借助临床医生的专业能力与自主人工智能
内容总结：
【医疗AI新突破：智能助手为医生减负，重塑和谐医患关系】
在美国，医师平均每周工作时间高达59小时，其中超过8小时被繁琐的行政文书工作占据。美国医学会数据显示，近半数医生正遭受职业倦怠困扰。这种现状催生了医疗AI技术的加速应用——环境智能助手正通过实时记录医患对话、自动生成结构化病历，将医生从文书工作中解放出来。
以医疗AI企业Nabla推出的环境智能助手为例，该系统能实时记录医患交流，自动生成精准度达"90%以上"的诊疗记录。执业医师、Nabla首席医疗官Ed Lee博士表示："面对复杂病例，过去完成病历需要45分钟，现在只需点击按钮就能获得精炼准确的记录，让我能全心关注患者。"
更值得期待的是，具备自主行动能力的智能体AI（Agentic AI）即将带来更深层次变革。该技术通过整合电子健康记录系统，可自动完成调取患者数据、生成诊疗摘要、提醒待办事项等系列操作。Nabla首席执行官Alexandre LeBrun描绘了这样的场景："心内科医生通过语音指令，就能在患者就诊前自动整理生命体征、检验报告和影像资料。"
为确保AI技术在医疗领域的可靠应用，专家强调需建立三重保障：首先通过系统培训使医护人员理解AI能力边界；其次让一线医务人员参与产品设计；最后建立包容透明的组织文化。目前这类AI系统已采用对抗训练模型等技术确保输出准确性，当存在不确定性时会默认采取保守策略。
业界共识在于，医疗AI的核心价值并非替代医生决策，而是通过自动化行政流程，让医护人员回归本职——成为真正专注于患者的医者。正如LeBrun所言："AI的首要使命是让医生回归医疗决策的本位。"
中文翻译：
与Nabla合作
释放临床医生与自主人工智能的协同效能
环境智能助手如何助力医护工作者节省时间、减轻职业倦怠并优化诊疗，重塑理想医患体验。
对许多临床医生而言，行政文书工作本身就是一项全职任务。从检查发现到治疗方案、检测结果和患者教育，医护人员必须在每个环节保持临床记录的准确、清晰和及时。而这一负担正日益加重：复杂的医保计费要求和不断变化的法规需要大量文书工作，电子健康记录系统往往操作反直觉、流程繁琐，严重拖慢工作效率。
美国医学协会数据显示，行政事务占据美国医生59小时周工作时长中的至少8小时，更多人会在业余时间额外花费超过8小时处理电子病历。这导致近半数美国医生出现职业倦怠。
但环境智能助手已能简化工作流程、加速病历记录、提升医疗系统整体效率，未来自主人工智能更将承担更多行政负担。这正彻底改变医护人员的时间分配方式，创造更多与患者深度交流的机会。
"我读医学院不是为了当文书员。应该有技术能替我完成这项工作，"医疗人工智能公司Nabla首席医疗官李爱德华博士表示。该公司于2023年推出首款环境智能助手，目前已被美国数百家医疗机构采用。
美国医生职业倦怠率趋近于零
数据来源：MIT科技评论洞察根据美国医学协会2025年数据整理
让医生回归医生本色
当前主流的环境智能助手已能实时记录、结构化整理并总结医患互动内容，将临床医生从耗时的文书工作中解放出来，使其能全心关注患者。"面对复杂病例，过去完成病历文档需耗时45分钟。Nabla极大改善了这一过程，让我能全神贯注对待每位患者。问诊结束时轻点鼠标，系统即刻生成条理清晰的精炼记录，"李博士指出该系统准确率"高达90%以上"，但最终记录仍需临床医生审核签字。
这种无间断的医患互动能促进更充分的眼神交流与更高质量的沟通。例如，当有替代性记录方式时，医生更倾向于在评估过程中口述思考内容。"我们原以为患者会担心AI设备监听，但实际上他们非常欢迎，"Nabla联合创始人兼首席执行官亚历山大·勒布朗表示，"患者在问诊时获得医生全程关注，当他们听到专业术语时，会感受到更优质的诊疗服务。"
据勒布朗介绍，Nabla系统还能通过自动化预检流程（在预约前审阅整理患者电子健康记录）以及医疗数据编码计费等功能，为临床医生提供进一步支持。该平台新增的内置听写功能让医护人员获得更统一的工作体验。这类智能助手任务有助于优化临床工作流程，降低机构行政管理成本。
自主人工智能的进化前景
包括Nabla在内的企业正致力于将自主人工智能整合进系统，推动现有智能助手功能再升级。勒布朗展望未来临床医生将与自主平台交互，该平台能联通既有工具系统，简化多步骤操作——如读取患者数据、在电子健康记录系统中执行操作、实时适配工作流程等。
"我们的平台将提供专业化、可定制、可组合的智能体，把零散工具整合成连续工作流，不再迫使医护人员在十几个独立系统间反复切换，"勒布朗解释道，"设想心内科医生准备晨间门诊时，通过几句语音指令，一个智能体从电子病历调取最新生命体征、检验结果和影像报告，另一个生成清晰的患者摘要，第三个标记出遗漏的随访心脏超声检查——这些在患者踏入诊室前就已完成。"
李博士认为自主人工智能的近期应用范围限于标准化非临床任务，但在治疗方案等临床决策支持领域同样潜力巨大，这些场景中人工智能可在临床医生全程监督下安全运作。
实现这一目标需要教育先行。"医学的魅力在于它是终身的学习过程。不仅要学习药物、诊断和治疗背后的科学原理，更要适应使用能提升病患护理质量的新工具，"李博士强调，"我们需要从AI基础知识着手，确保所有人理解其功能边界与风险隐患，真正明确它在患者照护中的最佳应用场景。"
他补充说，领导层必须战略前瞻，确保整个组织在AI认知与应用上协同迈进。"这需要让一线使用者参与流程设计，尽可能共同创作，并通过试点新方案推动组织学习。此外，需建立包容、真实、透明的文化氛围，才能为成功融合自主人工智能等转型实践创造最佳条件。"
安全融入工作流程
在医疗等高风险领域应用人工智能，需要在效率与准确性间谨慎权衡。"信任是医学的基石，"勒布朗指出，"赢得信任意味着通过准确性、透明度和对专业知识的尊重来赋予临床医生信心。"Nabla采用对抗训练模型等技术校验输出结果，并默认保守响应机制："我们优先保障精确度。只要存在细微存疑，默认会从输出中剔除该内容。"
新工具还必须与现有工作流程和平台无缝衔接，避免增加临床医生负担。"任何产品即便设计精良，若不能融入现有工作流程也毫无价值，"勒布朗坦言。在客户服务等领域构建新界面或平台相对直接，但这种方式在医疗领域既不可行也不可取。"这是由无数工作流程构成的复杂依赖网络。所有人都想革除旧系统，但不可能一次性全盘更换，"勒布朗解释道，而自主人工智能方案的优势在于"能在保留传统基础设施的同时优化流程"。
通过简化复杂系统、自动化常规任务、持续分担行政工作的耗时负担，自主人工智能在增强环境智能助手方面前景广阔。最终，这项技术的潜力不在于做出医疗决策或取代临床医生，而是助力医护人员将更多时间精力专注于核心要务——病患照护。"人工智能应聚焦决策支持与下游流程自动化，"勒布朗总结道，"它的首要使命是让医生回归医疗决策的本位。"
