AI responses in political debates seen as more real than human ones

Artificial intelligence can now convincingly mimic public figures in political debates, with audiences finding AI-generated responses more authentic and relevant than those from actual participants. A study by researchers at the University of Passau in Germany, published in Plos One, used GPT-4 Turbo to create fictional answers to questions posed in political debates. The AI was fed biographies of 112 public figures to help it imitate their style.

A significant majority of 948 UK adults surveyed rated the AI's responses as more authentic, coherent, and relevant than the real participants' answers. In some cases, the AI's response directly addressed the question while the human participant's did not, or they expressed entirely different viewpoints. This finding raises serious concerns about the potential for AI to be weaponized in disinformation campaigns targeting specific individuals.

“

The study conclusively demonstrates that humans consider AI-generated debate content to be more authentic than what these well-known public figures actually said.

"The study conclusively demonstrates that humans consider AI-generated debate content to be more authentic than what these well-known public figures actually said," stated Steffen Herbold, a co-author of the article. He warned that this "reveals the enormous disinformation potential of AI," urging society to be aware and critically evaluate all written information to prevent the uncontrolled spread of AI-generated falsehoods. Herbold also noted a strong public desire for transparency, with people wanting to know when AI has been used and how it was trained.

The researchers cautioned that their findings are based on a single debate format from one country and a single AI model, so they may not apply to all contexts. However, the study's implications for the integrity of public discourse and the need for robust detection and critical evaluation of AI-generated content are clear.

“

This reveals the enormous disinformation potential of AI, of which society must be aware to critically evaluate any written information and prevent the uncontrolled spread of AI-generated disinformation.