Taiwan in the bullseye of China’s AI-charged info warfare. With You-Hao Lai of DSET

Stop the world! Welcome to Stop the World, I'm Olivia Nelson. And I'm David Rowe. Taiwan is a target of more information operation attacks than any place on Earth. Most of it coming from mainland China and mostly aimed at dividing Taiwanese society and working in resolve to resist an eventual takeover by Beijing. AI is supercharging this steady barrage of disinformation as it is with so many fields in national security and elsewhere. And we're talking about this today with Yu Hao Li, who is a researcher at the Taiwanese Research Institute for Democracy Society and Emerging Technology. We start off by talking briefly about the news of the last week, which is the Global AI Conference that China hosted and Xi Jinping's speech to that conference, which lived was the first time he had personally addressed it, which was significant in itself. Yu Hao gives us his takeaways from that. He then provides his thoughts about Beijing's effort to dominate global adoption of AI, which means embedding Chinese tech into, with Chinese standards into global systems rather than American tech and also the consequences of that. So we talk some of that. Yu Hao then talks through how AI is changing China's information operations in terms of scale, speed and sophistication, the kind of narratives it uses and the strategic purpose of these operations. He gives some great examples along the way, which you really enjoyed, David. And the documented cases that really highlight how things are evolving. And then he finishes up by talking through Taiwan's whole of nation responses, which he puts in terms of a framework of four C's, cutting, clarifying, curbing and cultivating. It's a great framework. And there are a lot of lessons the rest of us can take away from this, because Taiwan has been doing a lot of work in recent years. It's really built up an impressive program with National Resilience. And so it's worth paying close attention to as are the remarks of Yu Hao. So let's hear from him. Yu Hao, thanks for coming on the podcast. Thanks for having me. So you were kind enough to join us at Sydney Dialogue, our tech conference last year. And you talked about China and influence operations and artificial intelligence and how they all fit together. We thought it might be a great time to catch up with you again to talk a bit more about this. You do a lot of really fascinating research along with your organization, Dset, based in Taipei, around this area. So mostly going to talk about influence operations. But you know all the news asked we cause some really, really fascinating time to be following artificial intelligence. Obviously, it's, I mean, it seems like every week these days is carrying a lot of eventful news and development. But last week the Chinese government Xi Jinping gave his speech, major speech on artificial intelligence at the world AI cooperation organization. I just wouldn't like getting your elections first of all on that. What did you make of Xi's speech and the creation of the world AI cooperation organization? What is the Chinese government trying to achieve here? Do you think? Yeah. So I feel like it's really interesting to see Xi attending the world AI conference in Shanghai and in person or the first time actually is the first time. So I believe that we know he has something to say, some important message to deliver. And I think at least time, a lot of message was really about defining what like why completely new established organization means. And I, for me, I feel like there's three important like message or as a significant, significant, I want to highlight here. And first it's more like, I feel like it's more about institution building. There's another step towards China leading how AI government standards and principles get set. And actually like China's PRC's premier Li Chang, he used to propose that the idea that there should be a China-led international organization on AI back in 2025. And this year it actually happened. So I believe that is taking turning its own organization approach into something always like real institutions behind it. And more importantly, it should be international organization. And second, I feel the important thing is the target audience of this organization is for sure the global south because when you see the 29 like signatories, there are largely countries from global south and they are already inside the digital sale grow and the belt and road initiative and also other China's broader digital cooperation programs. So this country trust Africa, Middle East, Asia, Latin America and they are mostly like from global south. So that's from a fully target audience side. I think that is also something we are worth watching. And the third is that I believe this is the strategy actually we've seen before is something we are already familiar with. Just apply to a new domain in AI. I mean China has been using infrastructure as a leverage to spread governance values and standards. So we can see that clearly in the initiatives like digital sale grow. So this time under YPO, it looks really very familiar that like she proposed like five thousand AI training opportunities for PRC's partner countries and also invited countries to adopt China's AI powered systems like meteorological warning system called Khmazu. He said that in his remark. And like he also emphasized China's support, long the long stand is important for open source AI development. And obviously let China again, bigger voice on governance standards simply by the few