探索Nabla更多洞见
本文由MIT技术评论定制内容团队Insights制作。内容由人类作者、编辑、分析师及插画师完成全流程创作，包括调查问卷撰写与数据收集。可能涉及的AI工具仅限通过严格人工审核的辅助生产环节使用。
英文来源：
In partnership withNabla
Leveraging the clinician’s expertise with agentic AIwith agentic AI
How ambient AI assistants are supporting clinicians to save time, reduce burnout, and enhance treatment, restoring the doctor-patient experience.
For many clinicians, administration is a whole job on its own. From examination findings to proposed treatments, test results, and patient education, clinicians must maintain accurate, clear, and timely clinical records every step of the way. And the burden is getting heavier. Complex billing requirements and ever-changing regulations demand extensive documentation. And systems for electronic health records (EHR) are often unintuitive and inefficient, with complicated workflows that slow the process.
This administrative load occupies at least eight hours of a US physician’s 59-hour work week, according to the American Medical Association (AMA), with many spending more than eight hours on electronic medical records outside working hours. This all contributes to burnout, which affects nearly half of US physicians.
But ambient AI assistants are already offering a means of simplifying workflows, expediting patient records, and improving the overall efficiency of the health care system, with agentic AI poised to take on more of the administrative burden in the future. This is transforming the way clinicians spend their time and increasing their opportunities to engage meaningfully with patients.
“I didn’t go to medical school to be a scribe. There should be technology that can do this task for me,” says Dr. Ed Lee, practicing physician and chief medical officer at health care AI company Nabla, which launched its first ambient AI assistant in 2023, now used by hundreds of health care organizations in the US.
Nearly 0% of physicians in the US suffer from burnout
Source: Compiled by MIT Technology Review Insights, based on data from AMA, 2025
Letting doctors
be doctorsbe doctors
Current ambient AI assistants, which gained mainstream traction in 2023, are already able to record, structure, and summarize patient encounters in real time. This liberates clinicians from the time-consuming exercise of writing notes, allowing them to fully engage with their patients. “For complex patients, it could take me up to 45 minutes to complete the documentation. Nabla makes that task infinitely better and allows me to give each patient my full, undivided attention. At the end of the visit, I click, and Nabla produces a thoughtfully crafted, concise record of what happened,” says Lee, who puts the accuracy of Nabla’s system in the “high 90s” in terms of percentage, with the clinician always responsible for reviewing and signing off on the final record.
