Imagine you're trying to decide if a behavior—like drinking or skipping sleep—is risky for your health. You might ask a friend or an AI chatbot. New research suggests who you ask could change your answer more than you'd expect.
In two experiments with 60 people, researchers compared advice from AI algorithms (using GPT-4) to advice from human peer groups. They found that the AI systematically rated health behaviors as more dangerous than human advisors did. When people thought a threat was relatively low, they preferred getting advice from other humans. But when the stakes felt higher, that preference disappeared.
Interestingly, when the AI disagreed with someone's initial judgment, people were more likely to significantly change their minds than when a human disagreed. This suggests AI advice can be particularly persuasive in shifting beliefs, even if it's more cautious.
It's important to remember this was a controlled experiment about judgment, not a test of AI in real doctor's offices or clinics. The study measured how advice influences perceptions, not whether following AI or human advice leads to better health outcomes. The AI wasn't acting as a medical tool, but as another voice in the conversation.