AI personas in polling: Stability vs. human voter volatility

Researchers have found that AI personas used in synthetic public opinion polling do not always mirror the changing opinions of real voters over time. Unlike human voters, whose opinions can shift between surveys, AI personas may maintain consistent responses depending on their design.

This discrepancy was highlighted in a study comparing AI personas and human voters. While human voters surveyed in October may provide different responses by January, AI personas might not exhibit the same level of opinion volatility.

The stability of AI personas in polling could potentially lead to inaccuracies in predicting real-world voter behavior. This is because AI personas, unlike human voters, do not experience the same external influences or changes in personal circumstances that can alter opinions.

Experts suggest that the design and programming of AI personas play a crucial role in determining their behavior in polling scenarios. If AI personas are programmed to be more adaptable, they might better reflect the dynamic nature of human voter opinions. However, this adaptability must be carefully calibrated to avoid introducing bias or inaccuracies.