Moving Beyond ‘Big Tech’ to Save Open Societies TGS 169
Repeating the category errors of some business as usual language, such as saying human resources or incentivizing corporations just propagates this category error in our thinking. So it's like trying to chart out a map, but with like very tilted lens, you can't perceive the world, right? When we see the internet of things, let's make it an internet of beings. When we see virtual reality, let's make it a shared reality. When we see machine learning, let's make it collaborative learning. When we see user experience, let's make it about human experience. Today I am joined by the Taiwanese Digital Ambassador at large Audrey Tang to discuss their work championing the integration of technology and transparency into government functions with the gold to further empower the voice of people in policy decisions. Audrey Tang was the first digital minister of Taiwan from 2016 to 2024,
where they were dedicated to promoting a radical level of government transparency with aims to make all government information, data, and resources as accessible to the public as possible. Today, we discuss a few of their past. Successful projects as well as the philosophy of plurality, which guides all of their work in a global environment where the topics of tech and artificial intelligence can feel esoteric and out of reach for ordinary people. The projects that Audrey has introduced in Taiwan and beyond have resulted in real humans communicating and enacting effective policy changes. Personally, I was naive on this topic and I was blown away by what Audrey and their team were able to accomplish, and I wonder what the world might look like if more communities, more countries, the whole world followed this lead. In my opinion,
this episode highlights the more hopeful side of the great simplification where technology could be used towards more pro-social community, ecologically aware, oriented goals. Additionally. If you are enjoying this podcast, I invite you to subscribe to our substack newsletter where you can read more of the system science underpinning the human predicament, and where my team and I post special announcements and new written Franks and other such snippets related to the great simplification. You can find the link to subscribe in the show description. With that, I am pleased
to welcome Audrey Tang. Audrey Tang, welcome to the program. Hello. Good time everyone. So glad to be here. So you, um, already have quite an amazing resume with lots of successful movements and governance initiatives in your country of Taiwan, especially over the last 10 to 15 years. You became the first Minister of Digital Affairs in Taiwan from 2016 to 2024, and now you are Taiwan's Cyber Ambassador at large. Mm-hmm. Um, but from what I understand, you'd been studying
and working in coding and digital innovation for quite a long time before that. But much of your journey into Taiwanese politics began what was called, um, the Sunflower Movement. Maybe we could start there. Can you tell us a bit about what that movement was? What was your role and experience within it and how it affected your current, uh, worldview and, and work? So back in 2014, the Taiwanese societies deeply polarized. The president at the time was enjoying, uh, 9% of approval, which means that in the country of 24 million anything, the president ma says 20 million people are, you know, not so happy with it.
And so at the time, the parliament was trying to rush through a trade deal with Beijing and using this. Basically, oh, it's inevitable. Uh, the GDP will grow, uh, we'll enter an acceleration phase. Uh, if we don't sign it, other people will sign and then we will lose out, and so on. So forth. Uh, this kind of logic. But then there's, uh, people who deeply, uh, think about the repercussions that it has, not just on our system of telecommunications. For example, Huawei and ZTE will be able to enter and monitor, um, our communications, but also the, uh, impact on environment, our labor on many other things. And so, um, in March at a time, uh, people took matters with their own hands.
So we peacefully occupied the Parliament for three weeks. Now a crucial difference is that we're not, uh, protesters who only demand something like against something we're demonstrators that showed a alternative. And so we developed a lot of tools, like of the half a million people on the street and many more online. You can show up to a citizen assembly like conversation. You can enter your company number,
uh, and then you can very quickly see how exactly does the trade deal affect you. And then you can have a conversation with a dozen other people who are also interested in this matter, to think about ways, uh, to basically regulate, uh, future trade deals of this kind. And so every day we read out, uh, like a plenary what was agreed, uh, that day. And then every day we push it. Uh, a little bit more on the low hanging fruits. Uh, that's, uh, basically under debate. And so after three weeks, we managed to agree on a set of very coherent demands and the Speaker of the Parliament basically say, okay, we'll adopt it, go home.
And so it's a very rare occupy that really converged instead of diverged. And so at the end of that year, I was tapped as a reverse mentor, as a young advisor to the cabinets, basically for each and every incoming polarized, uh, topic. Instead of fighting out on social media, which isolates people into this anti-social corners, we want to make something like the occupied parliament space that we did build that year without.
Literally occupying the Parliament. And so I basically built many digital public squares to tackle things all the way from Uber in 2015 to counter pandemic in 2020, all the way to generative AI and so on in 2023 and 24. And so by 2020 already, the approval rate is back to more than 70% because we systemically discovered the uncommon grounds that can pull people together despite their very polarized ideologies or political affiliations. Um, I have so many questions, Audrey, so, mm-hmm. Uh, I'm, I'm glad you're here today. Let me set the context a little. We, uh, in the world today realize the algorithms and social media and the polarization and the echo chambers, and the inability to really have civic discourse about the things that matter, and we don't even know what's true.
