveillance state or capitalism in our response to the pandemic. So this is a very recent example that kind of transcends this cultural barrier because of a shared urgency.

Yeah. When I said our process agile, it has a specific software development meaning. It doesn’t only mean that it’s fast, but it also means there’s a continuous integration of feedbacks. So when we say we fix the system, we deploy a new version every Thursday, it means that we go through this mini diamond, like people calling 1922, saying that there’s something wrong with mask distribution or the MPs making interpolations and so on.

So we do a approach a fast decision, quote-unquote, lowercase d decision, every week. But then that decision, lowercase d, is open to be forked by the social sector and also any good ideas gets amplified in real time and we schedule it for next week’s deployment and we continue to do that. It’s more than one year now, our central epidemic command center. So this continuous delivery, continuous integration is the backbone of being inclusive while being fast.

In my experience, anything that we can very easily explain to six years olds or I guess ELI5, five years old, is good for facilitation. And I mean it by, for example, Pol.is is a kind of AI, but it’s not deep learning. It’s just principal component analysis. And you can actually explain K-means clustering and principal component analysis quite easily, as opposed to, you know, the latest convolutional or transformer models of deep learning. So that’s already a quick — a easy win.

The other thing is about the modality for engagement. For there’s some people who prefer text-based conversation. There’s some people who prefer a visual, like mind map and so on. There’s people who couldn’t really think through without a back and forth conversation and so on. And machine translation as well. We run the cohack.tw with the machine translation powering Mandarin English bi-directional conversations as well.

But all these functions are very, very simple to explain. They’re very, very simple to correct if there’s any biases. It doesn’t really substitute judgment. So as long as the machine learning is for collaborative learning, assisted collaborative intelligence, I think is a boon. But as long as we spend more time explaining it than the time is saved by automating away, then I don’t think it’s a good idea.

Mm-hmm. I think that’s great as long as it’s democratized. And I don’t mean it as cheap and accessible like the 21st century use of democratization. I mean the old meaning of democratization, meaning citizens’ control, and also co-governance.

Thank you, all.

Thank you. Live long and prosper. Bye.

Maybe it will give them a more objective view…

Is this a kind of international comparison study?

Excellent.