Audrey Tang

Yes.

Yeah. So my main research interest now is to automate as much as possible for a community of people who just have layperson's interest in an agentic system, to steer this agentic system the same way as actually how political leadership works, because we don't know the details of all the expertise.

I know something about machine learning, but I don't know everything about the digital ministry. But I can make spaces where people articulate their goals, sometimes conflicting ones. I'm pretty good at conflict resolution or conflict co-creation. So I don't need to understand the details of any of the expert arguments. I only need to know facilitation. And once it's facilitated so that people can all live with this shared rough consensus, that's our policy. This is how we make multistakeholder policy.

So imagine that each participant is a superintelligence, something that is higher in cognitive bandwidth than human beings, and then human can just play the role of essentially a jury or an arbiter and so on, and say that, okay, based on this back and forth debate or whatever, I say that we now agree on something that is still within the continuous norm of civilization, and you may now proceed on these constraints.

And as the intelligence artificial ones are becoming more and more symbiotic with humans, like cyborgs and so on, I think this is one way out of the alignment problem, that we just incrementally increase the ability for people who are good at facilitation, but not as good as the AIs are in cognitive capabilities to shape this kind of debates or forums or whatever for continuous alignment. So if I have a magic wand, I'll just accelerate that research.

Yes, they debate each other.

They don't have to know the details. Most judges don't.

Yeah, exactly.

Yeah, because the flaws in the argument will be pointed out by the opposing side.

A judge or a jury is to make sure that instead of just the technical conversations, the result of this judgment will meaningfully continue the blended volitions of civilization so far, right?

So instead of having one disruptive case that sounds good and sounds reasonable and sounds rational, that disrupts the societal fabric and forces people to invent 16 new words for aunts and uncles, a constitutional judge will say that, okay, so based on your debate and narrative and so on, here is our interpretation that somewhat changes the societal norm, but in a smooth, continuous fashion that takes into account all the arguments about human rights, equality, and whatever, right, but without going all the way of saying the winner takes all.

And now we need to invent 16 new words for aunts and uncles. So a judge is at this point more of a kind of buffering role, of ensuring the epistemic continuity, of ensuring the plurality is still here, instead of going all the way to singularity, so that the society still makes sense at the end of each, this kind of arbitration, so that it doesn't converge to a point where there is no meaningful diversity after that point.

Yes.

Would then take the agreed-upon common understandings or form of, as well as the points of dispute. One side would say that society needs to be organized this way, and the other would say society needs to be organized in that way in response to the transformative technologies.

Many a time without this kind of structured debate or co-creation, what we often hear from acceleration people is that, for example, somebody who is pro-progress and puts a minimum stake on continued human participation, would say that something that optimizes the utility of this or that function on the progress side is worth the sacrifice of human participation, or meaningful participation, or the decentralization of power or checks and balance, or whatever. Or that another may make the case that if you do not impose the draconian measure, a limitation on compute or whatever, then the risk will outweigh all the progress gains or whatever.

And so you need to implement this particular security measure, and so on. Without this structured co-creation or overlapping consensus, one would expect the site that developed the most persuasive AI systems to basically win an argument and constrain the future toward their particular viewpoint.

But the point I'm making is that we don't need to accept any particular progress-safety trade-offs, or progress-participation trade-offs, or participation-safety trade-offs. We can make a space so that only the ones that take care of those overlapping concerns and also open up more spaces for nuanced overlapping in the future.

So instead of discounting the future, we put a premium on future possibilities. And the judge will always prefer those solutions, those arbitration results, that open up for more possible overlapping between those three different value concerns. That's my main argument.

Yeah. So that's democratic participation, like not leaving people behind. That's progress, which is efficiency, utility, automation, all that. Prosperity, all that. And safety, which is against existential agentic risk or whatever.

So the safety, the progress, and the participation arguments are sometimes posed as making trade-offs. But once you start getting into the trade-off mindset, you're foreclosing possible futures where all three are taken care of simultaneously.

My main argument is that if you do not take care of all three simultaneously, then you do not get to a sustainable future. Your light term becomes narrower and narrower every time there's a stronger breakthrough in artificial intelligence.

Nothing about us without us?

Yes.

Yeah, these are the two main allied lines.

So to not leave people behind is to ensure their continuous safety. So this is the participation-safety edge. And to make a co-creative solution that solves wicked problems that cannot be solved independently. That is the participation-progress edge.

The progress-safety edge is to concentrate power.

Right? The progress-safety edge, really, the industrial-military complex is all about being accountable to each other by reducing the nexus of power, the variety of nexus power, so that two-body systems are more stable than three-body and all that.

And so to concentrate meaningfully as hierarchies or as whatever other topologies is to ensure continued progress and safety, because they can more meaningfully make confidence-building measures toward each other when the number of players is small. But to the explosion of emerging new voices and people who are epistemically involved.

I think the word guardrail is perfect to describe it. So instead of guide rails, which is more participatory, guard rails are there because the powers know better than individual citizens.