Uh, I am not an expert on that other than I. I am an expert in knowing how it important it, it is to solve these issues if we're going to have, uh, any hope of solving the larger issues that I discuss on this platform. So you just mentioned that instead of protest, you wanted to have alternatives. Mm-hmm. And, um, I'd like you to unpack that a little bit because so much of our, uh, postmodern, uh, critique of the world is just pointing out what's wrong and what's bad, and it's just like an anger sort of thing instead of actually proactive. So can you mm-hmm. Describe why, why that's so important, uh, and, and your experience with offering alternatives. Yeah, definitely. So I'll use one recent example a year ago, um, about, uh, March,
uh, 2024. We saw a problem online, uh, with a lot of deep fake advertisements, uh, running fraudulent, uh, ads that pertains to, you know, sell crypto or sell stocks or so on. In Taiwan is always from Jensen, Juan, uh, you know, the Nvidia guy, uh, the richest Taiwanese, uh, and sometime also from other entrepreneurs. And then, uh, if you click on Jenssen's, uh, likeness, um, he actually talks to you, not just chat, but also, you know, voice and the whole deal. Uh, and that's because
the generative AI has grown to such a point where it can run such persuasion, um, what we call info attacks, uh, with no human supervision. And so, uh, to solve that. We sent SMS text messages to 200,000 random numbers in Taiwan from 1 1 1. That's the trusted number. People know it come from the government asking just one simple question, how do you feel about the information integrity online, what to do about it? And so people gave us, uh, their ideas and then. A thousand, 2000 or so, uh, people, uh, volunteered, uh, to basically have online conversation.
And now at the end, we did not engage all the thousands of people. Uh, we chose 450 people that is a statistical representative of the Taiwanese population in terms of place they live, age, bracket, gender, so on and so forth. And so this microcosm, this mini public, um, deliberated online for almost a day. And the way it works is that people enter, and it's like a Zoom call with nine other people. So 10 people each in each room and the 45 rooms deliberated about the potential responses to this incoming issue of the fake fraud. So maybe one room would say. Okay. Uh, if Jensen did not sign off on that advertisement, it should actually be assumed as skim.
We shouldn't assume human unless proven otherwise. We should assume skim unless, uh, proven by the human. Another roommate say, uh, if Facebook doesn't secure the signature and somebody gets skimmed outta $5 million, then Facebook should be liable for that $5 million because otherwise they would just pay the fine, which is, um, you know, negligible. Uh, and another room says if Facebook also, you know, doesn't even agree on this framework, we should slow down. Connection to the Facebook servers so that the business goes to Google, uh, and so on and so forth. And so all these ideas are facilitated not by human, but by the room itself as a AI facilitator that encourages the quiet people to speak up and make real time transcripts and identify what we call sensemaking, uh, the uncommon ground between those rim.
And then we read it back to everyone and people agreed more than 85% regardless of their party affiliation on the package of measures. And then we check with the stakeholders, the big tech in April, and they really cannot lobby against it because there's no fraud, uh, party. And we can show that everybody agree on that. And then finally, in May, we push out the draft. And it's one of the very rare legislation in Taiwan where all the three parties now of which have a majority, um. Just fast track through. And so now this year, if you open Facebook or YouTube in Taiwan, you just don't see any fraudulent advertisements anymore.
That's a solved problem. And that is because we can show that that was the sense-making result from this broad listening exercise. So this was an anecdote, but you can get the intuition. That is pretty amazing. I actually didn't know that, but let me ask you some questions about that. So you said, uh, you started with 200,000, you got it down to several thousand, and then you chose mm-hmm.
450 mm-hmm. Based on demographics and then they were in 45 rooms of 10. Mm-hmm. Mm-hmm. And, and so there would be, because that itself kind of reflects Dunbar's number of sorts that you have to bring it down mm-hmm. To a manageable Yes. Human interaction level and then scale a little upwards. So did each room of 10 come up with its own kind of verdict? Yes. And then Exactly. And then you compiled those 45 verdicts in a, in sort of a, a way that is exactly the case. And so the, uh, of the 45, uh, 30 rooms,
uh, were from lay people and 15 rooms were from practitioners, like people who are actually media people or social media, uh, professionals. And we made sure. That these, uh, cross, um, pollination, uh, works in the plenary. So, uh, people had one segment of conversation and during the plenary we weave together those questions and suggestions and so on. We read them back, uh, with interpretation by experts, and then we enter the second segment, uh, which then, uh, basically ratify uh, on this, um, plenary, uh, conclusions.
The good thing about AI is that previously you will need a lot of people to like read individually those comments in order to make sense. But now AI can do that without hallucinating, so you can get a pretty grounded report based on those 45 rims, individual verdicts. So what about someone that wasn't part of the 200,000? Um, you said there's 20 some million people in Taiwan and they see the results of this, wouldn't they? Their initial reaction be, oh, this was just some AI scam that put this together. Why, why should I believe what, what ended up being in in legislation? Yeah. Uh, part of the reason why is that we've been doing this for 10 years, and so starting from 2015, uh, during the Uber consultation where again, we just ask people how do you feel about someone with no professional driver license, driving to work, meeting a stranger on an app and charging them for it, um, people already had like more than 100 of those online either petition or the online sortition, uh, or this kind of, uh, conversations.
So people can refer to the prior experiences and they know they can kind of force a response, uh, just by going to the national participation platform and to get 5,000 other people to, uh, basically produce a counter signature, uh, so that for any regulation or for any policy. If they're not happy about the draft that we come up with, uh, if they get 5,000 people, they can force another round of this exchange. How scalable is this? Can't this be applied to almost any issue in, in the world and, and technically, um, maybe not politically, but technically in any country in the world.
Yeah, I think the trigger point, uh, really is that you need a topic that is urgent enough and politically is not the sole purview of an existing department. Mm-hmm. So if it is already a single department, then they tend to feel that they've already got a solution figured out. They do not actually
need the collective intelligence. And if it's not urgent, then it does not warrant this kind of instant sense making technologies you can afford to do that over years and so on. So just a couple weeks ago in California, uh, we launched, uh, engaged California and the first topic to be discussed, uh, is how to recover from the wildfire for Eaton and poly Sade. And that is the kind of topic that has this urgency for clarity and is far from a
single department's purview. And so I do think that for this kind of topics like Calfornia is. 40 million people. It's not a scale thing. It is the, uh, will of the people and the actual urgency for clarity, these two, um, merging together that creates opportunity to launch this sort of platform.
So there's the technology itself, like what it does, but then kind of separate from that is the people's trust in the technology. Mm-hmm. And you said since you did it for 10 years in Taiwan, there was like a social approval mm-hmm. Because people were used to it. What's the threshold beyond which people believe this? Like, could this happen in the United States, um, now on some issue that isn't existential, but, but is interesting to people and, and relevant to their lives? Yeah, I think it's also now, uh, ongoing in Bowling Green, Kentucky, uh, for the, uh, better bowling green, uh, consultation. Uh, and so it's not like urgency, urgency,
but uh, obviously people do feel that there is some value, uh, in closing the loop of the conversations in the neighborhood, the mayor paying attention to it, and then using AI to figure out what's the uncommon grounds, um, despite the differences that people have in the society and how those measures, uh, can really improve people's lives. And closing the loop and telling the people who initially propose those ideas is because these words you wrote, and of course the other 3000, uh, people that this measure was taken, was there any evidence that within the 45 groups of 10 people, each or any other recent example that the 10 people themselves. In the, the process of discussion and debate that was facilitated, facilitated by AI that they learned and changed their mind, or they, they altered their position on the issue? Yes, definitely. Uh, if you look for the, uh, deliberative Democracy lab in Stanford,
uh, which we partner with, uh, for both Engaged California and for this information integrity consultation, uh, they have a lot of research. And the most, uh, important takeaway for me is that this inoculation works in the long term. So not just p do people, uh, entertain. The other side's, uh, visions, uh, in a kind of surprising validator kind of way. So, I, I may not like your politics, but your suggestion makes sense to me. This actually influenced their decisions even like a year after such exposure to a citizen assembly, so that when they vote, uh, they tend to look at the actual measures, uh, the actual issues at hunt.
Uh, instead of just jumping into partisan politics. And the people, the 10 people in each group, did they know that the facilitator was an AI and not a real human? Yeah, because it's not an avatar or anything. You just see, uh, that the transcript, uh, appears, uh, as you speak. You just see a kind of little poke, uh, when you've been too quiet, uh, and so on. So it's not like a AI pretending to be a human facilitator. It's more like this room itself has a facilitating function.
So, so in addition to facilitating, uh, different, uh, priors and, um, ideologies, it also, um. Equalizes in a, in a different way. Because if you get 10 humans together, uh, various power laws ensue and one or two or three of the people are gonna do 80 to 90% of the talking. Mm-hmm. But this actually upregulates the quiet and downregulates the, the chatty.
Yes, that is correct. And the reason why is that we do want the voices, um, that, uh, reach this uncommon ground, uh, to have some way of, um. Amplifying their reach. This is in stark contrast with the antisocial corner of social media where the only most polarized, most extreme, the dunking, uh, that gets amplification because that's a broadcasting network. It's not a conversation network. And so in weaving together a conversation network, we want to upregulate the kind of voice that resonates with the entire room. And to do that,
you probably have to make sure that people, uh, take turns, uh, listen as well as speak. It's really quite impressive, and I am not such a fan of ai, uh, to be blunt, but this, this is one of the good, good sides of ai. Yes. I think that's because it's using AI as assistive intelligence. So
just as the assistive technology you are wearing the eyeglass, uh, it's not replacing a human in the human to human relationship. Rather, it is enhancing the human to human relationship. And this assistive use of AI also respects the dignity of the. People, uh, in a conversation so that they feel they can steer, uh, this conversation, not your eye glass steering, uh, the conversation. And so
I think when we talk about ai, we often think in a kind of automating fashion, like replacing a human in a human to human relationship or reducing humans to machines. But assistive kind of intelligence doesn't do that, is task only and is not trying to be this general, super intelligent that dictates the human's logic. And so it's not about aligning, uh, humanity to. The digital AI logic, it's about the individual digital tools like eyeglasses that can align to the human to human logic. This is very impressive, Audrey, and I know it's, uh, it's approaching midnight in Taiwan. Mm-hmm. And, and your clarity, uh, on this is, is very, um, helpful. Uh,
let me, let me take a step back in, in your history. Mm-hmm. Eventually, your, your work with the Sunflower Movement turned into some other projects, uh, gov Zero and Paul. Mm-hmm. Um, can you give a brief account of what those two projects were and, and specifically how they relate to a concept that you describe as demonstrating rather than protesting. Definitely. So G zero V tw, that's the domain name, uh, was registered before Sunflower in in 2012, um, by some of my friends.
Uh, I joined almost full-time, uh, in 2013. And the way we work is we look at all the government services, like something the Go v tw, and if we don't like it, whether it's budget or something, instead of, uh, just, you know, protesting that it's bad, we actually make a better version as something that G zero V tw. So I talk about the National Participation Platform, join the gov tw, and if you don't like that, you can change your O to A zero and go to join the G zero V tw, uh, which is the gov zero version. But because gov zero is always. Free software, uh, and open culture, meaning that our products, uh, are Forex.
That's to say alternate versions of the government versions. But we also relinquish sufficient amount of copyright so that if a government wants to, they can always merge it back into government service. So quite, uh, famously, uh, during the pandemic, um, the gov zero people developed a alternate way to do contact tracing that does not compromise privacy at all.
So instead of government version, the government simply say, okay, let's use the gov zero version. And that resulted in Taiwan, you know, not locking down any cities, uh, during the three years and actually held for, until Omicron, uh, which is no mean feat. And TSMC just keeps running. Um, anyway, I digress. And so the gov zero, uh, try many different things, but including Polis and Polis was. Before generative ai, before language models for sensemaking, um, you can think of it as a, um, visualization of where people stand on a issue. So for Uber, for example, we ask people to chime in and they go online and they see a fellow citizens, uh, feeling. For example, somebody may feel that, uh, undercutting existing meters is very bad, but search pricing.
