Artificial intelligence is increasingly being used to generate fake survey responses that imitate human opinions – raising concerns among researchers about whether public opinion polling can still be trusted.
The practice, known as “synthetic surveys” or “silicon sampling,” involves using AI systems such as OpenAI’s ChatGPT to simulate thousands of human responses to political, social and commercial surveys.
According to analysis published by The Conversation, the approach is spreading partly because traditional polling has become increasingly expensive and difficult.
Polling costs rising as response rates fall
Researchers say fewer people now respond to surveys and polls, forcing polling firms to spend more money contacting larger numbers of potential participants.
The article notes that a typical 10-minute survey involving 1,000 human respondents can cost tens of thousands of dollars.
AI-generated responses are dramatically cheaper.
Researchers can create “synthetic respondents” by giving AI systems instructions such as:
- age;
- political beliefs;
- education level; and
- geographic background.
The AI model then generates answers designed to imitate how a real person with those characteristics might respond.
Because large language models include built-in randomness, repeated prompts can produce thousands of slightly different answers.
Critics say simulations are not real opinions
Some researchers warn that synthetic survey systems may create the illusion of public opinion without actually measuring what real people believe.
The article compares the problem to using AI to estimate someone’s body temperature instead of measuring it directly with a thermometer.
“Researchers who poll AI systems instead of people are not measuring public opinion,” the article stated. “They are only simulating it.”
Critics also warn that AI systems inherit biases from their training data and may distort the views of groups underrepresented online.
Unlike traditional polling methods, many commercial AI systems operate as proprietary “black boxes”, making it difficult for outside researchers to fully understand how answers are generated.
AI still has useful roles in survey research
Despite the concerns, researchers say artificial intelligence can still improve parts of the survey process.
The article notes that AI tools may help:
- simplify confusing questions;
- reduce repetition;
- translate surveys across languages;
- summarize open-ended answers; and
- analyze large volumes of responses more efficiently.
Some researchers are experimenting with hybrid systems that combine smaller real-world surveys with AI-assisted analysis.
The article argues that the key issue is transparency.
Using AI to assist survey design may be acceptable, researchers say, but replacing human respondents entirely risks weakening the connection between decision-makers and the public.
Bigger concerns about trust and democracy
Polling and survey research play an important role in modern politics, business, healthcare and public policy.
Governments, companies and researchers often rely on surveys to understand public attitudes before making major decisions.
The article warns that if AI-generated responses become widespread without clear disclosure, public trust in polls and social research could weaken further.
At the same time, researchers acknowledge that declining participation in traditional surveys presents a real challenge.
The article concludes that AI will likely become more deeply integrated into polling and social research – but argues that human voices should remain central to the process.
Leave a Reply