“For complex patients, it could take me up to 45 minutes to complete the documentation. Nabla makes that task infinitely better and allows me to give each patient my full, undivided attention. At the end of the visit, I click, and Nabla produces a thoughtfully crafted, concise record of what happened.”
This kind of uninterrupted patient engagement can lead to better eye contact and a higher quality interaction. For instance, clinicians tend to verbalize their thought process more when there is alternative notetaking during a patient evaluation. “We originally thought that patients would be worried about an AI device listening, but actually they are very excited,” says Alexandre LeBrun, co-founder and chief executive officer of Nabla. “They get the full attention of their physician during the visit, and they love when they hear technical language as they sense they get better care.”
According to LeBrun, Nabla’s system can further support clinicians by automating pre-charting, reviewing and organizing a patient’s information in their EHR before an appointment, and coding medical data for use in areas like billing. Nabla has also expanded its platform with a built-in dictation capability, bringing clinicians closer to a unified experience. These kinds of AI assistant tasks can help to streamline and enhance clinical workflows and contribute to a reduction in institutional administrative costs.
The promise of
agentic AIagentic AI
Agentic AI, which companies like Nabla are currently working to integrate into their systems, promises to take the success of existing AI assistants a step further. LeBrun is looking to a future in which clinicians interact with an agentic platform that links to all the tools they already use and simplifies multi-step interactions, like reading patient data, acting within the EHR, and adapting to workflows in real time.
“Rather than forcing doctors and nurses to click through a dozen separate systems, our platform will provide specialized, customizable, and composable agents that turn disconnected tools into a single, continuous workflow,” LeBrun says.
“Imagine a cardiologist getting ready for their morning clinic. After a few voice commands to instruct the system, one agent pulls the latest vitals, lab results, and imaging reports from the EHR, another generates a clear patient summary, and a third flags a missed follow-up echocardiogram. All before the patient even walks into the room,” LeBrun explains.
“Rather than forcing doctors and nurses to click through a dozen separate systems, our platform will provide specialized, customizable, and composable AI agents that turn disconnected tools into a single, continuous workflow.”
Lee says that agentic AI’s near-term scope includes standardized and protocolized non-clinical tasks, but he sees promise in areas like treatment options and other types of clinical decision support, where AI can safely operate with clinicians always “in the loop.”
To get to this point, education is essential, says Lee. “The beauty of medicine is that it’s a lifelong learning process. It’s not just learning about the science behind medications, diagnoses, and treatments; it’s about adapting to the use of new tools that will ultimately improve the care of the patients you treat,” he explains.
“We need to start with the basics of AI, making sure everyone understands what it is and how it works. Not how the programming takes place but more around what it can do, what it can’t do, the risks and pitfalls, and then really understanding where it fits best in the care of patients,” says Lee.
Leadership must look ahead strategically and ensure the entire organization is moving forward with its use and understanding of AI, he adds. “Part of that journey is involving frontline users to be part of the process, co-designing whenever possible and conducting pilots of new solutions so the organization can learn,” Lee says. Additionally, “a culture of inclusivity, authenticity, and transparency needs to be in place so you can be in the best position to be successful with transformative efforts such as incorporating and integrating agentic AI into the ecosystem,” he says.
“Part of that journey is involving frontline users to be part of the process, co-designing whenever possible and conducting pilots of new solutions so the organization can learn.”
Safely integrating
into workflowsinto workflows
Applying AI to high-stakes sectors like health care requires a careful balance between productivity on the one hand, and accuracy on the other. “Trust is everything in medicine,” says LeBrun. “Earning that trust means giving clinicians confidence through accuracy, transparency, and respect for their expertise.” Nabla uses techniques like adversarial training models to check outputs, and it defaults to conservative responses. “We optimize precision. If we have a slight doubt, we prefer to remove something from the output by default,” says LeBrun
“Trust is everything in medicine. Earning that trust means giving clinicians confidence through accuracy, transparency, and respect for their expertise.”
New tools must also interweave with existing workflows and platforms to avoid adding more complexity for clinicians. “Any product can look great, but if it doesn’t fit well into your existing workflows, it’s almost useless,” says LeBrun.
In sectors like customer service, it is straightforward to build a new interface or platform, but that approach isn’t feasible—or desirable—in health care. “It's a complex network of dependencies with so many workflows and processes,” says LeBrun. “Everybody would like to get rid of these things, but it's not possible because you would need to change everything at once.” Agentic AI approaches offer great promise to sectors like health care because they can “improve the process without getting rid of the legacy infrastructure,“ LeBrun explains.
By simplifying complex systems, automating routine tasks, and continuing to take on more of the time-consuming burden of administrative work, agentic AI holds great promise in further augmenting ambient AI assistants. Ultimately, the technology’s potential is not in making medical decisions or replacing clinicians, but in supporting health care workers to dedicate more of their time and attention to their main priority: their patients. “AI should focus on supporting decisions and automating everything downstream,” says LeBrun. “The first role of AI is to get physicians back to the state where they make medical decisions.”
Discover more insights from Nabla here.
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. This content was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.