seen the whole AI ecosystem and is infrastructure. So that is something I believe is what I see from least organization establishment and also the implication of that. Yeah. Yeah. Look, I think that's all spot on and I think I mean, broadly speaking, those are the same takeaways. I must say that I got out of sheer speech. So I think we're on the same page there. Clearly China is on a push to win the AI adoption race. That seems to be a pretty common interpretation. I think that naturally flows from a lot of the points that you were making there in your answer. Talk through your, in your view, what the implications of that, what does it mean for the rest of the world? What does it mean for Taiwan? What does it mean for Australia? What does it mean for other countries? If China leads the US in perhaps below frontier level, but good enough, everyday use both by enterprise and consumers, that kind of unofficial intelligence that basically just becomes the default that the world starts to choose. Partly it has to do with the open source open weight nature of it, of course, which makes it much more affordable for most users. There's a big question, a big debate going on at the moment about whether China will continue, whether Chinese government will continue to encourage Chinese companies to go open source, open weight, given that the implications for revenue for those companies and whether it's just an affordable, affordable over the long term, but also then the control risks start to grow, of course, as well. I mean, open source, open weight models, some introduced source of risks. So that's a big question. So this is rather roundabout way, but I'm just interested in your view on the implications of China with the adoption race, and particularly around things like sovereignty of data and control of data, which I know is something that you've done some previous work on. Yeah. So yeah, as you say, I believe China is on AI industrial policy. It keeps pushing to accelerate the global adoption of Chinese AI technologies. And just like you mentioned, at least on the model side, one of the most important events or big news in the past few days is that moonshot just released K3. And actually it's almost catch. I mean, like from the performance, if you see performance is almost catch most frontier models in Western camp, including Cloud and also open chatributes mostly frontier models. So besides that, I still believe like open weight and open source, you can see it has consistently been a defining feature of China's AI development. But before I dip deeper into the implications of that, I want to highlight something, my colleagues research. It has done maybe a few months ago, because from what he told me from his research, he wants to highlight that this kind of open source open weight approach is not something really like tap down national strategy. You can also see that from like a bottom up character or bottom up survival strategy that as Trump's role-concentrement crisis of Chinese companies fighting to serve on their Western's cheap controls. Because lately, Bell is trying to open source based AI comments. My call to call it AI comments is more like a public comments for the Chinese AI companies to cut down, to bring down the cost of developing frontier models. So this is something I just want on the highlight here. It's not just a top down strategy, it's more like bottom up. But still, I built these open way to open source development. It's something that Chinese or PRC government wants to see. Because that's really something that you can leverage. Because open way to open source can help the diffusion of China's AI models globally. And based on that, you can leverage infrastructure power over governance, then, or setting, and other things. So that comes to the implication that you just want to talk more about. I feel like, because she mentioned in his remark that he wants to against overstretching the concept of national security. But what I find it really interesting, or ironic to science 10, is that Chinese AI development might be most, where it can say, more constraint by national security considerations and almost anywhere else. Because it can see that constraint gets built deeply and directly into how data is collected, how models are built, and how algorithms are designed based on she's concept of holistic national security, in Mandarin, called Don Quixia Anchen Quan. So by regulation, like China's generated AI development must build in this kind of security, security concerns or considerations, that is something we can't neglect. So then it comes to my January report, if you want to talk more about the data collection side. In my report, what I tried to show is that when these Chinese AI services or systems expand overseas, that kind of expansion itself will help Chinese government or the Chinese AI ecosystem to have more global data collection. Because these services tend to build multiple channels or pathways for data to flow back into China. I identified at least three pathways. But firstly, it's more like directly stored those data in PRC's jurisdiction. And another channel is intra-group sharing with Chinese parent companies or entities. And then you can also see like, or many actually, every, all the service providers, they all maintain or contain the legal provisions that enable direct government access to those data. So I feel like that doesn't just give China more data. The broader implication or deeper implication is it also quite least expands China's own data governance model out into the world through those AI