LONDON: Artificial intelligence (AI) impersonations of political debaters now come across as more "authentic, relevant and coherent” than the real thing, according to a new study.
A team led by Steffen Herbold of the University of Passau tested around 900 people in Britain with a mix of real and AI-generated versions of politicians speaking on British TV.
"Overall, participants rated the AI-generated responses as more authentic, coherent and relevant than the real ones, with the differences statistically significant in every comparison,” the researchers said, summing up their findings, published in the scientific journal PLOS One in July.
The results of the research should serve as a warning of the "enormous misinformation potential of AI,” according to Herbold.
At the same time, the findings show that people want more transparency around the technology – not only clear markers showing that material is based on AI but information on how the bot in question was "trained,” or what information or sources were used to inform the machine as it generated its output.
The team said they used GPT-4 Turbo to come up with "impersonated responses” to audience questions thrown at 112 "public figures” who have appeared on the show and whose Wikipedia pages were fed into the chatbot as a prompt.
One of the reasons that people were taken in by the AI responses, it seems, is that they stuck more closely to and more effectively addressed the question than the politicians did.
"The overlap between the question and the response text is significantly higher in generated responses, indicating that the panel members do not always address the question directly," the researchers wrote. – dpa
