AI Replaces Real People in Opinion Polls

A new method termed silicon sampling is emerging, where artificial intelligence agents simulate survey responses instead of gathering opinions from real individuals. This practice, exemplified by an AI poll showing a majority trusting doctors and nurses, offers a cost-effective alternative to traditional phone or web polling by removing the human element entirely.

Historically, opinion polls served to gauge public sentiment and guide policy. However, polling has faced implementation challenges, leading to reliance on statistical models that rebalance data based on pollster assumptions. This modeling introduces biases, as demonstrated by differing results from the same data set given to multiple pollsters.

Silicon sampling exacerbates these issues by using AI trained on past data to predict current and future behavior. Critics argue this focus on prediction distorts the fundamental purpose of polling, which is to capture current public opinion. There is concern that this method may produce results even more heavily skewed than traditional polls.

Companies are investing heavily in silicon sampling, developing digital twins or virtual representations of respondents for market research and other applications. If unchecked, this trend could lead to a significant decline in trust for public opinion work and social science research, potentially causing essential knowledge about society to become artificial.