It’s a speech note from ENCATC Congress 2023 (European network on cultural management and policy) Education and Research Session, Main Track. The citation for the license is CC 4.0 BY.

by Yen-Lin (mashbean) Huang, M.D.

Taiwan has two big problems related to its position in the world. One is about language, and the other is about public discussions. Even though Taiwan and China are close neighbors, our language and culture have become quite different over the last 80 years. We use Traditional Chinese, but China uses Simplified Chinese. We are a democratic sovereign state, but China is not. Because of this, some things are not written in Simplified Chinese. However, China has invested significant resources to develop open-source Language Models, making AI Models containers for their ideology. This has become an established fact.

Recently, there was a new about AI from Taiwan's leading research center, Academia Sinica. The Chinese Knowledge and Information Processing group, known for its work on cultural preservation and digitalization, made an AI model for Traditional Chinese, but people found out it sometimes gave answers favoring China. This happened because they used open-source LLM from Simplified Chinese due to a limited budget. As a result, they took down the model for now and said they would be more careful in the future.

Creating a language model requires tremendous funding. It’s a task only centralized governments and big tech can undertake. The cultural gap between Traditional Chinese and Simplified Chinese will likely narrow in the future, but not in a direction that Taiwan would prefer.

The second problem is Public Sphere Dysfunction. Taiwan is dealing with Information manipulation and how it's used as a weapon. We have an open online space for discourse, but China has been influencing stories for a long time. Also, big social platforms that favor China are now common for students. Policymakers find it hard to adapt to these fast-changing online discussion areas.

We are embracing the awkward dilemma. When discussing National security, there's an easy risk of falling into Censorship of speech. When talking about Freedom of speech, it's hard to sidestep Information manipulation. Generative AI only acts as an amplifier, intensifying conflicts. As the saying goes, "As righteousness rises an inch, evil rises a foot."

Biologist E. O. Wilson once said, "We have Paleolithic emotions, medieval institutions, and godlike technology." Currently, Generative AI embodies this godlike technology.

Earlier this year, a group of AI experts and public figures issued a Statement on AI Risk. They believe that "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

We believe that new technologies like LLM, which could shake up current democratic systems, should be adopted sooner by democratic groups. This early use can help democracies spot their weak points and adjust better, stopping autocracies from getting ahead.

To discuss how to have effective, high-quality discussions in the Digital Public Sphere, such as on Social Platforms, we must address its underlying foundation, like Digital Public Goods. For instance, Open Source LLM is a vital public good. We need to establish an open, fair, and transparent mechanism for AI development and use to ensure everyone can access AI equally. At the core is the Digital Public Infrastructure. Taking AI as an example, data governance, especially data related to Taiwanese culture, is crucial. Without data, there can be no LLM specific to Taiwan.

In Taiwan, we are no strangers to thinking outside the box and reaching a consensus on various issues. The public must participate in policymaking to strengthen unity. Such an investment in the democratic process is yielding significant returns by fostering shared values and resilience against authoritarian expansionism. We refer to this as the Taiwan method.

Using AI as an example, we should categorize it into three pillars: Education, Decentralisation, and Deliberation. Through Education, we aim to strengthen digital literacy, ensuring everyone has experience using AI and can make independent judgments. Through Decentralisation, our goal is for everyone to run LLM on their personal computers. Only in this way can AI truly serve as a universal service, overcoming biases, and ensuring that personal data isn't uploaded to the cloud. Lastly, through Deliberation, we are exploring methods of democratic alignment and partnering with international organizations, seeking an AI Constitution that aligns with our own.

1 - Education

In education, I once participated in a discussion meeting with the Ministry of Education, deliberating on how emerging technologies like AI could be integrated into the 12-Year Basic Education Curriculum Guidelines. I also took part in local experimental education forums in Taiwan, discussing how alternative educational systems could utilize AI effectively. Our conclusion was straightforward: questions that teachers can easily answer might no longer be the most valuable. Instead, we need to enhance students' critical thinking and practical skills. Hence, digital literacy is a specialized field, and we must intentionally maintain the digital public sphere.

