AI’s built-in sycophancy means it is predisposed to agree with users, which can fuel hypochondriac tendencies — or cyberchondria, as it’s been dubbed. “The danger about LLMs is that they are architecturally biased to ‘know the answer’ and be helpful,” Lahav says. “Most general AI solutions are literally programmed to explain how a tingling arm could be multiple sclerosis, rather than telling you it’s likely nothing. A human doctor’s most valuable tool is the ability to say, ‘You seem fine, let’s just monitor this; go home and sleep.’ AI lacks that clinical instinct.”