I think it's just that the bitrate is too low. When you collect preferences, if you leave everything to a referendum proposition, it's literally one bit per proposition. If it's among eight parties, you don't often get eight, four parties, then it's just two bits per one.

So if you are forced to make your preferences known and aggregated with this not even eight-bit quantization, then you're guaranteed to get something that's very skewed. And which is why increasing the bit rate of democracy is as important as reducing the latency.

We've mostly talked about reducing latency. When we talk about real-time contact tracing, pandemic, infodemic, it's about responding here and now instead of four years later. But equally important is the increased bitrate. So people can propose nuanced co-creations, like repurposing the Taipei Main Station seating, a floor, and so on. Imagine if you're forced to do that only by voting among four different candidates. You cannot express that.

Right. And people would watch on a national broadcast as a journalist asks all the questions in real-time and so on. So it shortens what's usually regulatory speaking is usually a 60-day cycle into 24 hours. So that's reducing latency.

Expressing preferences in a way that is not individualized and that is structured to allow more dimensions.

So like all our ideas of Polis, because you get to type that statement instead of choosing among five survey questions, it automatically increases the dimension every time somebody types a new statement.

Yes. So we're taking all these tools, collective intelligence, and all that, and using them in the public sector. My ministry has 600 full-time staff. Any 30 of them can sit on a petition on a local discourse instance. And we always live stream every month everything that passes, like 30 likes, and if it has 100 likes, and it's a discourse forum. So of course, with more structured and unstructured inputs, I am basically pre-committed to implementing everything that 100 of our 600 people agree on. There's also negative votes, so like net 100 votes and so on. This is not like the top-down industrial power you see in Yes, Minister or other movies. This is an internal collective intelligence system. And we do that now also with Talk to the City. We're just deploying it internally so that you can even just hold book reading clubs or public hearings or multi-stakeholder conversation, focus groups, whatever, and then just throw everything into Talk to the City for that kind of visualization. So I think bureaucracy itself, because the staff, all of them, are very good at multimodal communication. It's part of their entry exam, right? So they're by definition capable of high-bandwidth input into policy conversations. So you just need to install such apparatus in place in order to structurally respond to it every week, every couple of weeks, and finally every month in a live stream.

No, that's for the citizens. So this is alluded to as internal Join. This is what's called MyModa, which is the Moda internal instance of co-creation.

They actually do if it's on the Join platform. One of the registration requirements is to verify your SMS. And this is also because there's participatory budgeting on the Join platform, so you can only participate in your resident municipality or city. And for that, we need your SMS code.

Well, we learned all these from other places. The pro and con two columns with their reply button to each. We learned from Better Reykjavik from Iceland. The PB dashboard and everything design, we learned from Barcelona and Madrid, which is why we keep saying jurisdictions, so that I can just say Barcelona. And then for many of the pioneering uses of discourse, like rough consensus, consent instead of consensus, we learned from Loomio, which is New Zealand. We're just one tentacle in this collective intelligence lab. Recently, Sitra from Finland deployed Polis nationally in their national conversations, so we're also learning from that. They made Polis GDPR compliant, which we're looking forward to merging.

Yeah, well, our T-Road system is an homage to the X-Road system of Estonia. I think Estonia shines because of its commitment to public code. It can afford to do what anywhere else would be called too centralized deployments of electronic voting or whatever, by the virtue of them being public code.

So not just open source in code for programs, but also code of conduct, code of regulations and so on. That shows how the public sector is integrating people's contributions, including bug bounties, of course, but also meaningful extensions and so on to their public code. The X-Road has now evolved into the Nordic Institute, the NIIS.

So that Finland and Iceland, I think, are co-maintaining with Estonia this fundamental system for data exchange, the data fabric layers. So we look to Estonia for this infrastructure of stuff.

Maybe just Scratch. I wouldn't call Scratch computer science.

Yeah. Which is why we insist on competency instead of literacy in education. In the curriculum, starting 2019, the basic education, we change media literacy, digital literacy, data literacy, whatever, to media competence, data competence, and digital competence.

Because literacy is when you act as an individual consumer, but competence is when you work as a community co-producer. So not only it is about creating social impact, but it is also about the mindset of common good. Like, a student would only get high marks in competency for media or digital or data if instead of comprehending the material or whatever, if they translated it in a way that makes a positive social impact.

In fact, one of the most viral petitions we get from the Join platform to ban plastic straws from bubble tea takeouts is a civics class assignment to a senior high school. The civics teacher basically said that there's a Join platform, find something, make it go viral, start a movement, without specifying which or how.

So this is a classic competency curriculum, and then the student, being 16 or 17 years old, gets a lot of empathy from 16 and 17-year-old people who care about sustainable environments and very quickly forced a shift in how bubble tea and straws interact with each other. So I think these kinds of examples let people who are 16 or 17 know that they are actually leaders, agenda setters in democracy.

Instead of not yet mature enough to vote, instead of apprentices, they get to be like, almost mentors to show a possible new direction outside of the existing power balance, that we can actually have a new design that takes care of the decomposability of straws or whatever.

Yes.

Well, one thing is to introduce her to Cofacts.