During, uh, high demand. That's very good. Uh, so somebody may, may have this statement, you can agree, you can disagree or you can pass, but there is no ance, so no room for truth to grow. And so it is in a synchronous way, simulating a little bit of the 10 people room dynamics by highlighting what's the most resonating idea.
And so you see your avatar being sorted to one room and this room, uh, have these kind of agreements, but you also see across all the different cluster, different rooms, what are the ideas that are currently gaining grounds that everybody, regardless of where they're coming from. Do agree. And so after three weeks in 2015, we agree on the set, a very coherent idea about Uber, which we then pass into law so that, uh, the local co-ops and so on can also operate. And Uber is a legal taxi fleet, uh, in Taiwan for quite some years now. So the idea is to use asynchronous, uh, contribution, uh, and discovery of the uncommon ground so that even if we don't have, uh, the language models, uh, to weave things together, people can still kind of see the community notes that flows to the top.
And the same algorithm has been adopted by YouTube, by meta, and by X as the community knows algorithm. Wow. So, um. Embedded in there, uh, is your emphasis on data about feelings, specifically the feelings of the citizens living under these laws and, and regulations that a government enacts? Mm-hmm. Why is that so important to incorporate, uh, those values into decision making?
And by the way, do you know, um, Nora Bateson and her work in what's called, uh, warm Data Labs? Mm-hmm. Yes, I've heard of, I've not worked directly, but yes. Okay. But go ahead. What, what about, uh, data and feelings? The integration of that? Mm-hmm.
Yeah. Uh, first of all, I think we're all experts of our feelings. Uh, and so that is actually what can easily resonate with our fellow citizens. Had we start our Uber consultation with, what's your ideal economic model for sharing economy versus extractive gig economy? Uh, probably nobody will come, right? Because it was, um, like. Extremely abstract, but feeling is not abstract at all. Feeling is very personal. And so based on feeling, then people want to take care of each other's feelings. So you can see like the Uber driver, the taxi drivers,
the passengers, the people worrying about rural development and someone, um, they all center around shared feelings. And so naturally when people start proposing ideas, those idea that take care of everybody's feelings will float to the top. And so this speaks to a very different ethical, uh, foundation of policy making. This is more about the ethics of care. That is to say, how much do we want to take care of each other instead of what's. Single abstract value, um, like in a scholar value sense, do we want to optimize? Right? Uh, and care also has the benefits of, um, its positive sum.
So if I take care of your ideas, then you are probably going to propose an idea that also take care of my feelings, uh, as opposed to if you put it to referendum or something as Uber did in other jurisdictions, maybe 51% people feel they have won, maybe 49, feel they have lost, but their feelings are hurt and are therefore more likely to engage in negative sum uh, conversations from that point onward. So what did those projects, um, tell you about the divisiveness and polarization of the societies where they were enacted, and did people respond well to, to these technologies? Like, oh, I, this feels more, uh, positive, some and, and caring, and, and did they notice that? Yeah, definitely. Um, so, uh, we can, uh, look at very objective numbers, uh, especially the very young people in 2019. We changed our curriculum. So instead of, uh, the standardized answers, you know,
that the East Asians are very famous about, uh, we switched. To prioritize, uh, the civic competencies, uh, namely, uh, autonomy, that's curiosity, interaction with people who are unlike you, and also the common ground, uh, the ability to construct common good. And so the idea here. Is that if we do not have this shared, uncommon ground in for young people, young people will feel they're very detached from politics. They're just 14, 15. They have no way to contribute to agenda setting, even though they do know, uh, what is actually better, uh, for the planet and people.
But by making sure that the young people have agenda setting power, uh, in setting, for example, e petitions or even becoming, as I mentioned, cabinet level advisors and so on, the Taiwanese 15 year olds, according to ICCS in 2022 are now. Populative world when it comes to the agency. They feel that they can affect the society for people and planet issues, and they still, uh, maintain the number three to number five, uh, PS a score. So people are also happy that their stem isn't actually degrading. It's not the trade off, uh,
but I think the young people's empowerment as well as the depolarizing effect across, um, religious, uh, urban, rural, uh, age brackets, uh, and these, uh, Taiwan is also the least polarized among OECD equivalents A couple years ago. That's amazing and important, uh, because there's two issues. One is using this technology to actually change policy and regulations and, and things. But the other is,
irrespective of that, this technology, um, uh, suppresses apathy and provides agency, which is essential in our current world because there's more and more people with, um, mental illness and, and just checking out because it's so much, because they don't feel they have agency against all the things that are going on. So this technology could be really important just as a, a vector to, um, to increase the feeling of agency. Yes. Yes. And it also has what we call a pre bunking effect because if there's already a polarized fight between the two memes, uh, then trying to arbitrate it, um, like especially from the government, uh, tend to just, uh, kindle the fire even more and people become even more polarized, uh, and fuel conspiracy theories and so on.