infrastructure. If you see AI services or systems as kind of infrastructure. And that is also something I think we can see through the open source, open weight model diffusion. They also project projects or trying to diffuse the world view and the values and also the norms embedded inside those models. And that is actually why when I read Washington's, I mean, the US own AI action plan, it also emphasized that the model holds some kind of geostrategy, a geostrategy value and also cost for the American open models to try to fund it on American values as well. So you can also see that the size, they also see the importance of the model diffusion and also the values and the work views that might become part of the model diffusion and becoming infrastructure of the future information environments. Yeah. No, I think you raised a really good point there about the action plan. And I remember it well too. And I think it's been referred to back to a few times in the past week as some of the analysis and commentary has been made about the Chinese, the latest speech from Xi and related issues. Let's move on to the influence operation side, which is sort of the main thing that we'll be talking about. First of all, I'm just interested in a summary, a sort of overview of how AI is changing the way that China is carrying out influence operations against Taiwan. What have you observed over the last couple of years or so about how that is evolving? Sure. So you probably know that Taiwan has been maybe single most targeted country in the world. Sure. Yeah, foreign disvermation or information manipulation or open of decade according to VDEMs data based research. So it's not bad. What was changed with AI? I would say AI has, I would say, is more like massively increased the capacity to manipulate across the whole life cycle of information manipulation through production of information and also the distribution of information and then to the reception of information. Yeah. And before the AI edge, I would say it used to be more scattered. It's more like a scattered, scattered, but now it becomes more constant, more industrialized and become more technology empowered efforts that help the malicious actors to manipulate the whole information environment to try and find some like infrastructural notes to manipulate and try to intervene different stages of the life cycle of information operation. So that is highly of other states that I want to highlight here. Okay. So presumably, I mean, the volume I would imagine has increased. I mean, if you automate certain aspects of it and you reduce the cost, you reduce the, I mean, the highest cost usually is people, isn't it? So, you know, the more, particularly on the production side, but as you say, the distribution and the reception all have implications too, but on the production side, most obviously, it's easy to see how, rather than taking a person 10 years, 10 hours, I should say, to craft a, you know, a particular meme or something like this or, you know, to come up with some sort of design that is some, that is going to be appealing to an audience, automating all of that just makes it much faster and cheaper. So perhaps you could speak a little bit about the production side, but I'm particularly interested as well in just how it's evolving from, I suppose, you know, disparate or isolated pieces of information into a more kind of comprehensive information ecosystem of, you know, that is manipulative or that is misleading and that is trying to have a kind of strategic effect. So can you talk a little bit more about how it's becoming more sort of sophisticated in that way? Sure. And I think I can try to explain that through maybe two examples. The first is a research from a Chinese group called FactLink. It's a newly established organization working on like Fact Checking and also like Fact Checking Related Research and actually most of their family members are friends of mine and I learned a lot from their work. Yeah. I believe one important research or report that they're trying to observe a great recall movement last year in Taiwan. If you, I believe like many of your audience might not be familiar with this but last year in the US, Taiwan went through a nationwide wave of recall votes like citizens launched this kind of recall movement and mostly against obsession lawmakers. And throughout this process, like FactLink observed plenty of information manipulation during the movement. That is something like you can expect, right? But what they did things back in the first place that there's a clear difference they see, they see through at least this new wave of manipulation compared to what they observed in the past. So the clear difference, the content was not about a false plan anymore. Now they see more about volume, more about the flood of Qi. AI generated short videos designed to provoke and manipulate emotion. So they take or they read is that we've already moved from this information campaign to propaganda, driven manipulation. So once a lot of something important to them, they're going to be like, "Oh, I'm going for Lemmy's because if now the problem is no longer about true sponsors or falsehood, land for land as a check back in group, fact check himself, start losing his footing. So that's something for Lemmy's really concerning. And Lenn, I also want to clarify that there, what were they trying to get Taiwanese people to think or to feel? Were they trying to, instead of getting them to try and believe something that wasn't true, they were trying to provoke their emotions and their feelings about this