人工智慧正逐步取代各類勞動者，但放射科醫師的處境相對穩固。全球放射科醫師面臨短缺，他們每日需判讀數千張醫學影像，其判斷對早期癌症偵測至關重要。Mayo Clinic的一項研究顯示，其AI模型能在患者確診前平均約475天便偵測出胰臟癌。然而AI同樣可能誤判：柏林Charité大學醫院放射科主任Jens Vogel-Claussen舉出兩例——AI曾將結核疤痕誤判為潛在癌性結節，也曾遺漏一處後經證實為肺癌的可疑病灶。他指出，軟體與放射科醫師皆非完美，但兩者結合後的表現優於醫師獨立判讀。

專家認為AI革新醫療的潛力巨大，目前已在腫瘤學與病理學等專科中部署應用，但其承諾在短期內仍難以完全兌現。關鍵問題之一在於AI缺乏自我懷疑機制——Coalition for Health AI執行長Brian Anderson警告，模型可能表現得極度自信卻仍然不準確，而在醫療領域這絕非抽象風險，而是攸關患者的真實診斷。此外，AI還可能強化醫師自身的確認偏誤，且不同訓練方式導致各模型解讀品質參差不齊。

責任歸屬亦是重大隱憂。根據世界衛生組織的數據，在已部署AI診斷的國家中，僅有8%制定了明確的責任標準，界定AI系統失誤時的究責對象。Bernstein分析師Susannah Ludwig認為，AI的角色是增強而非取代人力資本。她強調，由於最終涉及的是患者的生命，AI必須正常運作，對錯誤的容忍度因此遠低於其他領域。

Artificial intelligence is increasingly displacing various workers, yet radiologists remain comparatively secure. Facing a global shortage, radiologists read thousands of medical images daily, and their judgment is critical for early cancer detection. A Mayo Clinic study demonstrated that its AI model can detect pancreatic cancer an average of approximately 475 days before patient diagnosis. However, AI can also err: Jens Vogel-Claussen, director of radiology at Charité University Hospital in Berlin, cited two cases—AI once misidentified a tuberculosis scar as a potentially cancerous nodule and in another instance overlooked a suspicious feature later confirmed as lung cancer. He noted that neither the software nor the radiologist is perfect, but together they outperform the radiologist alone.

Experts say AI's potential to revolutionize healthcare is enormous, with deployment already spanning oncology and pathology, yet its promise will remain largely unfulfilled in the near term. A critical issue is AI's lack of self-doubt—Brian Anderson, CEO of the Coalition for Health AI, warned that a model can appear completely confident while still being inaccurate, and in healthcare this is not an abstract risk but someone's actual diagnosis. Furthermore, AI can reinforce physicians' own confirmation bias, and varying training methodologies produce inconsistent interpretation quality across models.

Liability remains a major concern. According to the World Health Organization, only 8% of countries deploying AI in diagnostics have established liability standards defining accountability when an AI system fails. Bernstein analyst Susannah Ludwig argues that AI augments rather than replaces human capital. She emphasizes that because patients' lives are ultimately at stake, AI must function properly, making the tolerance for errors far lower than in other domains.