Uh, but this kind of technology allows us to discover, uh, the uncommon ground and share it as pre bunking. So one very early, um, example, pre is pre bunking. Yes. So it's not debunking Oh, oh, it's not debunking. It's pre bunking. Yes. Debunking is After something goes viral, you say, oh, that's not quite the case. Uh, pre bunking is that before something goes viral, you already say, by the way, this is actually like this. Right? So, so it's, uh. Many people feel that if they pre bunk each other, they are less likely to be polarized. And there's many ways to pre bunk and humor is one large part of it. So in early 2020, uh, when people are not sure what the coronavirus
interaction with Musk are in Taiwan, we already observe as in other places, like one side says, because we had a SARS experience a few years ago, people feel only N95. The highest gray mask are useful and every other mask are actually, you know, a, a scam or something. Uh, and the other side says it's ventilation is aerosol. So wearing a mask hurts you and wearing N95 hurts you the most, right? So, uh, if we just let these two polarized memes, uh, grow, then they tend to fight each other and people will, um, basically polarized into mask anti-US camps. Uh, but the science, uh, was still not very clear then. So we basically. Pushed
out the meme of a uncommon ground, uh, very quickly, and it's a Shiba inu, a very cute dog, putting her pole to her mouth, saying, wear a mask to remind each other to keep your dirty and wash hand from your face. So, so that's a uncommon ground no matter which part you are. You probably agree that hand washing is good. Uh, we actually measure tap water usage. It actually increased, uh, and because the dog is just so cute, if you laugh at it,
the next time you see somebody wearing a mask or not wearing a mask, you would just think about, uh, you know, hand washing, uh, which is like not polarizing at all. Uh, everybody washes their hands. So just like, there's no pro fraud camp in Taiwan. There's no, you know, anti hand washing camp in Taiwan. And so it just diffused the polarization into just, you know, hand washing. There's also songs about it and the cute dog dancing and things like that. So, um. Yeah, this is like literally, I'm. Soaking this all up because I think it's, it's so important and I take our current social media landscape as a, as a given, and I, I've stopped using Facebook and, uh, I do use the other things to post the content of, of this website, but I'm become really disenchanted with social media and this, this is exciting to learn that these things are possible.
LL let me continue. Ultimately, I believe you've rooted your work, uh, in the idea of plurality. Mm-hmm. Uh, which I think is the name of the book you co-authored with Glen Weil. Mm-hmm. Can you describe what is at the core of, of pl plurality?
Yeah, certainly. So, um, singularity. Means, uh, an AI that can improve itself, uh, increasingly without human uh, control. And at some point the AI can automate everything there is to automate about AI research. And then either, I guess, grow a self preservation instinct and refuse to develop the next generation of AI and kind of see us as competing carbon-based species, or, uh, they don't get that. Uh, and just recursively self-improve and serve not themselves, but maybe, you know, a CEO.
And then the CEO becomes. Transhuman, uh, and then become a very different species than the rest of us, right? So that's singularity. Thank you for that. I've, I've heard that word a lot,
and that was the best description, uh, as horrifying as it is. Um, but thank you, you for that. Please continue. Uh, it is, uh, kind of losing the race, um, of humanity, right? Yeah. It's not a race of, um, ascension, uh, as sometimes portrayed, but for the rest of us, it's just the humanity race loses. And so plurality says, um, instead of making an AI
that's even more powerful by the day recursively, we should, uh, actually enhance the way that people can work across differences. So design each piece of technology. It could be ai, it could be immersive reality, many technologies, um, with this eye on fostering. The differences, but seeing the conflict that ensues not as fire to be put out, but as energy can be harnessed, uh, for co-creation. And so any sort of technology that enhances this collaboration across differences, uh, is in the direction of plurality. So instead of a vertical race, uh, of takeoff, uh, escape velocity, you see a lot of space based, uh, metaphors. Uh, the plurality is entirely horizontal. It is,
uh, a lateral diffusion of technical capabilities. And each capability is steerable by the community that's deploying it. And so the more we invest in plurality, the better we're prepared, uh, to face all the emerging harms. That's. Being caused by advanced AI and so on. And the hope is that at some point people will just discover that this is a better, a more worthwhile direction. Maybe it's not worthwhile at all to replace our, uh, human race, uh, with some other, uh, silicon based, uh, stuff. Un unless you're the CEO. Yes. Uh, and as we have seen, uh, when people said to the CEOs
of big tech, uh, from Taiwan that you need to be liable for whichever scams advertisement that you put on because you've been earning, uh, advertisement dollars from those scammers. Uh, and the entire society is paying the consequences, the cost of such negative pollution, externalities. This is the kind of plurality technology that quickly lets the decision makers reign in the CEOs. And so I do believe that this steerability comes from the button up, but it also does need, uh, endorsements, uh, from the regulators, uh, to say basically, okay, it's not my idea. It's like a trade, uh, negotiation. It is the people's idea. So is that kind of, uh, a plurality is kind of like a decentralized singularity. Well, it's
a acceleration for decentralization, for democracy, and also for defense. Uh, so Vitalik Buting caused this d slash acc or defensive democratic decentralization acceleration. So it is an kind of acceleration in that we want, uh, the most possible equitable way of diffusion. Uh, but it's accelerates not in the sense of self-improvement, like the vertical singularity one. This could be applied in a lot of different areas. I'm specifically interested in how it could be used for the ongoing battle of what the future of social media could look like.
Mm-hmm. Especially with our. Aims of, of this podcast and your work and a lot of our colleagues and people in the world for a pro social, uh, future, what would be specific features of a social media platform rooted in the ideas of plurality and how would those look different than the platforms we have today? I'm, I'm sure you've thought about this and if not, are working on it. Yes, certainly. So, uh, I co-authored a paper called Pro-Social Media that talks about this. The idea very simply put is that in your newsfeed, instead of being ranked by the engagement or addiction that it generates, it can rank instead by the various communities that you belong to and how much coherence, how much uncommon ground each post can generate between those communities.
