Can healthcare artificial intelligence be more human than humans?
Google researchers say they've developed AI that is more accurate — and empathetic — than real-life physicians.
The Big Tech company created an experimental large language model, built on actual EHR data and medical appointment transcriptions, to conduct medical interviews, according to a Jan. 11 study on the preprint site arXiv.
Twenty actors posing as patients participated in online text-based medical consultations — and weren't told whether they were communicating with AI or real-life clinicians, per the study. The bot outperformed the human physicians in 24 of 26 conversational measures, including showing empathy, being polite, and putting patients at ease. AI also matched or surpassed the primary care physicians in all six diagnostic categories.
The researchers stressed the technology is only experimental and wasn't tested on actual patients. The study also has yet to be peer-reviewed.
"We want the results to be interpreted with caution and humility," study co-author Alan Karthikesalingam, MD, PhD, a clinical research scientist with Google Health, told Nature for a Jan. 12 story. "This in no way means that a language model is better than doctors in taking clinical history." For one, AI may have appeared so considerate because, unlike humans, it doesn't get tired, he noted.
But the researchers say the tool shows that AI has "significant potential" to transform the collection of clinical histories and diagnostic conversations in healthcare.