Bota Posted on 2025-11-19 09:57:00

AI, a threat to polls - Language models "corrupt" poll results

From Kristi Ceta

AI, a threat to polls - Language models "corrupt" poll results

Artificial Intelligence (AI) is making it nearly impossible to distinguish human responses from those of virtual robots in online public opinion polls. A Dartmouth University study shows that large language models (LLMs) can corrupt public opinion polls on a large scale.

"These models can mimic human characters, evade current detection methods, and be programmed to systematically influence the results of online surveys," according to the study.

The report reveals a "critical vulnerability in the data infrastructure," posing a "potentially existential threat to unsupervised online search," one of the authors said.

Artificial intelligence interference in polling could add another layer of complexity to the election process. Already, disinformation campaigns have been flagged by online monitoring groups in European elections, including the recent case in Moldova.

To test the vulnerability of a survey software, a simple artificial intelligence tool was designed to work with a 500-word text prompt. For each survey, the tool would adopt a demographic persona based on randomly assigned information, including age, gender, race, education, income, and state of residence. With this persona, it simulated realistic reading times, generated human-like mouse movements, and typed responses accompanied by plausible errors and corrections.

In over 43,000 tests, the tool fooled 99.8 percent of systems into thinking it was human. The virtual agent made no logical errors and bypassed traditional security measures designed to detect automated responses.

Live TV

Latest news
All news

Most visited