So. Each of us have very different like spiritual, professional, family and so on circles. And it's often the case that we ourselves are also figuring out how to take something that we feel cherished from one context across to another context. And the idea is that there are creators, uh, on social media that specialize in creating this kind of bridges so that people can understand the other community more and vice versa just by viewing and engaging with such content. And so for each post, you can then see of the communities you belong to, uh, which communities find this to be bridging. And which communities find this to be debatable. So it's like the polish interface, but apply to social media. We already have
that in the form of community notes, but it is kind of a debunking thing. Uh, you already have a trending, polarizing post, and then you can look at the community notes, uh, to have this kind of resonance and bridging. So the intuition is to move this into the main feed so that the main feed itself becomes, uh, pro-social. And in the paper we talk about,
for example, I'm involved, uh, in advising the Project Liberty Institute, uh, who, uh, works out a new economic model, uh, for TikTok if the people's bid succeeds, uh, in buying TikTok. Us. And so instead of the advertisers paying to bid for the highest bid, uh, getting the attention of each individual kind of strip mining the social fabric and making each person look at a wildly different feed, uh, this idea is recreate this common experience so that people can know, oh, um, your community and that community are enjoying this, um, together. So a little bit like those 10 people in the same room, people will be able to know that this is, uh, white resonance with the extended communities and it creates kind of a Super Bowl effect and things like that. And we conjecture
that the communities as well as brands will, uh, pay for this kind of shared experiences. So how prevalent are these various technologies? Some of the things, uh, the project you, you've mentioned in Taiwan. Mm-hmm. And is there any evidence that on some, uh, um, group of issues that Taiwanese population is less polarized than mm-hmm. Than other countries? Definitely, as I mentioned, across urban, rural, across, uh, age groups, across religion and so on.
Taiwanese people are the least, uh, polarized. And we can also simply compare the pro-social ranking algorithm that's deployed in LinkedIn, uh, versus say in Facebook, um, LinkedIn Q rates. Uh, its, uh, feed in a way that is not maximizing the time you spend on advertisement, but rather on the, uh, cohesion, the coherence that we just talked about. And so feed is quite different, or LinkedIn In Taiwan? No, no, LinkedIn globally. Uh, when they first introduced the newsfeed to LinkedIn, they were very intentional and then they curated this kind of common ground bridging, uh, posts from. Business leaders, uh, from people who follow, uh, who are followed by a lot of people on LinkedIn.
And then they gradually, uh, open up commenting and things like that. But the whole idea is to shape a norm where, uh, engaging with the feed actually adds to your, uh, sense of social cohesion instead of, uh, distracting, uh, from, its like Facebook did since 2015. So what are the barriers to this scaling, uh, pro-social, plurality based, um, social media? What, why isn't this I. Taking off more, this feels like something that people would want, uh, of all political ideologies and and backgrounds.
Yeah, definitely. Um, and it is true that I've been talking with, uh, many different people on different sides of ideologies and they all feel that it's time to move past peak polarization. And I do think that what we need now is, uh, both strategies. One is working with, uh, free software communities that runs those smaller but still very respectable sized, uh, networks such as Blue Sky with 30 million people on one side, and also choose social on the other, which is also free software. Uh, and, uh, in a way to show that we can bridge.
The contents so that people across true social and blue sky can find the uncommon ground, the, uh, surprising validators. So this is what we're doing. And the other is just to take an existing network like TikTok and just change its algorithm. And the idea of people spit is that TikTok needs to interoperate. Meaning if you post on TikTok, you should be able to consume the same content and link to the same friends on Blue Sky or on true social or on any other places.
And so people will then be able to curate their own experience instead of feeling locked in to the core, uh, recommendation algorithm of TikTok. And so this gives us much more grounds to experiment with the pro-social ranking. Just like everything else in our world though, isn't, um, our global economic system, uh, our national economic system, our corporate economic incentives are based on, uh, dollars and we get clicks for dollars. So, you know, when we use social media, we get some benefit and mm-hmm.
A lot of times it's dopamine based instead of oxytocin based. Mm-hmm. Um, to, to make a generality, but it results in an economic, uh, uh, gain for some individual or corporation. Mm-hmm. Does, does this still, um, does this combat that at all? Or how, how does that play into this? Yeah, the, the hope here is just as LinkedIn has demonstrated, there is a way to pay for. Common experiences and oxy toin based, uh, feelings, uh, while still making sure that whatever advertisement, whatever messages, uh, that you pay, um, can result, uh, in like Super Bowl, uh, which is the kind of pinnacle of common experience. And then you can build narratives and brands and so on in a way that individualized dopamine hits, uh, really cannot. Seriously. I, I think our culture has like a massive dopamine hangover. Um,
they may not know that, but we're so depleted. Uh, it's like we've all been on this Las Vegas junket and have lost all our coins and our brains are kind of fried and we're hungry for serotonin and oxytocin. Other of our ancestral neurotransmitters that we've been craving, and we get that through community and community engagement and social interactions, and the fact that we can possibly get that from social media mm-hmm. Is encouraging. Um. Don't you think? Yes. Uh, and there's a famous study a year and a half ago,
uh, a average undergrad in the US using TikTok. If you ask them to move off TikTok, then you will have to pay them almost $60 a month so they lose that much utility like fomo. Uh, while everybody else is still on that hamster wheel, but if there's a magic button you can press that can transplant everybody around them and themself into some other like non dopamine based, uh, platform, uh, then they're willing to pay you almost $30 a month. Um, and so it's obvious we're in a product market trap. Mm-hmm. Everybody lose utility
on the hamster wheel, but the first one to move off suffers so much fomo so that nobody want to be the first that moves off. Hmm. That's quite profound and dopamine is still worth two x, uh, serotonin and oxytocin in our current economic system. But that might change. Yes, that might change. Um, so, so you are,
uh, in your work, um, you're very specific in your projects and initiatives about the use of language and the importance of it. So why is language so important, uh, in these movements and for civic engagement and, and participation in general? Yeah, I think repeating, uh, the category errors, uh, of some business as usual language such as, I dunno saying human resources, uh, or incentivizing corporations, uh, just. Propagates this category error in our thinking. So it's like, um, trying to chart out a map, uh, but with like very tilted, uh, lens, uh, you, you can't perceive the world, right? Uh, if you use that sort of category error, um, where it's, and so in 2016, uh, when I first entered the cabinet as the digital minister, uh, I made a word play because in Taiwan, digital shuway also means plural. So I'm not just a digital minister, I'm also the minister for
plurality. I. So even though there's no ministry at the time, the ministry will come in 2022. I still wrote a job description, uh, as a shuway uh, minister. It goes like this very quick. When we see the internet of things, let's make it an internet of beings. When we see virtual reality, let's make it a shared reality. When we see machine learning, let's make it collaborative learning. When we see user experience, let's make it about human experience.
