BOSTON — A generation of physicians is entering practice with virtually no ability to diagnose illnesses without artificial intelligence, according to program directors at major teaching hospitals who describe the trend as "concerning but inevitable."

The shift has occurred rapidly. Medical students who began training in 2023 now demonstrate profound deficiency in fundamental clinical skills—listening to heart sounds, palpating abdomens, interpreting physical symptoms—having spent their residencies primarily typing patient data into diagnostic AI and implementing its recommendations.

"They're excellent at prompt engineering," said Dr. Barbara Kellerman, Chief of Medicine at Johns Hopkins. "But several residents appear confused when I suggest they actually touch the patient."

Young physicians defend the approach. "Why would I waste time with a stethoscope when the AI can analyze an ECG more accurately?" asked Dr. James Reeves, a first-year internal medicine resident. "Physical examination is a legacy skill. Like cursive writing."

The consequences emerged sharply last month when a widespread AWS outage disabled diagnostic AI systems across the Northeast. Emergency departments reported that attending physicians under 35 largely stood paralyzed, unsure how to proceed without algorithmic guidance. Several patients were instructed to "wait until the system comes back online" for treatment of clearly visible fractures.

Medical educators acknowledge the problem but consider it unsolvable. "Students simply won't learn skills they perceive as obsolete," explained Dr. Kellerman. "And they're not entirely wrong. When the AI is available, human clinical judgment is mostly just expensive theater."

Insurance companies have enthusiastically endorsed AI-dependent practice, noting that algorithmic recommendations are easier to audit for billing purposes and less likely to deviate from cost-effective care pathways.

Several teaching hospitals have discontinued courses in physical diagnosis entirely, replacing them with seminars on "AI-Augmented Clinical Reasoning" and "Effective Prompt Design for Differential Diagnosis."