AI-Generated Opinion Polls: A New Frontier or a Threat to Democracy?

As the June local elections loom, a new technological frontier is emerging in the realm of public opinion: artificial intelligence. While traditional opinion polls have long been a barometer of public sentiment, their reliability has often been questioned due to issues like low response rates. Now, 'silicon sampling' offers a novel approach, using large language models to simulate responses based on detailed demographic profiles. This method, while innovative for market research, raises profound questions for the political landscape.

“

You are a 40s working woman, college graduate, living in Seoul, middle class.

The core of this AI-driven polling lies in its ability to create 'AI personas' – virtual individuals whose responses are generated based on vast datasets. This allows for rapid simulation of public reaction to various scenarios. However, this very efficiency raises concerns. Critics point out that AI models may struggle to capture the nuances of public opinion, particularly from marginalized groups or those who do not actively participate in online discourse. There's a risk that AI might oversimplify complex issues or default to socially acceptable answers, thus failing to represent the true diversity of thought.

“

The large language model infers how a person with that profile would respond based on patterns learned from its training data.

Furthermore, the fundamental nature of AI polling – generating responses rather than reflecting them – presents a significant challenge to democratic principles. In a democracy, public opinion is meant to inform policy and governance. When opinion itself is simulated, the line between genuine public will and artificial output blurs. While current election laws in South Korea may restrict the public announcement of AI-generated polls, the possibility of their use in internal campaign surveys remains. This could lead to a situation where manufactured 'public opinion' influences political campaigns and, consequently, election outcomes, without the transparency and accountability expected in democratic processes. The Hankyoreh, as a publication committed to robust public discourse, views this development with significant concern, urging careful consideration of its implications for the integrity of our democratic institutions.

“

The large language model cannot capture voters who do not voice their opinions online, and tends to process the opinions of minority or peripheral groups with stereotypes, and tends to avoid controversial answers.