particular issue of, sorry, of vote recalls? Can you just explain a little bit further what, you know, what was their intention with that? What were they trying to actually achieve in the Taiwanese society? I feel like maybe the broads' implication of that is, I feel like all the time information and manipulation for the attackers, they don't really focus on, and actually don't need to focus on like using fake or false content to achieve their goals. But you can also try to say something is true, but maybe you interpret that in another way, or maybe you can use some misleading narratives to try to amplify the existing division in the society, right? So I believe what they see or what I observe is that for a long time, like fake or fake content or falsehood is not necessarily the key. But for now, because AI, especially when ready AI, come a available tool, it's very important for them to produce a lot of content as scale. So for them, they can produce a lot of content and try to find different ways to provoke emotion, try to produce different kinds of misleading narratives. So they'll help them to further amplify their capability to do this kind of misleading or manipulative content. And that is not necessarily related to falsehood or fake content. So that is something they've been seeing as through what is happening in the infinite I recall, right? And tell us about the other instance that you were going to mention. I think you mentioned two examples. Yeah, just fine to highlight another example is something you might be familiar with is a PRC military linked it firm called Galaxy. Yeah. Okay. So like last year researchers at Vanderbue, Vanderbue, University of US released several, maybe hundreds of pages of these companies internal document. And after that, this year Taiwan's double think them, double think that they put together a detailed analysis on this, on those documents as well. And I believe this is a good case, for example, to know how AI actually has been used in foreign information manipulation and interference across information production, distribution, and also reception. So like production side, the firm, I mean, Galaxy, it claimed that they put on a automated platform powered by a generated AI to manufacture lots of manipulative content as scale. And then on the distribution side, they also claim that they build a system that can deploy large numbers of AI personas to orchestrate manipulation. And that is something interesting is that because it echoes a report I read maybe a few years ago, it was a report launched by Rand, a US based thing tank. So back in 2023, they warned that AI's impact on information manipulation would not necessarily state confide to message. It can also extend to the messenger itself. And now it's the like as a person as a messenger as a distributor, they become harder to detect and easier to like deceive the receiver of those information because of AI generate the power of generated AI and increasingly powerful generated AI can help them to become more and more human like and more and more tailored to Taiwan, Taiwanese people's language habits or social context. So that's something on the distribution side. Air personas can be a more and more concerning threat. And also on the reception side, that is also something I believe we can also see is because the content is built on large scale profiling of the target society including Taiwan and it customized to hit people's emotion as hard as possible based on large data collection and analysis and then profile. So I believe these two examples can showcase how AI has been using in information manipulation and trying to amplify trying to enhance its power. So there are some concerning consequences but I can talk about lowly ground. They're sure are and I've got to say I've all of that I find the reception side in a way the most terrifying of all the idea that it is very personalized, very customized to target me as an individual or to try and manipulate me as an individual and that perhaps the way I am responding to something is not actually my own organic responses that itself is actually being sort of tailored and abused almost by some external actor. I mean that is just a basic human level. I find that to be a really, really terrifying idea. Tell us a bit about some of the sort of key narratives. I mean what is China trying, what are the main China is China prosecuting Taiwan? What are they trying to get Taiwanese people to believe think feel? All right so actually this is also something like several Taiwanese organizations and groups has been researching for years. For example I double think that that's the organization I mentioned earlier. It has mapped out many sets of themes like narratives about government corruption or about Taiwan's democracy being incapable of doing job is incompetent and also narrative meant to weaken resistance to identification. So I can give you some example if you think that would be a little bit. I mean that makes sense. Trying to erode trust in the government. That's an obvious one and trying to weaken people's resolve to resent. Convincing people that one day somehow or other Taiwan is going to be incorporated into mainland China. There is nothing you can do to stop that. So you might as well do it the hard way. You might as well do it the easy way. Otherwise we can do it the hard way which is going to be much more costly and much more painful for you. That all makes sense. But yeah, maybe give us a couple of examples that you found interesting. So from what I read during the 29th election, there are many fans