On the Conversational Persuasiveness of GPT-4

On the conversational persuasiveness of GPT-4

Early work has found that large language models (LLMs) can generate persuasive content. However, evidence on whether they can also personalize arguments to individual attributes remains limited, despite being crucial for assessing misuse. This preregistered study examines AI-driven persuasion in a controlled setting, where participants engaged in short multiround debates. Participants were randomly assigned to 1 of 12 conditions in a 2 × 2 × 3 d…

Artificial intelligence already argues more convincingly than humans. According to a new study by Swiss researchers, the AI model GPT-4 works more coherently than human discussion partners in online debates in 64.4 percent of cases. However, only if it has personal data of the interlocutors such as age, educational status, gender or political attitudes. Without this information, the persuasiveness of AI is at the same level as that of its human …

One study showed that the great models of artificial intelligence language can be 64% more persuasive than people in online discussions – if they had been told about you.

GPT-4 Outsmarts Humans in Debates With Basic Personal Data

A new study has found that artificial intelligence (AI) can become much more convincing in arguments when it knows a few details about the person it is talking to.

Research.

Unlike humans, who need time and effort to adapt their arguments in a debate, LLM models transform their message instantly and on a massive scale. This gives them a disproportionate advantage in political campaigns.