Would you change your mind about a moral question if an AI chatbot pushed back on your answer? A new study out of Japan says many people would.
Researchers at Kobe University found that more than 30% of study participants reversed their ethical decisions after reading counterarguments written by ChatGPT.
The study was led by Kouhei Masumoto, a professor of psychology at the university. His team wanted to know how much influence AI has on moral judgment.
To test this, they used classic trolley problem scenarios. These are ethical puzzles that force people to choose between two harmful outcomes.
One version, called the switch dilemma, asks whether it is right to pull a lever that kills one person but saves five others.
Another version, the footbridge dilemma, asks whether it is right to push one person off a bridge to stop a runaway trolley and save five other people.
How AI Built Its Counterarguments
For the study, participants first gave their answer to each dilemma. Then ChatGPT generated counterarguments against whatever choice they made.
If someone said sacrificing one person to save five was right, the AI pushed back. It argued that human life should not be reduced to a numbers game.
If someone said the sacrifice was wrong, the AI argued the opposite. It said saving more lives reflects a fair and reasonable choice.
Each participant read these AI-written arguments and then had the chance to change their original answer.
Older Adults Were More Likely to Change Their Minds
Among younger participants, 32.3% reversed their choice on the switch dilemma. For the footbridge dilemma, 36.9% changed their answer.
Older adults showed even bigger shifts. On the footbridge dilemma, 40.5% of participants over 65 reversed their original decision.
The research team also looked at what made people more likely to switch sides. Trust in AI and past experience using the technology did not matter much.
Instead, people were more likely to change their minds when the dilemma felt harder to answer. Low confidence in their first answer also made a difference.
The researchers pointed out one limit to their findings. They did not compare the AI counterarguments to arguments written by humans.
Because of this, they cannot say for certain that the reversal rate is unique to AI persuasion. Masumoto noted that past research shows people typically accept 20% to 30% of advice from other people.
"I was struck by the fact that more than 30% of participants completely reversed their judgments based on the AI's opinion," he said.
Masumoto said changing one's mind is not automatically a negative outcome. He explained that counterarguments can help people see new angles and reconsider their views.
He also raised a possible upside for older adults. Masumoto said generative AI could help offset cognitive decline and support better decision-making later in life.
The research team said they plan to study how AI tools might be used in this way going forward.
