New research from Cornell Tech shows that biased AI writing assistants can subtly shift users’ opinions on social issues—even when people are warned that the AI may be influencing them. (Source: Image by RR)

Participants’ Views Moved Toward the AI’s Suggested Arguments

New research from Cornell Tech suggests that AI-powered writing assistants may influence not only how people write, but also how they think. In two large experiments involving more than 2,500 participants, researchers found that autocomplete suggestions generated by biased AI systems nudged users’ opinions toward the positions embedded in the AI’s suggestions.

Participants, as noted in techxplore.com, were asked to write essays on controversial societal issues such as the death penalty, fracking, genetically modified organisms and voting rights for felons. Some participants received autocomplete suggestions from a writing assistant programmed with a specific bias, while others did not. Surveys conducted before and after the writing tasks revealed that those who used the biased AI assistant tended to shift their views closer to the AI’s stance on the issue.

Surprisingly, the influence remained even when participants were explicitly warned that the AI might be biased. Researchers tested several mitigation strategies, including informing users before and after the exercise about the bias, but none significantly reduced the effect. Even when participants knew the AI might attempt to influence them, their attitudes still shifted toward the suggestions they received.

The findings raise broader concerns about the psychological influence of AI writing tools that are increasingly embedded in everyday platforms such as email clients and productivity software. Researchers warn that biased AI systems — whether intentionally designed or inadvertently trained — could shape public opinion over time by subtly guiding how people express their thoughts and arguments.

read more at techxplore.com