from some like we hear messengers were fake accounts. They're trying to manipulate dispute over imported acts and also the US court labeling. They have something trans people has been discussing in 2024. And this kind of like social dispute got weaponized into plans like the ruling parties is corrupt and also like the government tolerates the scammers, something like that. And also like Beijing's own like framing like we're in Mandarin we call it Liang Aichan like two shores one family and also like Xie Nong Yuxue means like block a block thicker land water. They have been using these kind of narratives and trying to position PRC or China as a family not a not rattling adversary and trying to soften or trying to reduce the resistance to an education that way. So let's let's think about that. It's interesting that there are positive narratives as well as negative. I don't want to say I'm necessarily endorsing that or embracing it. But I mean obviously Taiwan has a it's right to make up its own mind about and to you know be the its own determinant of its future. But you know that's it's a less objectionable narrative than some of the other ones I suppose. Yeah, yeah. And I can also give you some other examples about like how maybe PRC has been using this kind of it has been using the information organization and trying to promote like narratives that tries to maybe detect the trust of maybe allies, like maybe like Taiwan, US, or even like Japan. So for example, another important research organization in Taiwan called IORG, they documented many distinct narrative variants for many years. You can see many problematic narratives like Taiwan will be abandoned by the US or US is not a good friend to Taiwan, they are a false friend, they are not the honest friend. And maybe some blood narratives has been even more powerful in the past few years because of the geopolitical change or some like past happening in the past few years. So that's something I believe we can also see if you want me to talk more about how this narrative would harm the alliance or maybe the trust between different countries if you tell you the democratic can. So I think if you like-- Perhaps just touch on that, I mean it's interesting, apart from anything else, one concern that I have is that sometimes Taiwan's allies are making it a bit easier for China in that way. I think that some of Donald Trump's commentary after his visit to Beijing for instance, when he didn't say all that much while he was actually in Beijing, but afterwards he did a couple of interviews with Fox News that could hardly have been encouraging for the Taiwanese people. And so I suppose I mean one concern that I would have is just that sometimes some of the work is being done for Beijing by other countries, there are probably other examples as well. But perhaps just give us a bit of a bit of a flavor of your view on all that. You mean like more examples about how this kind of narrative has been using from the PRC's drive? Correct, correct. Yeah. I would say as far as I remember, there's a one concrete example is that when the US was through from Afghanistan in 2021, there was a narrative like they have Afghanistan tomorrow would be Taiwan. That's this kind of, yeah. Wow, that's kind of like Taiwan would be abandoned by you by its allies. This kind of narrative spreading through more than a hundred posts on social media. And but the problem is that sometimes this kind of narratively they originate from Taiwan's own outlet. But then that would be accoled or like further amplified by Chinese like state media or maybe some of their supporters. So one thing we can see here is that most some most of time many of these kind of narratives lay originally from divisions that already exists inside Taiwan. So these kind of operations they often often ultimately they don't invade invent new divisions. They just amplify what already lay there and trying to manipulate whatever to them. Yeah. Classic, classic. And I mean, I mean, I'm not an expert on this, but my impression is that the Russians have traditionally been very, very good at that. They understand the the adversary society very, very well and know exactly what those divisions are that already exist authentically in the country. And they're very good at finding those sort of sensitive points that they can push at in order to drive those divisions wider. Do you think that China is getting better at that sort of more nuanced, I suppose, style of information or influence operation? Yeah, I think so. And that's also relates to what I mentioned earlier because when like China or PRC like gets more and more like personal data or like personal information like from a target society, it can build up a more nuanced understanding of the target society and trying to find some like most important divisions or disputes happening in the target society and then trying to manipulate that, trying to find out the most like influential narratives than to achieve their goals. So I believe there's something we have seen witnessing with the help of the emerging technology especially the gendered AI. And also we can also see like some similar playbooks on narratives like in how Russia attacked Ukraine and also how China or PRC attacked Taiwan. So similar narratives like attacking governments of capacity and also trying to attack the trust between the allies and also trying to convince their people in the target society that inification is not a bad thing. So you can also see similar things happening in different authoritarian regimes there are tactics and narratives. Interesting, thanks. All