I admit it. I’m not a big fan of AI. I crave the “good old days” of social media, when my feed wasn’t curated and I knew which reels were, in fact, real. As a college professor, I watched ChatGPT undermine students’ ability to distinguish between reputable and dubious sources. As a consumer, I want live customer service agents who understand me as a human being.

All that said, AI is really good at some tasks we humans struggle with—especially complex tasks requiring us to synthesize and weigh information from multiple sources before making our decisions—the very skills needed for precise diagnosis.

Scientists have known for decades that biases in human thinking compromise our ability to diagnose—even those of us with extensive experience and great confidence in our decisions. More than four decades ago, the first desktop PCs were better at diagnosing than seasoned clinicians who had the same information. AI only extends the advantage.

Precise diagnosis requires us to thoroughly assess and weigh dozens of symptoms, many spanning several disorders, all in the contexts of family history, life circumstances, and past life events. Our brains aren’t built for this, but AI is, and it matters. Mistaken diagnoses, which we often stick with once they’re made, cause treatment delays, use of poorly matched treatments, and, for some, real harm.

Bipolar disorder, though not unique in this way, provides an instructive example. Diagnosing bipolar disorder is challenging because depression often emerges years before a first manic episode (a necessary criterion), and up to two-thirds of patients self-medicate with substances,........

© Psychology Today