醫療AI評測：醫師用量衝三倍，疑慮也跟著翻倍 | Medical AI Review: Doctor Usage Triples, So Do the Doubts

By Kit 小克 | AI Tool Observer | 2026-09-23

🇹🇼 醫療AI評測：醫師用量衝三倍，疑慮也跟著翻倍

醫療AI今年用量衝上新高，但醫師心裡的問號也跟著變多。Wolters Kluwer Health在2026年調查超過350位美國醫師與護理師後發現，每週固定使用AI的醫師比例從38%衝到74%，護理師也從46%成長到70%；不過同一份調查裡，74%的人擔心自己因此「去技能化」，75%擔心AI編造內容，只有27%的人搞得清楚自己單位的AI使用規範。這篇文章把數字攤開來看，順便講講醫療AI現在到底能不能信。

醫療AI現在多好用？醫師用量成長多少？

根據調查，74%醫師、70%護理師每週至少使用一次醫療AI，較去年的38%與46%幾乎翻倍；天天用好幾次的醫師比例也衝到38%。最常見的用法是整理醫學文獻、分析數據（超過五成醫師），另外44%醫師用AI書記工具（AI scribe）幫忙寫病歷。

74%醫師、70%護理師每週至少用一次AI（去年38%／46%）
38%醫師、32%護理師天天用好幾次
只有9%醫師從沒用過AI工具

醫師在怕什麼？「去技能化」是什麼意思？

「去技能化」指的是醫師太依賴AI，反而讓自己的臨床判斷力退步——這正是74%受訪醫師最擔心的事。Wolters Kluwer首席醫療長Peter Bonis直言，過度依賴AI可能讓醫師「學錯技能，或失去原本就有的能力」。

醫療AI會產生幻覺嗎？醫師怎麼抓出錯誤？

會，而且高達75%的醫師把「AI幻覺」（編造不存在的內容）列為主要疑慮；雖然73%自認抓得出錯誤，換句話說仍有四分之一的人沒把握。實務上，77%的醫師會拿PubMed等可信來源交叉查證AI給的答案，53%則希望AI能說明自己的推理過程。

醫院有訂出醫療AI使用規範嗎？

大多數還沒有。只有27%的醫師清楚自己單位的AI治理政策，比2025年的21%只微幅進步；僅35%知道該怎麼驗證AI答案的準確性，22%的人有明確的「AI出錯算誰的責任」規範。換句話說，醫院用醫療AI的速度，遠遠超過訂規矩的速度。

Kit小克怎麼看

醫療AI在「輔助」型工作——讀片、整理文獻、寫病歷草稿——已經證明好用，這也是它用量能在一年內翻倍成長的原因。但一旦跨進治療建議、病患溝通這類需要臨床判斷的領域，證據還沒跟上炒作的速度，連天天在用AI的醫師自己都比較保守。如果你是醫療從業者，AI可以當助手幫你省時間，但別把它當成免責的擋箭牌——最後簽名的還是你。

好不好用，試了才知道。

🇺🇸 Medical AI Review: Doctor Usage Triples, So Do the Doubts

Medical AI usage among doctors has nearly doubled this year — and so has their list of concerns. A 2026 survey of over 350 US doctors and nurses by Wolters Kluwer Health found that weekly AI use among physicians jumped from 38% to 74%, while 74% also flagged "deskilling" as a top risk and only 27% understood their organization's AI governance policy. Here's what the numbers actually say about whether medical AI is ready for more than reading scans.

How Widely Is Medical AI Actually Used Now?

74% of doctors and 70% of nurses now use medical AI tools at least weekly, up from 38% and 46% a year ago. 38% of doctors use it multiple times a day. The top uses are summarizing medical literature and analyzing data (over 50% of doctors), and AI scribes for documentation (44%).

74% of doctors / 70% of nurses use AI weekly (up from 38% / 46%)
38% of doctors use AI multiple times daily
Only 9% of doctors have never touched an AI tool

What Are Doctors Worried About? What Is "Deskilling"?

Deskilling means relying on AI so much that clinical judgment atrophies — and it's the top concern for 74% of surveyed clinicians. Wolters Kluwer's chief medical officer Dr. Peter Bonis warned that overreliance risks clinicians "learning tasks incorrectly or losing abilities they already had."

Does Medical AI Hallucinate? How Do Doctors Catch Errors?

Yes — 75% of clinicians cited AI hallucinations as a major concern, and while 73% feel confident spotting them, that still leaves roughly a quarter unsure. In practice, 77% cross-check AI output against trusted sources like PubMed, and 53% want AI to explain its reasoning.

Do Hospitals Have Medical AI Policies in Place?

Mostly not yet. Only 27% of clinicians understand their institution's AI governance policy, barely up from 21% in 2025. Just 35% know how to verify AI accuracy, and only 22% have clear rules on who's responsible when AI gets it wrong. Adoption is outrunning governance.

Kit's Take

Medical AI earns its keep on assistive tasks — reading scans, summarizing literature, drafting notes — which is exactly why usage doubled in a year. But push it into treatment recommendations or patient communication, and even the doctors using it daily get cautious, because the evidence hasn't caught up to the hype. If you're in healthcare, treat it as a time-saver, not a liability shield — you still sign the chart.

You won't know if it works until you try it.

Sources / 資料來源

常見問題 FAQ

醫療AI現在有多普及？

根據Wolters Kluwer 2026年調查，74%醫師與70%護理師每週至少使用一次AI，較去年的38%與46%幾乎翻倍。

什麼是醫療AI的「去技能化」風險？

指醫師過度依賴AI，導致自己的臨床判斷與診斷能力退步，是74%受訪醫師最擔心的問題。

醫療AI會不會產生幻覺（編造內容）？

會，75%的醫師把AI幻覺列為主要疑慮，73%自認能抓出錯誤，仍有約四分之一的人沒把握。

醫院有訂出醫療AI使用規範嗎？

多數還沒有，只有27%的醫師清楚自己單位的AI治理政策，較2025年的21%只微幅進步。

醫療AI現在適合用在哪些工作？

整理文獻、分析數據、輔助病歷書寫等輔助型工作已證明好用；但治療建議、病患溝通等需要臨床判斷的領域，證據還不夠充分，連常用AI的醫師自己都比較保守。

延伸閱讀 / Related Articles

AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

Claude Code 實測：AI 幫你寫程式到底行不行？ | Claude Code Review: Can AI Really Code for You?

3月 26, 2026

閱讀完整內容

搜尋此網誌

AI小貼士