…For example, one room said, “Let’s make sure all ads on social media are displayed as ‘probably scam’ unless digitally signed by someone using a digital signature KYC (Know Your Customer) we can trust .” Another room said, “If somebody loses seven million dollars to an investment scam that’s not digitally signed, let’s hold Facebook or other social media liable for the full seven million-dollar damage.” Another room said, “If TikTok does not settle in an office and ignores our liability rules, let’s slow down the connection to their videos so their…
I wanted to say first of all, Meaningful transparency can be helpful in building trust in AI technologies. But it is important that due care is applied when disclosing information about AI models, such as through the “submission” of LLMs for evaluation. This is especially because such comparison requirements, such as submitting LLM for evaluation may conflict with other importantly principles such as privacy, security and the protection of trade secrets. We also want to note that the EU AI act, for instance, does not mandate companies to submit their LLM or evaluation. Rather, it…
So, like tomorrow, we actually sent 170,000 SMS, random SMS, to Taiwanese people using the trusted number 111, so they know it comes from the government. And out of the people who filled in the survey, we chose 400 people that are representative of the Taiwanese population statistics. So, we know that these people are the microcosm of Taiwan. And tomorrow, for the entire afternoon, to the evening, we will send the data to the government. And in the afternoon, to the evening, they will meet online, in the stand for online deliberation, and…
…When you start from a lack of trust in people, when you see the role of government as being to manage or deliver for people, you end up with this kind of centralized power that is crushing. I find it so inspiring to see and hear Audrey talking about this. Seeing people as citizens, as capable contributors, means that you use technology entirely differently. The starting point is how do we make the most of what people are capable of and assist them to do it, rather than how do we supersede.
…It could be zero trust , it could be zero knowledge, proofs, and things like that across all ministries because they did not have the incentive for investment. But now that we’re attacked, they suddenly have, and we just take that opportunity and then reconfigure our infrastructure. I hope that answers your question.
So, the first thing we addressed with radical transparency is to radically trust the people; if the people see a public service that is not designed well, instead of protesting on the demand side, they can switch to the supply side by demonstrating how to do it better. If they feel that the contact tracing is not respecting privacy because of radical transparency, they can design better contact tracing system that preserved the privacy and indeed helped Taiwan to last until Omicron and we never locked down any cities and we reported one of the…
…No trust , no legitimacy, no credibility, because nobody could get fishing licenses that summer, right. It undermined their economic success. But for many people, the fishermen, the only institution that still had legitimacy was the University of minnesota. Right.
…Once it’s certified, then both the public sector and the private sector and the social sector can all use the same FIDO compliant ways to implement zero trust and our architecture as to make sure that it remains interoperable across sectors and we don’t have to independently retrain or recertify or things like that. That’s the first question. In terms of TAIDE, I think TAIDE explicitly is designed not to answer open-ended questions. It is just designed for the use case where all the bits in the output can be found in…
…Some swore by BNT (procured with help from TSMC, our most trusted institution); others preferred the locally developed Medigen; others only Moderna. We turned “anti‑vax vs pro‑vax” polarization into a healthy “my soccer team vs yours” competition that kept attention on vaccination. There was little oxygen for anti‑vax politics in Taiwan. But again, targets were set by CECC; civic tech didn’t propose alternatives to vaccination—because legitimacy was high.
…This is not a new idea; the “Clean Network” was formally announced around 2020 and essentially stated a preference for trusted technology partners at every link in the supply chain. At the time, it was mainly focused on the telecom industry, but the concept has now extended to many other areas, such as drones.
…They have a Taiwan trustworthy add-on engine, and we took the collective input from Taipei City for them to tune constitutionally their AI model.
…But I still think the co-op idea the principle, the main principles are still very instructive in the sense that when people collaboratively own the modes of production, and I mean the production of data, including collection and curation and processing, then people tend to think in a human-relation kind of way, and data collection becomes the beginning of a trustworthy relationship.
…So by building this kind of exoskeleton of troll hugging and practicing Aikido for more than, I guess, 100 collaborative meetings this way, inviting trolls in through the cracks, letting the light in, then we rebuilt the trust . The approval rate was at 9% in 2014, but by the time we first met in 2020, it was over 70%.
…Our science minister is training now, as we speak, the trustworthy AI dialogue engine, which is a language model specifically trained not to be agentic or self-conscious, or frontier, or AGI, or whatever, but for the very simple need of translating between the national languages of Taiwan and summarizing them and so on, like everyday tasks.
…So the most important thing here is to make sure the tools are trustworthy , they’re repeatable, and they do not hallucinate in their social cultural context.
…So, what we’re doing, essentially, is to make sure that in the actor-behavior-content model, instead of looking at the content, which is going to be impossible with generative AI anyway, we go back to the accurate model and make sure that trustworthy actors are easily identified as such.
…And with just a few seconds of voice print, now AI can very meaningfully synthesize the acoustic model so that over the phone you never know who’s actually speaking there without a code word or zero trust fingerprinting digital signature or some kind. And this is a large threat vector. It means social engineering will probably always succeed if people are not aware of the voice cloning, and even video cloning now, capabilities is caught off guard and aware.
…And really, it’s very toxic for the online trust , because not only are those celebrities unhappy about it, it also makes this whole web3 crypto thing look like a scamming only activity. So, it’s one of the top threats that we detected.
I don’t know if that was the best place to out, but also because there was no trust on either side, that that would be a credible non-partisan arbiter, and the person or people running that would play that role. This is maybe one example. But this is it’s not necessarily a question other than it’s an observation, but you created something here which I think is very unique, and I really hope we can replicate it in our system so that we can have these conversations and not think, Hey…
If people see that, in order to deploy AI in a believable, trustworthy fashion, that inspectability really is very important, then I do feel that it is possible instead of just running all the leading AI on supercomputing data centers. What Jensen Huang said is a personal supercomputer becoming a viable alternative so that people who run those GPUs in smaller Mac mini-like configurations can then run those smaller models, and even if there are different architectural choices, even if there are trap doors in models such as DeepSeek, then people, the tinkerers, as…