2 - Decentralization

In terms of Decentralization, taking generative AI for images as an example, we are currently witnessing a battle between Proprietary Software and Open Source Software. Engineers might be able to design fantastic AI, but they can't singlehandedly address all issues between humans and artificial intelligence. What can open-source AI achieve that closed AI cannot?

Let me share a case. Have you ever seen the animation "Dragon Ball"? There was a fan who loved "Dragon Ball" so much that he fine-tuned a model to produce images of the Nimbus cloud, a recognizable feature from the show. He even uploaded it online for other Dragon Ball enthusiasts to download. We can't eliminate biases, just as we can't prevent someone from loving "Dragon Ball." Even less so can we stop someone from creating an AI that produces the Nimbus cloud from "Dragon Ball" and then sharing it with fellow fans. Such customized images can only be achieved when run on personal computers, combined with an open-source AI model, and further augmented with various small-scale, custom-made models. This embodies the concept of plurality. Our goal isn't to erase biases, but to allow more individuals to create diverse and heterogeneous algorithms, contributing to this universe.

3 - Deliberation

The last pillar is deliberations. We employed Polis to discuss AI, an open-source tool for online deliberation, to collate viewpoints from Ideathon attendees regarding the future course of AI. Polis has a precedent of being utilized for legislating on matters related to emerging technologies in Taiwan, such as Shared Economy like Uber in the past. Post-voting, this platform facilitates the formation of distinct opinion clusters, assisting participants in identifying how their perspectives differ from the broader consensus.

We have two purposes when it comes to deliberation: enhancing local digital citizen engagement and forging international partnerships for collaborative progress. For this year, AI Democratization stands as a pivotal and highly relative pilot initiative. In collaboration with the Collective Intelligence Project, we have initiated Alignment Assemblies in Taiwan and staged our pilot project at the Ideathon, a hackathon designed to ignite the imagination.

In addition, we have taken the initiative to upgrade to Polis 2.0. In alliance with the AI Objectives Institute, we have deployed their ‘Talk to the City’ utility to visually map the seed statements and voting results into varied thematic clusters. These clusters offer an interface for users to engage with and discern the ‘collective wisdom’ derived from the deliberative process. You can discuss with the virtual collective opinion leader through Talk to the City with the assistance of LLM.

To round off, Moda has launched on-site AI Democratization workshops. By partnering with specialists in deliberative democracy, we aim to evolve the discourse further, leading to a result that encapsulates both raw data and conclusive findings.

Upon authorization from participants, this treasure trove of deliberative data will be shared publicly online. We extend an invitation to global democratic networks—including AI policymakers and AI model developers —to utilize these insights. As Taiwan serves as a fertile testing ground for AI Democratization, we welcome nations worldwide to leverage our collective intelligence as we cautiously, yet optimistically, navigate the future.

Lastly, let's discuss Taiwan's Law and Policy. According to the theory of American legal scholar Lawrence Lessig’s Pathetic Dot theory, Code, Laws, and Markets shape our digital society. However, we still have the opportunity, through Digital Democracy and especially the Deliberate process, to build Collective Intelligence, emerging as a fourth force shaping society—Norms.

In 2019, Taiwan passed the "Cultural Fundamental Act." The section related to Cultural Technology aims to include comprehensive digital governance and enhance citizens' digital participation. However, achieving this objective within the Ministry of Culture is challenging. The potential reason might be the lack of a collaboration mechanism between those familiar with emergent technology and the bureaucrats responsible for cultural policy. What we need most at this moment is a platform or system that can bridge and facilitate cooperation between the two.

Conclusion

First, we discussed the threats Taiwan faces from China. Taiwan's Geopolitical Crisis includes a vulnerable culture and public sphere dysfunction, with AI evidently exacerbated this situation. However, we still have a turning point. The solution lies in the Taiwan model and international cooperation, encompassing Education, decentralized AI, and deliberation. The next step might be the legal framework for cultural technology, enabling multi-stakeholders to collaborate and fostering pro-democracy technological development.