right, perhaps we can sort of spend the last few minutes just talking a little bit about Taiwan's response and how Taiwan has been dealing with this. Tell us a bit about this sort of whole of society model that Taiwan has adopted. So I have the impression that different sectors of society, government, civil society, the technology industry, the general public have figured out a pretty good way to sort of coordinate and complement one another and work with each other. Is that right? Just tell me a little bit about that. Yeah, I think you are right. And trying to map the Taiwan's the whole of a savvy defense approach is also something this has been working on. Like we published two reports on this in the past few years. One is about how generate AI is affecting information environment. Another one is exactly how Taiwan's public savvy defense model builds resilience to maintain truth. And like, so throughout these research projects, I usually use a framework. I called a 4C strategy to talk about these efforts because I believe they follows how information gets produced, distributed and received this this framing is something I have been using for a long time. So this 4C is more like first is to like cutting production and then clarifying falsehood and then curbing dissemination and then to cultivate in media literacy. So I believe it is a useful framework to see how like different actors, I mean government, society, civil society and maybe public companies how they trying to do things at a different level. So I can study from like cutting production. It's more like, I feel like it's mostly a job for a government and also AI model providers. For example, in 2023, Taiwan's government amended the law to raise penalties for defec disformation targeting candidates and then Taiwan's prosecutors also set up a tax force to trace where defecs originate during the election. Yeah. And another example is it's my mind not a lot related to Taiwan's context, but it's like something I believe is worse highlighting is that so like a few months ago, open night, open AI released a threat report and then it documented a Chinese law enforcement account that asked chat GBT to throw up a plan to smear Japan's incoming Prime Minister Tashikashi, but the chat GBT refused to do so. So I feel like what I mean by like cutting production is more like you need to try to eradicate the production information upstream. So maybe you should do that through maybe like tailored criminal penalties and also those who hold the infrastructure for information production like AI models. So open AI tries to prevent their models from like being misused by malicious actors. It's one standpoint. Another example is how government might be able to leverage their like a regulatory tools to stop or at least raise some cost for the minister's actors to produce this kind of misleading content. That is something I believe is very important to see at least this level. I mean cutting production level. There are also the other three dimensions I believe is really important. And just before you get onto those presumably there's no way to cut production in mainland China though, correct? I mean that's something that there's just not much that the Taiwanese government can do about. Sorry, could you start again? Sorry, I'll start the question again and we can edit. Before you get onto the other three C's just tell me unless I'm misunderstanding there's no way for the Taiwanese government to cut production in mainland China itself, correct? They don't they just don't have any. So presumably. Yeah. So that's why I proposed that this should be a framework. This should be a 4C framework because there's no silver bullet to address this. You can not. And anticipate that we can eradicate the information for content production in mainland China. So there should be other three pillars of the slide that you can use. So when it comes to clarifying falsehood, I believe some Let it something people are more familiar with Taiwan's experiences. Because here, like both government and civil society, also they both put real efforts in here. For example, since 2018, Chinese government has been using a strategy called they called triple-2 principle, or you can say, two-to-two principle. Meaning that trying to debunk within two hours and then trying to through two images and with two hundred, no more than two hundred words. And also like during COVID-19, they also have daily press conferences to respond to those problematic plans in real time. And I believe all these kinds of efforts help to debunk problematic or false information in real time. But let us something the government can do on one hand. But we all know that not everyone believes what government says. So that's why we also need some people, some actors from civil society, especially those independent fact checkers. And Taiwan do have a few. I mean, there are several organizations working on this, including like FactTrack Center and also my GOPAN. These two organizations, they are internationally recognized fact checkers. Yeah, but there's another one called CoFACTS. It's more like a community driven fact checker organizations. They try to trend a bunch of fact checkers. They are just citizens. They are willing to, they are voluntarily willing to take questions or some submissions from people on their platforms and then trying to do a fact check and then trying to release their result on the platform. And you can choose if you want to, if you want to believe it or not. So it's more like a more professional version of community knows