And whenever we hear that the singularity is near, let's always remember the plurality is here. Nice work, Audrey. Thank you. Um, I do think law, um, um, language is so important. Like fossil fuels. They're not fossil fuels. Mm-hmm. Um, they're fossil hydrocarbons. We're just choosing to use them as fuels. Mm-hmm. As, as one example, or we refer to the United States consumer spent more this month, like we're human beings who buy food and other things. We're not necessarily consumers, uh,
unless the true ecological sense, but yeah. Language is super important. Mm-hmm. Um, mm-hmm. Yes. Because we're marketing to each other. Yes. Consumer of foods is like, you know, referring to your users and, and it sums this,
you know, drug subscription, uh, case, right? So I think when I say user experience, uh, should be, instead, human experience, we're pointing out the same thing. That is to say there's much more to being human than just consuming something or getting addicted on something. So I've heard you, um, describe liberal democracy as a sort of social technology. Mm-hmm. That should be in constant innovation, uh, alongside other technologies. Mm-hmm. H how would you describe the current state of innovation for democracy itself and what is needed for it to keep pace mm-hmm. Uh, with other things in parallel that are going on in our world,
like artificial intelligence and other disruptive technology? Yeah, that's a great question. So, um, I analyze, uh, democracy as a communication technology that has, uh, bandwidth and latency. Bandwidth is how much, uh, information can each citizen communicate to their communities and also into decision making. So if you have a referendum, that's one bit of information. If you have a, um, votes on mayor, uh, with four plausible candidates, that's two bits of information. The problem is that the emerging technologies, they change our world in a way that demands, um, solutions to what's called wicked problems. Meaning that issues that require coordinated action of many, many different parts of the society. But if each part of society.
Can have two bits, three bits of information uploaded, then that's not sufficient information to piece together a solution, a kind of jigsaw puzzle, uh, to the wicked issue. And this is one part, and another part is latency. If you have to wait for four years for the next, uh, mayor or the next referendum and so on. Well, um, many incoming transformative threats, uh, can change the society to the point of no return in less than four years. And so think, um, not just pandemic, but also the info dynamic, uh, the polarization issue, the, uh, generative ai, power scams, phishing, and so on.
So all of these, you, you do not wait for four years and start a new referendum or vote in a new mayor or things like that. You. Immediately gets people together and very quickly gets much more bits than just a vote. Maybe you get conversations which is much more bits, or instead you get, uh, reflections on each other's posts and so on. Like in poll, no matter which way, you need to close the loop very quickly so that people know that within weeks or at most months, your idea results in the steering of the, uh, direction of the technology and its responses. And then people can come around again and again to learn the steerability. So I'm the cyber ambassador and cybernetics in Greek means steering. So this is about the art of steering.
I didn't know that. Um, so is there a risk that if we don't continue to innovate, uh, democracy as it is today mm-hmm. And all the liberties and freedoms that we've come to take for granted, uh, in our generations, that democracy will simply become obsolete in the face of accelerating AI towards the singularity and the changing global political landscape. Um, how worried are you about that and how do those concepts, uh, interrelate? I think, uh, there are various ways that people can see the incoming crisis, which is not just one but many. So some people say PO crisis, uh, but they're all isomorphic in the sense that if you see one crisis, you've also seen the shape of some of the other crisis as well.
So like a meta. Crisis. And so I do feel that, uh, our experience when it comes to whether it is occupying the parliament peacefully and keep it peaceful, or whether it is about countering the algorithmic, um, dispatch of Uber and of social media and, uh, the infoam and also the pandemic, uh, and generative AI harms and so on. Each of these examples, uh, shows that maybe a crisis, uh, as in weight is both in a danger and an opportunity. And so the shared danger is likely to. Make sure that people see the societal resilience as not a nice to have, but rather something that people must contribute to. So the wildfire, um, recovery issue,
uh, on engaged California is a great, uh, result of this infrastructure level building. And then when such a topic comes in, then people can pivot and respond very quickly to it. So I'm not pessimistic at all. I feel that each of those incoming threats actually accelerate the diffusion and the common knowledge of the people that democracy does need, uh, improvement as the social technology. Audrey, why are, are concepts, uh, like responsibility, liability, inclusivity, and transparency, um, important, uh, for creating and maintaining an open democratic governance system of, of the type that you've been describing? I.
