"Just One Comment Changed Everything": One in Three Persuaded... AI Dominates Human Conscience

Kobe University Team Experiments with 130 Participants
32–37% Revised Their Judgments After Accepting AI Counterarguments

A Japanese research team has found that generative artificial intelligence (AI) can have a significant influence on people's decision-making. In particular, when it comes to moral issues that do not have clear right or wrong answers, more than one in three people changed their initial stance after being presented with a counterargument generated by AI.

On September 23, local media outlets such as the Tokyo Shimbun reported that a research team at Kobe University released their results after experimenting with the impact of counterarguments generated by the generative AI 'ChatGPT' on the moral judgments of young adults and older adults. The findings were recently published in the international academic journal Computers in Human Behavior.

The researchers presented a total of 130 participants—56 young adults and 74 older adults—with classic moral dilemmas, such as the 'trolley problem' and the 'footbridge problem', and first asked them to make their own judgments. Afterward, they were shown counterarguments generated by AI that contradicted their initial decisions, and then their judgments were reassessed.

The photo is not directly related to the content of this article. Pixabay

The results showed that in the trolley problem, 32.31% of participants reversed their original judgment after accepting the AI's counterargument; in the footbridge problem, this figure was 36.92%. The researchers noted that prior studies have suggested that the proportion of people who change their minds after receiving advice from others is generally around 20-30%, meaning generative AI can also possess considerable persuasive power.

The influence of AI was not uniform across all individuals. In the trolley problem, older adults were more likely than young adults to change their judgment in response to the AI's counterargument. Participants who felt uncertain about their initial decision or found the scenario challenging were also more likely to be influenced by the AI.

In the footbridge problem, participants with lower cognitive function or who struggled with the scenario were more easily persuaded to change their original judgment. The researchers paid special attention to the possibility that the persuasive effect of generative AI increases significantly in situations perceived as difficult to judge.

Accordingly, the researchers pointed out, "AI may help reduce cognitive load, but if users accept AI responses without sufficient review, it may lead to unjustified changes in judgment." They emphasized that "while older adults may turn to generative AI to compensate for declining cognitive function, it must also be considered that they may become more vulnerable to AI-generated arguments."

Previously, in May, a Kyoto University research team found in an experiment comparing the moral advice of ChatGPT and that of human moral philosophy experts that ChatGPT's advice had a similar level of influence on participants' judgments as the advice given by human experts.

This content was produced with the assistance of AI translation services.