that has been developed by acts or other social media platform in recent years. But if they only confide the fact checkers to those has been trended, they are still a community. There are still many people. They are not, because like for the Taiwan fact checkers and also my GOPAN, maybe there are only a few people. They are, most of them, they are journalists in the past. So they are only a few people and they are trend to do fact checking and they got a certification from some like international organizations to proof that they are able to do this. They are capable of doing this. But for the co-facts, they are more like a voluntary group driven by county citizens. So there are different models trying to do this in different ways. And I believe that's something like helpful and also really healthy because we really need more people trying, they are willing to engage with this kind of fact checking or debunking efforts. Absolutely. And it gives that sense of a layered approach. You really need to take lots of different styles and different approaches with different layers to them. And I think that seems to be a real highlight of Taiwan's. But so on. Do you want to just quickly mention curbing and cultivating? Yeah, sure. So like for curbing dissemination, that is something I, for myself, I believe that is most importantly, especially in the HIP. Because as Wata Brede mentioned, the HIP has already flooded the information environment. So for now, those fact checking organizations, they do a great job. I need to admit that. But for now, because of the floating of those agents generated by county, fact checking more and more becomes a cat and mouse game. You don't know what I mean. And also, so that is something I believe the government and the civil society group and also platforms need to work together trying to curb the problematic spread or dissemination of those misleading information. So in Taiwan's context, it has been for years. When civil society groups, they notice some abnormal patterns of information spread. They will flag it to the platforms and the platform will decide whether to take down those content or even remove some bot accounts or fake accounts. So it's more organic mechanism trying to identify to detect problematic dissemination of information and then trying to curb them from reaching more people, more audience before they can do it, before they do so. So that is something I believe is even more important for now, because as I just mentioned, floating becomes one more serious problem in Asia. Now, and the last pillar is cultivating literacy. And that is something I would say is the most important defense line for long term, because according to a report I read before, it's released by Carnegie Endowment for International Peace. They find that it is a literacy for long term, it's the most effective approach to counter information environment, information manipulation. But that is also something you need to put most resources into it. And you can see the effect of that unless you invest into it for a long enough time. So that is also something we need to do for sure. But for now, I was still emphasize the importance of curbing dissemination. That the four pillars I just mentioned, let's read the framework, you need to try to enhance every element of dimension of that, trying to address the problem of information environment. And that is something I believe the Taiwan's whole society defense mechanism trying to teach the world, trying to give the lesson to the world. It's brilliant. It's a wonderful framework. I suppose we can all have our favorites. If yours is curbing, I might say mine is probably cultivating. I do feel really strongly about building digital literacy, starting from kindergarten or primary school right through life and making sure it spreads right across society. So encouraging parents to teach their kids to question what they actually read on Instagram, or see on Instagram, or read when they do a Google search or increasingly an AI request. So personally, I feel if I had one wish, frankly, this is ridiculous. But if I had one wish in the world, I would probably ask all democratic governments to invest heavily in a very sort of cross-society digital resilience and digital literacy program. Because I think that's-- Yeah, I suppose. That means resources, right? Sorry. I say, like, that means love resources that you believe. It does. But I think it'll be worth it over the long run, because I think without that, without having people's critical thinking as the ultimate backstop-- and now, sorry, I'm just editorializing-- but without having that, I think that countries are really going to struggle. And I think it's investment that would be well worth it over the long term in the coming decades. Yeah, yeah, I suppose. But I also want to-- I still want to emphasize that we cannot rely on these strategies alone. I mean, we cannot only rely on mediators to stay alone, because it risks putting the entire burden of responding to the state level attack onto individual citizens. Sure. So you still need another three pillars or another strategy to address these threats altogether. Absolutely. And that's a very valid point. All right, you have-- I'm conscious of time. You've been very generous. It's been great talking to you. But look, I really appreciate it. It's a good chance. So thanks for coming on Stop the World. Thank you, and look forward to my future conversation. You'll be welcome back to Australia any time. How about love to? Thanks for listening to Stop the World. We're back with another episode soon.