Yeah. Um, I learned this, uh, when I entered the cabinet, um, because, uh, in 2016, uh, I entered the cabinet with some of that doge energy, you know, uh, wanting to make everything transparent, want to make a procurement, like a leaderboard of people comparing, uh, and things like that. Shortening the, uh, tax filing from three hours to three minutes, uh, through direct file, um, and so on. And so all these, like what we did that in like 2016 and so on, but we very quickly found out people in the career, public service, the career public servants. They also had the same idea, and they are also like great reformers. They actually know how to do things better. It was just they lack a air
cover. There's no one who say, uh, if you do this well then, um, it's you who get a credit and if you do this, uh, but it doesn't work and I can take the blame. Uh, and so I made sure that we align our, this energy of democratic innovation to the. Languages and the logic, uh, that the career, public service, uh, especially the, uh, planning and research and development departments use. And so in Taiwan we have the National Development Council, and to them always, uh, transparency, accountability, uh, is I. The norm. Uh, and if we add participation and inclusive participation at that to it,
they want to know that this participation is accountable so that we can regulate, um, this institution into new institutions, not just challenging and taking down existing institutions. So we announced our every move, everything like the join platform, the participation office and so on. Uh, instead of just doing it, uh, as code, we said, okay, six days from now we're going to do it and here's a public commentary period. And we made sure that there's no exceptions. Everything needs to be pre announced publicly this way. And so even though that each of our move takes like 60 days more, I think we want much more support from the career public service because they can see that. I'm designing myself out,
so to speak. Uh, if I'm no longer the minister, all those institutions, the new designs are still around because it conforms to the logic of the bureaucracy. I imagine that there are many other countries in the world, some countries are very interested in copying your success in Taiwan. Mm-hmm. And others are also afraid of, of implementing some of these things. Mm-hmm. I mean, in your opinion, should countries, uh,
be doing more to regulate social media platforms to be in line with these principles and, and what are some of the, uh, the benefits and risks to such government oversight and any comments there? So for this kind of broad listening and sense making, I think the smaller the polity, the easier it is to implement. Uh, to your point about Dunbar's number, pretty much any polity, if it's just 150 people, they don't have to run a sortition. They just invite, uh, everybody right to a conversation. Uh, and we do see that in many countries, like in Japan there's a long tradition of citizen assemblies, but on a hyper local level, like literally township level, uh, and that, uh, has worked well.
Do we have the technology to do that at a township level now? Yes, we do. It's the same technology. It's just easier to implement, uh, and gets buy-in from a mayor of a town as opposed to say, you know, a federal government. Right. So it's usually easier to start. I want you to finish answering this question,
but just so I understand, I. In the United States right now, people in Topeka, Kansas, or Red Wing, Minnesota mm-hmm. Or Sebastopol, California mm-hmm. Could access some existing technology right now to do Oh, yeah. Some of the things You're ta what, what would that be? Yes. What TE technology. Yeah. As, as
I mentioned, the Bowling Green, uh, process is ongoing, right? So if you just search for Bowling Green, uh, Kentucky, uh, sensemaking, uh, or polis, uh, or better Bowling Green, uh, you can see exactly how it's done. It's all open source, not just the P platform, but also the sensemaking tool. Uh, they're all free software free for anyone to use. And so there are some US states, uh, with.
Citizen assembly tradition already in an in-person kind like in Oregon. Uh, and so in that sense then it's not about convincing them to move online, but rather using digital tools to augment the conversation and to improve its reach. So like Democracy Next has been working with Oregon people on that. So the Bowling Green and the Oregon, there are, there are entities that are working and chaperoning that process. Mm-hmm. But in theory, anyone listening to this show could look at the Bowling Green example, access the source code, and start something in their own community. Yes, definitely you can roll out Polis installations. Uh, at PL is, and the sales making tools, uh,
you just search for Jigsaw, sensemaking and Polis, I think now have integrated that logic. So it can also use language models to do a very balanced reporting of people's ideas. So you can close the loop like literally within, um, a minute or so for the mayor, uh, to maybe read every morning. Let, let me ask you a, a related question. Not, uh, to do with democracy per se. Um,
but I've noticed, um, over the years, um, decades of convening groups of. High status scientists and activists, that everyone's got an opinion and they're very smart, and you get 80 or a hundred people together. But what ends up happening is when you're in person or when your name is attached to something, people, since we're social primates, uh, and we compare and look at status metrics, they defer to the senior wealthiest, or most famous, or most influential per person in the group. And so they don't mm-hmm. Let their, their real thoughts, um, be known. So I'm wondering, the technology that you just described about the Bowling Green, could that be used mm-hmm. In an institution itself where there's 200 people and you really wanna know
what people are thinking without fear of saying the wrong thing and getting demoted or anything? Is, would this apply to those situations as well? Yes. And, uh, there are technologies for the in-person, uh, kind like cortico, C-O-R-T-I-C-O, and develop out of MIT. This tool, uh, you can just put your phone or a round microphone on table, and then it ensures that the facilitator is guided by not just, uh, the conversation guide, the turn taking, you know, not letting the single senior person dominate conversation, but can also carry other conversations from previous, uh, talks, uh, to this particular conversation pod so that the conversation network can cross pollinate. So when the most senior person speaks something, the facilitator can then press a key and then a method, uh, place from some other conversations that counterbalances, uh, the point that was just made. Why didn't I know about this? And what is, what is holding this sort of technology back? Is it, is it awareness, uh, like in my case, or is it money or is it, um, big tech is, uh, afraid of these things, uh, or is it social organization? Why aren't these things scaling more rapidly? I think, uh, one of the main reason, uh, was that all these things run on oxy, toin and serotonin, right? And so it is a, it is a vibe thing. Once you're in this vibe,
uh, then it's more likely that you will participate in one of those, um, conversations and you will discover a very large rise on like, conversation network. But if you're dopamine bound, it's very difficult. Yes. So actually we need to heal people's dopamine addictions, uh, concurrently so that they move into this more, uh, um, zen, uh, holistic human experience. And then
obviously this is the type of social media that I would prefer rather than clicks and likes and, and unexpect reward of, of some goat that claps and falls down and a snake crawls under it. And woo, I never saw that before. Um, which doesn't really give us much meaning or depth or purpose to our lives Anyways. Oh yeah, definitely. In
2025-03-27