The many capabilities of artificial intelligence already extend to translation, art, coding and even inspiration, making it indispensable in numerous sectors. Yet deepfakes, where AI replicates faces and voices, erode trust. These are weaponised by authoritarian governments and cybercriminals, posing a serious challenge to our free and open societies.

A democratic approach to AI governance, not a technocratic one, is the best answer to what is an ethical and political conundrum of global proportions. AI should be run not by big tech, but through an inclusive, democratic model.

This requires broad-based participation to reach a sustainable consensus. There are no shortcuts; instead, it takes the involvement of large populations and diverse groups around the world.

The major generative AI models are each trained by a single organisation. This means that the biases of the creators are unintentionally echoed. The diverse cultural nuances of a global audience are often neglected or mishandled.

We need to move away from placing all our eggs in one basket of developers and open up opportunities for the people to participate freely and collaboratively. This brings us closer to a globally trusted AI assurance framework, with shared ethical standards and principles of use.

Fortunately, a roadmap for democratisation is emerging through the work of the Collective Intelligence Project ^[1], of which Taiwan’s Ministry of Digital Affairs is a part, along with industry players such as OpenAI and Anthropic.

Together, we launched Alignment Assemblies ^[2] to determine how societal input can shape the development of AI development in a way that better reflects the cherished shared values of the democratic world.

Already, the project has partnered with Anthropic ^[3] to train an AI large-language model on written principles—similar to a constitution—that were collectively designed by the people and mirrors more fully common expectations and hopes for the technology.

Researchers assembled a representative sample of 1,000 Americans drawn from various age, gender, income and geographic categories, to draft a constitution for an AI system. Constitutional AI is a method for aligning AI language models with high-level normative principles written into a constitution, comparable to documents such as the UN Universal Declaration of Human Rights.

The process was powered by Polis ^[4]—an open-source system for gathering, analysing and understanding what large groups of people think in their own words, enabled by advanced statistics and machine learning.

Participants gave their views on statements such as ‘AI should prioritise the needs of marginalised communities’ and ‘AI should prioritise the interests of the collective or common good over individual preferences or rights’.

Researchers then took the results and produced a constitution that was used to train a new AI model, which proved to be less biased than the original while being equally capable.

Taiwan is doing its own work in this area. In July, my ministry collaborated with Polis to explore topics on democratising AI futures. The results will soon be published as open data, facilitating the training of AI language models aligned with Taiwan’s distinct experiences.

Using this approach, the people get to contribute to and better understand the behavioural rules of the AI with which they will interact. This gives all and sundry a chance to help chart a safe and sustainable course of development for AI, as well as identify and mitigate related risks.

There are two sides to every coin. Certain emerging technologies bolster autocracies, but they can also revitalise democracy. How best to combat the pervasiveness of online harms? The solution lies in plurality, or technologies for collaborative diversity, to increase the bandwidth of democracy.

I envision the future of AI as starting with Alignment Assemblies. It is one in which the hopes, dreams and potential of all humankind are realised. This is because we are capitalising on a rich tapestry of diverse experiences—the wisdom of crowds—accumulated over thousands of years. Such an approach can surely shape AI into a partner dedicated to societal wellbeing.

As a responsible member of the international community, Taiwan is at the forefront of collaborative efforts ^[5] to deliver digital resilience for all. This is my ministry’s core mission, one in which we strengthen societal resilience by empowering civil society ^[6] through connecting with democratic networks.

Taiwan will keep building consensus with like-minded partners to ensure that global AI development is a race to safety, rather than a race to power, enabling continuous leadership in evaluating the most advanced AI models and jointly constructing a secure and stable ^[7] AI environment.

The truth of the matter is that democracies must confront formidable challenges from autocracies undermining the delicate equilibrium between societal wellbeing and individual freedoms. Our efforts need to go beyond piecemeal regulatory frameworks. We envision a trajectory harnessing the irresistible force of participation, progress and safety for the collective good.

When people from all segments of society come together constructively in the digital realm, they form spaces of progress and prosperity in which all voices are heard and respected. Only by bridging ideological divides and utilising farsighted initiatives like Alignment Assemblies can we deliver sound global governance of AI and ensure that we leave no one behind. The tools to effect real and lasting change are in our hands. Let’s free the future—together.