Artificial intelligence (AI) now poses a severe threat to the reliability of online public opinion polls, with new research demonstrating that bots can convincingly mimic human responses, evading detection with near-perfect accuracy. The implications extend beyond election interference, potentially poisoning scientific research that relies on survey data.
The Vulnerability Exposed
A Dartmouth University study published in the Proceedings of the National Academy of Sciences reveals that large language models (LLMs) can corrupt online surveys at scale. The core problem is that AI can now generate responses indistinguishable from those of real people, making it nearly impossible to identify automated interference.
To test this, researchers developed an “autonomous synthetic respondent” – a simple AI tool driven by a 500-word prompt. This tool was designed to simulate realistic human behavior, including plausible typing speeds, mouse movements, and even typos.
Near-Perfect Mimicry
The results were alarming: in over 43,000 tests, the AI fooled 99.8% of systems into believing it was a human participant. The bot bypassed standard safeguards like reCAPTCHA and even solved logic puzzles accurately.
This isn’t crude automation; the AI thinks through each question, acting like a careful, engaged respondent. The ability to generate believable responses in multiple languages (including Russian, Mandarin, and Korean) further amplifies the threat, allowing foreign actors to easily deploy manipulative campaigns.
Election Interference at Minimal Cost
The study highlighted the practical vulnerability of political polling, using the 2024 US presidential election as a case study. Researchers found that as few as 10 to 52 AI-generated responses could flip the predicted outcome of top-tier national polls during the crucial final week of campaigning.
The cost? A mere 5 US cents (4 euro cents) per response. This makes large-scale manipulation incredibly accessible, even to actors with limited resources. The ease with which AI can now skew polling data raises serious concerns about the integrity of democratic processes.
Broader Implications for Scientific Research
The threat extends far beyond elections. Thousands of peer-reviewed studies rely on survey data collected from online platforms. If this data is systematically tainted by bots, the entire knowledge ecosystem could be poisoned.
The study argues that the scientific community must urgently develop new, verifiable methods for collecting data that cannot be manipulated by advanced AI tools. The technology exists to verify real human participation, but the will to implement it remains a critical barrier.
The Need for Immediate Action
The findings underscore a fundamental weakness in our data infrastructure. The integrity of polling and the reliability of scientific research are now directly threatened by AI-driven manipulation.
Without immediate action, the credibility of public opinion data will erode, undermining democratic accountability and hindering scientific progress. The study concludes that implementing robust verification methods is not just desirable but essential to preserving the integrity of our data-driven world
