There is considerable promise and hype for the impact of artificial intelligence in medicine. Three large observational studies indicate that there are approximately 12 million American adults per year who are misdiagnosed. This serious shortcoming could certainly be improved.
If all data for each patient were available for "deep learning" — a subtype of artificial intelligence using neural networks — it would be possible to provide a more complete differential diagnosis with weighting (probability estimates). AI could also produce an up-to-date survey of the complete medical literature, and offer decision support for the interpretation and ordering of diagnostic tests.
Although the potential certainly exists, this capability hasn't yet been adequately demonstrated, and a major shortcoming is that all of each individual's medical data is frequently not available or aggregated.
By taking on pattern recognition tasks and many rote functions, it is possible that artificial intelligence will ultimately free up physicians to spend more time with their patients. This lack of time has not only led to deterioration of the patient-doctor relationship, but also plays a role in the misdiagnosis problem.
The bottom line: Ironically, more use of technology in the future might improve human touch.
Other voices in the conversation: