Akido Deploys AI Diagnostic Tool to Treat Homeless Patients

Akido部署AI诊断工具救治无家可归患者 自主医疗AI对于通过病例分流及减少对传统诊所的依赖来化解严重的医生短缺问题不可或缺。 在本次炉边谈话中，Akido首席执行官Prashant Samant与资深记者Mark Borden讨论了Akido的AI原生系统及ScopeAI技术如何变革医疗服务交付，特别是在针对弱势群体的街头医疗领域。讨论重点介绍了AI如何赋能非医生现场工作人员在临时营地直接采集全面的诊断数据，并通过长期医嘱或虚拟医生派诊来开具处方。Samant概述了Akido服务超过一百万名患者的业务规模，解释了利用1020万个真实世界临床病例进行模型训练的情况，并探讨了“人在回路”监管如何在扩大临床服务能力的同时保障安全性。 美国医疗体系因结构性服务能力问题而陷入严重困境，现有医疗提供者仅能满足管理当前患病水平所需的大约20%。 (0:04) 传统的既有医疗体系建立在资源稀缺的基础之上，迫使患者必须前往集中的实体医疗机构才能获取人类医学专业诊疗。 (1:14) AI原生医疗系统颠覆了传统模式，它普及了临床智能，并将医疗照护与对实体医护人员的直接、持续依赖脱钩。 (1:14) ScopeAI为非医生外勤人员配备了AI医生助手，可在现场进行针对性随访，单次问诊接诊即可提示多达八种潜在诊断。 (3:02) 在长期医嘱授权下，使用ScopeAI的外勤人员可以直接发放广谱抗生素，或远程调度医生即时开具管制药物处方。 (3:02)

Akido CEO Prashant Samant argues that the fundamental crisis in American healthcare is an acute structural shortage of provider capacity, operating at roughly 20% of needed levels. To address this gap, Akido has developed an AI-native healthcare model that pushes diagnostic capabilities directly to field workers while preserving licensed physician oversight. Decentralizing Care to Address Provider Capacity Shortages (0:04) According to journalist Mark Borden and Akido CEO Prashant Samant, American healthcare suffers from an acute structural capacity failure rather than bad actors, with the nation operating with only about 20% of the providers needed to manage population illness. Legacy healthcare systems centralize scarce medical talent in physical facilities, forcing patients to navigate fragmented specialist networks that function like distant oases in a desert. Samant contrasts this with an 'AI-native health system' that decentralizes clinical intelligence away from continuous reliance on physical facilities. Rather than functioning merely as administrative scribes, Akido's tools embed clinical decision support directly into the encounter to bring medical expertise directly to patients. Deploying Mobile AI Diagnostics in Street Medicine (3:02) Akido deployed its ScopeAI application to treat homeless populations, who often face extreme resource deprivation and delayed interventions that culminate in emergency department visits. Non-physician field workers and case managers take mobile devices into shelters and encampments, using ScopeAI to conduct dynamic diagnostic intakes that can surface up to eight potential diagnoses across multiple specialties during a single 60- to 90-minute session. Care workers distribute broad-spectrum antibiotics directly in the field under clinical governance council standing orders. For controlled substances such as Suboxone for opioid addiction, ScopeAI dispatches a remote physician for immediate virtual consultations, reducing treatment initiation from weeks to hours and lowering emergency room utilization. Clinical Verification, Human Oversight, and Regional Scale (8:25) Akido reports managing care for over one million patients across more than 100 physical, virtual, and street-level sites, including approximately 40% of primary care in Rhode Island and major cardiology volume in Southern California's Pomona Valley. The service operates within existing Medicare, Medicaid, and commercial reimbursement frameworks, avoiding out-of-pocket costs for patients. Addressing safety and diagnostic reliability, Samant explained that ScopeAI's investigative reasoning is benchmarked against a proprietary repository of 10.2 million longitudinal clinical cases gathered over six years. To maintain clinical governance and regulatory compliance, licensed physicians retain final review and sign-off authority on all generated medical orders, creating feedback loops to refine the underlying AI models. Reallocating Clinician Labor Toward Complex Care (13:23) Addressing whether artificial intelligence will displace human doctors, Samant asserted that autonomous healthcare is mathematically inevitable because the clinician education pipeline cannot produce sufficient providers to meet national demand. However, he stated that Akido's immediate objective is capacity multiplication rather than labor elimination. By automating routine diagnostic triage and data collection, the platform seeks to relieve clinicians of early-stage intake burdens. This workflow reallocation is intended to concentrate scarce physician hours on high-level medical decision-making where human judgment is most critical. While Akido reports reductions in emergency room visits and rapid operational growth, the long-term sustainability of autonomous healthcare depends on maintaining rigorous physician-in-the-loop oversight as diagnostic responsibilities increasingly shift toward algorithmic triage.

Akido 首席执行官 Prashant Samant 与记者 Mark Borden 指出，美国医疗体系深陷仅能满足约20%患病人群需求的严重产能瓶颈；Akido 提出“AI 原生医疗系统”，试图将临床诊断推理能力直接下沉至基层外展与弱势群体场景，摆脱对物理诊所和人类医生实时在场的重度依赖。 医疗供给结构性短缺与AI原生系统 (0:04) Borden 与 Samant 认为，美国医疗服务的糟糕体验本质上是结构性产能问题，而非个案过失。目前全美医疗提供者数量至多只能应对全社会20%的患病负荷。传统医疗体系建立在“人类医生智力资源稀缺”的基础之上，运作模式形同沙漠中的单点绿洲，迫使患者耗费高昂成本前往实体诊所，跨专科转诊更要在不同机构间奔波。 Akido 提出的“AI 原生医疗系统”旨在重构这种中心化模式，通过将临床智能下沉普及，实现“送水上门”，使看诊不再绝对依赖实体医疗场所或医生的全天候物理在场。Borden 强调，这种架构不仅限于记录病历的“AI 抄写员”（AI scribes），而是让医疗智能在真实诊疗接触中与一线人员及医生协同作业。 ScopeAI与街头医疗的一线救治 (3:02) 针对面临严重资源匮乏、成瘾与精神障碍的无家可归群体，传统按服务收费的门诊模式往往无法提供跨专科随访，导致病情拖延至危急并频繁冲击急诊。为此，Akido 推出了基于移动设备（iPad 或手机）的 ScopeAI 系统，由具备相关生活经历的个案管理师等非医师外展人员携带，深入帐篷和庇护所开展床旁问诊。这一街头医疗模式曾被美疾控中心（CDC）前首席医疗官 Deb Houry 在《时代》杂志撰文肯定。 在实际运作中，ScopeAI 扮演现场高级医师助理的角色，可针对复杂体征展开跨专科追问，单次60至90分钟的问询可即时筛查出最多8项潜在诊断。外展人员依据临床委员会制定的常备医嘱，可直接现场分发广谱抗生素；针对成瘾药物（如 Suboxone）等受控处方，系统能即时派单调度远程执业医师进行点对点视频会诊开方，将以往长达数周的等待缩短至几小时内。 区域规模扩张与现有医保体系整合 (8:25) Samant 介绍，Akido 目前已在全美管理逾100万名患者，拥有100多处涵盖实体诊所、虚拟平台及街头外展的医疗网点，且已显现出减少急诊就诊、提高疗程依从性的效果。在区域布局上，Akido 承担了罗德岛州约40%的初级保健服务，并覆盖了南加州波莫纳山谷地区主要的心脏病医疗业务，后续还计划向纽约市与旧金山湾区扩张。 在商业支付路径方面，Akido 完全依托 Medicare、Medicaid 以及商业健康险等现有医疗支付体系运行，受惠患者无须承担额外的自费费用。Samant 认为，AI 工具正发挥平权效用，为以往无力负担私人特需医疗的弱势群体提供了类似高净值人群所享有的个案管理与照护服务。 临床推理验证机制与人机协同定位 (11:23) 在面对关于算法准确度与安全边界的质疑时，Samant 阐述了 Akido 的双重保障措施：首先，系统基于过去六年积累的1020万份真实世界纵向临床病例库，以此对 ScopeAI 的临床探究推理链进行基准对照，重点解决如何通过交互式问答甄别隐匿致命体征的难题；其次，维持“人类在环”（human-in-the-loop）监管，所有生成的医疗医嘱均须由持牌医师最终审核签字，形成合规审查与模型训练的闭环。 针对 AI 是否会完全取代人类医生的疑问，Samant 预测自主化医疗从长期看具有必然性，因为仅凭培训更多医生已在数学上无法跟上需求激增；但眼前的核心诉求是通过自动化分诊初筛来倍增现有医护产能，从而让人类医生将稀缺精力集中在最需要复杂人类判断的关键决策上。 Akido 展示了利用 AI 延伸基层诊断、缓解特定弱势群体医疗缺口的实践可行性，但其模式的持续推广仍受制于既有医保报销通道的兼容度与持牌医生的最终背书；随着业务向超大城市扩张，非医师人员借助算法行使临床判断时的责任边界与准确度稳定性，仍是其需要长期应对的现实挑战。

## Background 该内容由行业频道 Ai4 于 YouTube 平台发布，记录了资深商业记者 Mark Borden 与 Akido 首席执行官 Prashant Samant 的一场炉边对谈。活动探讨了超越传统 AI 病历速记员（AI scribe）范畴的“AI 原生医疗系统”，聚焦 ScopeAI 如何深度主导实际临床问诊，以及基于三年弱势群体一线实践数据的人机协同（Human-in-the-loop）诊疗机制。 ## TL;DR 对谈直面美国医疗面临高达 80% 供给缺口与满意度垫底的严峻现状，阐述了 Akido 如何颠覆传统的中心化“绿洲模式”，构建主动送医上门的 AI 原生分布式医疗网络。通过基层个案工作者携带搭载 ScopeAI 的移动终端深入街头与收容所，系统能在单次问诊中排查多项复杂病症，并以常规医嘱授权与远程医生秒级连线实现即时干预和处方签发。依托全美庞大的 1020 万份纵向临床案例库和严格的人机协同监管，Akido 已服务超百万患者，致力于通过现有保险支付体系将高阶医疗关怀普惠化，成倍放大稀缺医生的服务产能。 ## Atmosphere & style 整场对谈兼具尖锐的现实批判与深刻的人文温情，主持人的提问直击医疗体制痛点，Prashant 的回应则展现出前沿技术落地复杂现实场景的务实底气。从剖析医疗匮乏现状到分享街头流浪成瘾者被治愈后反哺社区的动人故事，情绪曲线从沉重逐渐走向振奋，并在观众问答环节的机智幽默中传递出对自主医疗未来的坚定信心。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=4s) 美国医疗满意度长期垫底甚至低于车辆管理局（DMV），核心病因在于严重的医疗资源供给匮乏，全美现有医生仅能满足约 20% 的实际健康需求。 - [01:14](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=74s) Akido 推出全球首个“AI 原生医疗系统”，旨在颠覆传统需要患者跋涉就诊的“绿洲模式”，通过平民化临床智能实现“主动将医疗送上门”。 - [02:44](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=164s) 在前线街头医疗项目中，经过培训的基层工作者携带搭载 ScopeAI 的移动终端进驻街头帐篷，单次 1 至 1.5 小时对话即可梳理出约 8 项潜在多学科诊断。 - [06:15](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=375s) 系统实施严密的临床分级调度：常规感染与外伤依托临床治理委员会预设医嘱现场配药，而 Suboxone 等管制类药物则秒级接入远程医生视频审核处方。 - [06:15](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=375s) 对于物质成瘾等脆弱群体，ScopeAI 协同机制将传统长达数周的转诊排队周期压缩至数小时内，在患者产生戒断意愿的黄金窗口期即时锁定救治。 - [08:08](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=488s) Akido 已从初创企业发展为拥有超 100 家诊所、管理超 100 万患者的网络，承担了罗德岛州 40% 的初级保健服务及南加州波莫纳谷的心血管专科医疗。 - [08:08](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=488s) 技术普惠展现深远社会价值：曾被街头医疗治愈的前流浪成瘾者重获新生后，加入 Akido 成为前线个案工作者，实现生命影响生命的救助闭环。 - [10:43](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=643s) Akido 业务正加速向全美拓展，目前已正式落地纽约市和旧金山湾区运营。 - [11:28](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=688s) 系统基于 6 年积累的 1020 万真实纵向临床病例库构建基准，评估重心放在动态问诊与鉴别排查过程（Investigative Accuracy），而非单纯给出最终诊断。 - [11:28](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=688s) 所有医嘱与处置在提交前必须经过执业医生（Human-in-the-loop）最终核准，既确保合规安全，也为 AI 模型提供纠偏与持续迭代机会。 - [13:27](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=807s) 商业模式全面嵌入现有支付网络，覆盖 Medicare、Medicaid 及商业健康保险，避免将经济负担转嫁给患者。 - [13:27](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=807s) AI 的核心定位并非替代医生，而是在巨大供给缺口下实现产能倍增，将医生从繁重消耗中解放出来以专注于最关键的人文关怀与专业决策。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=4s) 美国医疗体系的根本困局与供给危机 主持人 Mark Borden（资深商业记者，曾供职于《财富》与《Fast Company》，并曾担任《纽约时报》CEO 演讲撰稿人）开门见山地指出，在当今所有公众议题与报道领域中，医疗健康无疑是最具话题度也最亟待改变的领域。他毫不避讳地提出一个尖锐的前提：“美国的医疗体系糟糕透顶（Healthcare sucks in America）。”在全美的各类消费者满意度调查中，医疗体验的评分常年垫底，甚至比前往车辆管理局（DMV）排队办事的体验还要糟糕。 Akido 联合创始人兼首席执行官 Prashant Samant（此前曾担任南加州大学医疗系统创新实验室主任）对此深表赞同，但他强调这并非源于某个人的过错或恶意。其核心根源在于极其严峻的医疗资源供给匮乏——全美根本没有足够的医生来支撑庞大的人口需求。即便按照最乐观的测算，面对当前人群的患病与健康管理需求，美国现有的执业医生和医疗提供者数量也仅能满足大约 20% 的需求。面对高达 80% 的医疗供给缺口，传统体系早已不堪重负。 - **消费者满意度垫底**：美国医疗健康服务的用户满意度评分持续低于 DMV（车辆管理局）。 - **极度匮乏的供给比**：全美现存的医生资源最乐观估计仅能满足实际患病人群需求的 20%。 --- ### [01:14](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=74s) 颠覆传统“绿洲模式”：构建 AI 原生医疗系统 针对医疗资源供给不足的痼疾，Akido 宣布推出全球首个“AI 原生医疗系统”（AI-native health system）。Prashant 解释了传统医疗体系与 AI 原生体系在底层设计哲学上的根本差异：传统医疗系统本质上是“围绕稀缺性建立的”，由于医生极度短缺，获取临床医学智慧（Medical Intelligence）就如同在荒漠中寻找水源。 传统模式就像是建立了一座“绿洲”，所有口渴的患者必须亲自跋涉到绿洲旁排队取水。然而，绿洲周围的位置极其有限；如果患者需要看不同的专科医生，就不得不长途跋涉前往另一座全然不同的“专科绿洲”，这也是为什么普通人往往需要辗转多个门诊才能查清自身病因。而 AI 原生医疗系统则彻底颠覆了这种以“绿洲”为中心的架构，改为“主动将水送到每个人身边”。它通过解除临床智能对特定人类医生物理存在的绝对依赖，将原本高门槛的医疗诊断能力平民化、普惠化。这远不止于辅助记录病历的 AI 书记员（AI scribe），而是让 AI 临床智能在实际问诊中与医生实现深度的并肩协作。 ```mermaid flowchart TD accTitle: 传统医疗绿洲模式与 AI 原生分布式模式对比 subgraph Legacy["传统稀缺模式：绿洲架构"] P1["多病症患者"] -->|辗转奔波| O1["全科门诊绿洲\n(容量极度饱和)"] P1 -->|重复挂号| O2["专科门诊绿洲\n(跨机构协作困难)"] end subgraph AINative["Akido 模式：AI 原生分布式架构"] Scope["ScopeAI 临床智能引擎"] --> Field["实地个案工作者 / 移动终端"] Field -->|送医上门| Bedside["患者所在场所\n(帐篷/庇护所/社区)"] Scope -.->|按需调度| MD["远程执业医生\n(关键环节把关)"] end ``` --- ### [02:44](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=164s) 深入街头：ScopeAI 赋能下的前线流动诊疗 《时代》（Time）杂志刊发了前美国疾控中心（CDC）首席医疗官 Deb Houry 博士的一篇探讨美国医疗乐观前景的文章，文中重点提及了三家关键公司：OpenAI、Anthropic 以及 Akido。其中，Akido 针对无家可归群体（People experiencing homelessness）开展的街头医疗项目（Street Medicine Program）被视作 AI 原生医疗落地的标杆范例。 这部分极度脆弱的边缘群体同时面临食物与水源匮乏、精神疾病、药物成瘾以及通讯手段缺失等多重打击。传统模式下，他们因复合病症需要往返于十余家不同的医疗机构，且因沟通困难、就诊耗时极长且难以通过常规保险结算，导致医生缺乏动力接诊，最终该群体往往只能在病情恶化至灾难性阶段时频繁涌入医院急诊室（ED）。 在 Akido 的体系下，经过培训的现场个案工作者或护理经理（许多人自身拥有无家可归的过往经历）背着双肩包、手持搭载 ScopeAI 系统的 iPad 或手机直接深入庇护所或街头帐篷。ScopeAI 作为先进的 AI 临床引擎，在与患者交流时能像经验丰富的医生一样实时提出最恰当的跟进问题。当患者从一个症状跳转到另一个健康隐患时，ScopeAI 会自动展开多维度的交叉调查。犹如把“豪斯医生（Dr. House）”的多学科诊疗能力装进了背包，仅需在一处帐篷里进行 1 到 1.5 小时的促膝长谈，系统便能梳理出多达 8 项不同的潜在诊断，在患者最舒适的环境下完成全面的健康评估。 - **多重生存危机交织**：街头患者常年面对物质匮乏、心理疾患与成瘾问题，难以进入固定诊所体系。 - **便携式多学科问诊**：由具备同理心的基层工作人员携带 ScopeAI 移动终端直接进驻帐篷与收容所。 - **全景式诊断输出**：单次 1 至 1.5 小时的现场对话即可挖掘出涉及多器官、多系统的约 8 项独立诊断结论。 --- ### [06:15](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=375s) 临床调度与人机协同：从常规抗生素到处方药 Suboxone 当 AI 在现场梳理出大量诊断时，整个流程通过严密的临床治理与即时分级调度机制展开“人机协同”（Human-in-the-Loop）： ```mermaid flowchart LR accTitle: ScopeAI 现场诊疗与临床分级调度流程 A["现场 AI 诊断采集\n(ScopeAI 梳理病情)"] --> B{"处置类型判断"} subgraph Protocol["常规医嘱通道"] B -->|常规感染/创伤| C["临床治理委员会\n授权常规医嘱"] C --> D["现场背包即时配发\n(广谱抗生素等)"] end subgraph Controlled["受控物质与高阶诊疗"] B -->|成瘾治疗/管制药物| E["ScopeAI 智能调度系统"] E --> F["实时远程连线医生\n(移动端视频会诊)"] F --> G["医生核准并签署处方\n(Suboxone 立即给药)"] end ``` 如果患者遭遇创伤感染，在临床治理委员会（Clinical Governance Council）事先设立并严格监管的“常规医嘱（Standing Orders）”框架下，一旦症状完全吻合，个案工作者即可依据授权现场从背包中取出广谱抗生素交付给患者。 然而，若患者同时患有甲基苯丙胺（冰毒）成瘾，且在当下展现出强烈的戒断与治疗意愿，希望启动药物辅助治疗（MAT）并使用丁丙诺啡/纳洛酮复合剂（Suboxone）时，由于该药物属于受控物质（Controlled Substance），法律上必须有执业医生的直接监管与处方签署。此时，ScopeAI 瞬间化身为精准的调度系统，在几秒内将远程值班医生接入现场手机的视频通话中。医生快速调阅 ScopeAI 整理的诊断依据并完成现场视讯复核，当场签署处方授权用药。Prashant 强调，物质成瘾者的治疗意愿窗口期转瞬即逝，传统转诊排队往往需要数周，而这套人机协同系统将这一周期直接压缩到数小时之内，在患者敞开心扉的黄金时刻立即锁定救治。 - **即时医嘱响应**：常见外伤与感染基于治理委员会授权的标准化流程，工作人员可现场开包给药。 - **关键药物秒级连线**：针对受控管制类药物（如 Suboxone），AI 自动调度远程医生无缝切入，完成合法合规的现场视频把关与处方签发。 - **显著改善结局指标**：白皮书数据显示，该模式极大降低了急诊室就诊率，显著提升了高危群体的治疗建档与长期留存率。 --- ### [08:08](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=488s) 规模化跨越与普惠关怀：将“礼宾级”医疗赋予最脆弱人群 Mark 指出，如果这套模式能够在洛杉矶环境最复杂、人群最脆弱的街头跑通，其拓展至常规诊所和全美医疗网络的能力将带来巨大变革。谈及 Akido 的成长规模与未来发展，Prashant 分享了一组对比数据：从一家初创企业起步，Akido 如今已拥有超过 100 家诊所（涵盖实体诊所、虚拟医疗与街头巡诊网络），管理并服务着超过 100 万名患者。尽管与服务约 1100 万人的全美医疗巨头 Kaiser Permanente 相比仍有差距，但 Akido 已经深度融入了美国基础医疗体系的核心版图——例如，Akido 承担了罗德岛州（Rhode Island）全州多达 40% 的初级保健服务，同时也是加州波莫纳谷（Pomona Valley）核心的心血管专科医疗提供方。 在商业扩张之外，Prashant 尤为自豪的是技术的包容性（Inclusiveness）。许多前沿医疗创新往往沦为高收入群体的专属特权，而 Akido 运用 AI 的平民化力量，将过去只有高净值人群才能享受的“礼宾级（Concierge-level）全方位健康关怀”，无偿且尊严地带给了加州乃至全美规模最庞大的无家可归群体。 在分享环节尾声，Prashant 动情地讲述了一个基层一线的故事：团队中有一名深入街头救助流浪人员的个案工作者，其本人过去就曾是一名无家可归且深陷毒瘾的流浪者。正是 Akido 的街头医疗团队在帐篷边发现了他，运用这套体系治愈了他的疾病并帮他彻底戒除毒瘾。这位重获新生的人深受感召，毅然选择加入 Akido，背起装满药品与智能设备的背包重返街头，去拯救那些正处于他曾经经历过的绝望深渊中的同胞。 - **百万级服务体量**：Akido 从初创公司跨越式发展为服务超 100 万患者、下辖 100 多家诊所的医疗体系。 - **重点区域核心支柱**：覆盖罗德岛州 40% 的初级保健服务，并在南加州波莫纳谷建立起强大的心血管专科支撑。 - **生命影响生命的闭环**：曾经的街头被救助者经系统化脱瘾与康复后，转型成为 Akido 前线个案工作者，实现社会价值与临床救助的双重闭环。 ### [10:43](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=643s) 团队成长与业务拓展前瞻 对谈在热烈的氛围中继续推进。主持人 Mark Borden 与 Prashant Samant 谈到了社区工作者的发展轨迹——许多最初作为基层协调人员的伙伴如今直接加入了 Akido 团队，深度参与到构建医疗未来的旅程之中。Prashant 感慨道，这种让基层工作者真正融入创新核心的模式，在其他行业或平台是极其罕见的。 Mark Borden 提到，在交流中 Prashant 透露了即将公布的重磅动态与重大新闻。Prashant 证实了这一点，表示公司很快就会有一系列重要消息释出，包括多位重量级人才的加入，以及全新业务区域的拓展。目前，Akido 已经正式在纽约市（New York City）和旧金山湾区（Bay Area）落地运营，更多详细规划也将陆续向外界披露。 --- ### [11:28](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=688s) 1020万病例库与问诊排查精度的评估体系 现场进入观众提问环节。一位观众表明自己是临床医生，并指出随着美国人口老龄化加剧，老年医学专家（Geriatrician）严重短缺（其表亲即为在拉斯维加斯执业的老年科医生）。他向 Prashant 提出核心疑问：在强调 Human-in-the-loop（人机协同）的背景下，Akido 究竟如何评估并确保 AI 系统在临床推荐中的准确度、相关性与整体质量？ ```mermaid flowchart TD accTitle: Akido Clinical Evaluation and Oversight Pipeline accDescr: Evaluation pipeline from 10.2 million longitudinal case studies to investigative accuracy and final physician oversight. subgraph Evaluation ["评估基准与训练数据"] A["1020 万真实纵向临床案例库<br/>（全美最大医学病例集）"] --> B["问诊排查过程评估<br/>（Investigative Accuracy）"] A --> C["最终诊断评估<br/>（Diagnostic Accuracy）"] end subgraph ClinicalLoop ["临床执行与闭环监管"] B --> D["ScopeAI 实时辅助问诊"] C --> D D --> E["处方/医嘱提交前审核"] F["执业医生（Human-in-the-loop）"] --> E E --> G["监管合规与系统持续学习"] end ``` Prashant 对此作了深入拆解： 1. **全美最大的纵向临床案例库**：在公司创立的前六年时间里，Akido 耗费大量精力编译了涵盖约 1020 万份真实临床案例与纵向病程跟踪的庞大数据库，堪称“全美汇编规模最大的医学教科书”。系统在此基础上与真实就诊过程进行全方位基准比对。 2. **重在排查精度而非单纯诊断**：Prashant 强调，如果已经拿到完整的病历全貌，得出最终诊断其实“相对容易”；真正的技术难点在于**排查与问诊过程（Investigative Process）**。同一种疾病在不同患者身上表现各异，病情究竟是轻微还是危急，完全取决于医生层层递进的问答逻辑与鉴别路径，Akido 的核心评估指标正是这种动态排查的准确度。 3. **医生闭环监管（Physician-in-the-loop）**：系统生成的任何医嘱与处置方案在正式提交前，均须由执业医生进行最终把关与审核。这不仅为 AI 模型提供了遇到差异时的纠偏学习机会，更构筑了符合监管规范的合规安全屏障。 --- ### [13:27](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=807s) 支付方机制与“人机协同”替代争议 当被现场问及“谁来为这套系统买单”时，Prashant 幽默而干脆地回答：“保险公司。”现场随即爆发出一阵笑声。他进一步阐明，Akido 的核心理念是——自主医疗（Autonomous Health）是不可逆转的必然趋势，因为仅靠传统的医学院培养机制，根本不可能填补日益扩大的医生缺口。然而，自主医疗必须无缝嵌入当下的支付与监管体系，绝不能将额外的经济与操作负担转嫁给患者。因此，Akido 在设计之初就完全适配现有的医疗报销网络，全面覆盖联邦医疗保险（Medicare）、联邦医疗补助（Medicaid）以及各类商业健康保险。 针对“AI 是否旨在完全替代临床医生”这一行业经典问题，Prashant 先是风趣地调侃了一句“完全是（Totally）”，引得全场大笑，随后认真澄清了技术定位： - **产能倍增而非取代**：医疗系统当前面临的真正危机是人力极度匮乏，“谈论替代还为时尚早，因为我们现有的医生数量根本远远不够”。 - **优化临床精力分配**：系统的真正目标是进行产能放大（Multiply capacity），将身处繁重高压岗位的医生与医护人员从机械劳动中解放出来，确保他们能把最宝贵的时间与精力，“投入到最需要人类专业判断与人文关怀的关键环节中”。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=4s) 美国的医疗体系糟糕透顶（Healthcare sucks in America）。 — Mark Borden > [00:04](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=4s) 全美现存的医生资源最乐观估计仅能满足实际患病人群需求的 20%。 — Prashant Samant > [01:14](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=74s) 传统医疗系统本质上是“围绕稀缺性建立的”，由于医生极度短缺，获取临床医学智慧就如同在荒漠中寻找水源。 — Prashant Samant > [01:14](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=74s) 解除临床智能对特定人类医生物理存在的绝对依赖，将原本高门槛的医疗诊断能力平民化、普惠化。 — Prashant Samant > [02:44](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=164s) 犹如把“豪斯医生（Dr. House）”的多学科诊疗能力装进了背包，仅需在一处帐篷里进行 1 到 1.5 小时的促膝长谈，系统便能梳理出多达 8 项不同的潜在诊断。 — Prashant Samant > [06:15](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=375s) 物质成瘾者的治疗意愿窗口期转瞬即逝，传统转诊排队往往需要数周，而这套人机协同系统将这一周期直接压缩到数小时之内，在患者敞开心扉的黄金时刻立即锁定救治。 — Prashant Samant > [08:08](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=488s) 将过去只有高净值人群才能享受的“礼宾级（Concierge-level）全方位健康关怀”，无偿且尊严地带给了加州乃至全美规模最庞大的无家可归群体。 — Prashant Samant > [11:28](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=688s) 如果已经拿到完整的病历全貌，得出最终诊断其实“相对容易”；真正的技术难点在于排查与问诊过程（Investigative Process）。 — Prashant Samant > [13:27](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=807s) 谈论替代还为时尚早，因为我们现有的医生数量根本远远不够。 — Prashant Samant > [13:27](https://www.youtube.com/watch?v=DtQPvDe9ZRI&t=807s) 系统的真正目标是进行产能放大（Multiply capacity），将身处繁重高压岗位的医生与医护人员从机械劳动中解放出来，确保他们能把最宝贵的时间与精力，投入到最需要人类专业判断与人文关怀的关键环节中。 — Prashant Samant ## Mentioned Prashant Samant, Mark Borden, Akido, Egospeed, ScopeAI, Ai4, Deb Houry, OpenAI, Anthropic, 《时代》（Time）, 《财富》（Fortune）, 《Fast Company》, 《纽约时报》（The New York Times）, 南加州大学（USC）, 美国疾控中心（CDC）, Kaiser Permanente, Suboxone, 车辆管理局（DMV）, 联邦医疗保险（Medicare）, 联邦医疗补助（Medicaid） ## Video description Prashant Samant - CEO, Akido Mark Borden - Founder, Egospeed A fireside chat with Prashant Samant (CEO), Akido, moderated by Mark Borden AI note-taking is table stakes. What happens when AI runs the actual clinical encounter? Akido's CEO Prashant Samant joins veteran reporter Mark Borden to make the case. In this fireside chat, they'll walk through a ScopeAI visit, grounded in outcomes data from three years of deployment across underserved and high-need populations. Then tackle the question every health system and regulator is really asking: where does a physician step in and override the AI? It's the moment that separates AI-doctor hype from the human-in-the-loop reality of care delivered at scale. What does the future of healthcare look like? What is the role of humans? Finally, thoughts on bots. [Source](https://www.youtube.com/watch?v=DtQPvDe9ZRI)

[0:04] Mark Borden: 美国的医疗体系糟糕透顶（Healthcare sucks in America）。 [0:04] Prashant Samant: 全美现存的医生资源最乐观估计仅能满足实际患病人群需求的 20%。 [1:14] Prashant Samant: 传统医疗系统本质上是“围绕稀缺性建立的”，由于医生极度短缺，获取临床医学智慧就如同在荒漠中寻找水源。 [1:14] Prashant Samant: 解除临床智能对特定人类医生物理存在的绝对依赖，将原本高门槛的医疗诊断能力平民化、普惠化。 [2:44] Prashant Samant: 犹如把“豪斯医生（Dr. House）”的多学科诊疗能力装进了背包，仅需在一处帐篷里进行 1 到 1.5 小时的促膝长谈，系统便能梳理出多达 8 项不同的潜在诊断。 [6:15] Prashant Samant: 物质成瘾者的治疗意愿窗口期转瞬即逝，传统转诊排队往往需要数周，而这套人机协同系统将这一周期直接压缩到数小时之内，在患者敞开心扉的黄金时刻立即锁定救治。 [8:08] Prashant Samant: 将过去只有高净值人群才能享受的“礼宾级（Concierge-level）全方位健康关怀”，无偿且尊严地带给了加州乃至全美规模最庞大的无家可归群体。 [11:28] Prashant Samant: 如果已经拿到完整的病历全貌，得出最终诊断其实“相对容易”；真正的技术难点在于排查与问诊过程（Investigative Process）。 [13:27] Prashant Samant: 谈论替代还为时尚早，因为我们现有的医生数量根本远远不够。 [13:27] Prashant Samant: 系统的真正目标是进行产能放大（Multiply capacity），将身处繁重高压岗位的医生与医护人员从机械劳动中解放出来，确保他们能把最宝贵的时间与精力，投入到最需要人类专业判断与人文关怀的关键环节中。

[0:04] Mark Borden: healthcare in America is severely broken [0:04] Mark Borden: worst consumer product in America, ranking below the DMV experience [1:14] Prashant Samant: water to the people [8:25] Prashant Samant: concierge-level experience

[0:04] So just kind of start with a little background.  Uh I'm a journalist. I worked at Fortune Fast [0:08] Company. I was a speech writer for the CEO  of the New York Times. Lately I've been [0:12] covering healthcare. I think it's one of the most  interesting um spaces in in journalism or in the [0:17] cultural conversation. I think one of the most  interesting companies that are actually engaged [0:22] in this space is Aikido. Pashant is the CEO of  Aikido. Um, prior to that, he was the director of [0:28] the innovation lab at UC USC's, um, health system.  Um, I think we should just start by coming out [0:35] and saying it. Uh, healthcare sucks in America.  Like, do you agree with me? Healthcare sucks. Um, [0:42] healthcare is consistently rated as the worst  consumer product. America, it is worse rated than [0:48] the DMV experience. Uh, and the fundamental issue  is that it's not any individual's fault. It's not [0:54] a person's fault. It's not because of mal intent.  It's just that we do not have enough doctors to [1:00] take care of the population. Most optimistically,  we have 20% of the providers that we need to take [1:06] care of our current level of sickness in this  particular population right now. Right. Okay. So, [1:14] earlier this month, you announced that you're  launching the world's first um let me get this [1:21] right, AI native health systems, right? Um, so  what is an AI native health system and how is it [1:28] going to make American healthcare suck less? Well,  let's start with what a legacy health system is. [1:33] So, a legacy health care system is designed  around scarcity. Um, and the scarcity here, [1:40] because we have an under supply of doctors,  is access to medical intelligence. So, [1:45] when you're designing around that, it's kind of  like you're building an oasis that everybody has [1:50] to come to to drink water. Um, and it's there  and people want to serve you water, but there's [1:55] just not enough room around that oasis. And  then if you have to go to another specialist, [2:00] it's like going to a whole another oasis. I think  everybody here has been to multiple employants to [2:04] get an understanding of what's going on with them.  um a AI native health system is inverting that is [2:11] uh making it instead of it being uh oasiscentric  where you get water you're bringing water to the [2:16] people and you're doing that by tackling that  fundamental primitive issue by making clinical [2:23] intelligence much more democratized by divorcing  its reliance on uh access to a human being. So [2:30] beyond the scribe, you're talking about actually  this um AI medical intelligence actually working [2:39] hand inand with the physicians in the appointment.  Correct. Um so I don't know how many people read [2:45] today in Time magazine uh Dr. Deb Howry,  former um chief medical officer for the CDC, [2:51] um wrote sort of a contrarian piece then that not  supporting the healthcare stocks, but what she's [2:57] actually optimistic about uh for the American  health care system. Um three big companies were [3:02] named uh OpenAI, Anthropic, and Aikido, your  company. And in particular, she pointed out the [3:09] street medicine program. Um this is a vulnerable  population, very difficult to treat. Tell me, [3:16] take us through what an encounter with a clinician  and somebody living in the street is like and how [3:23] you've changed that that kind of engagement.  Well, first of all, I mean, really honored to [3:29] be mentioned in that piece. Uh, I love Devi.  I think that, you know, Dr. Ary is an amazing [3:36] public servant. She was the CMO of the CDC for  10 years. Uh, amazing infrastructure that she's [3:41] helped build. Um, and I think that, uh, our street  medicine program focused on people experiencing [3:47] homelessness is a really amazing example of what's  possible with an AI native healthcare system. So, [3:53] um, I think it's helpful to understand it when  you justos it with what happens now versus what's [3:56] possible in an AI native world. So, right now, um,  people experiencing homelessness are dealing with [4:02] a lot of issues. They have very limited access to  health care, very limited access to basic human [4:07] services in general. Right. I mean, it's like food  scarcity, water scarcity, mental health issues, [4:11] addiction. Absolutely. And sometimes not even a  way to communicate, right? I mean, correct. And [4:16] you know, so there are physicians and providers  that create availability for these individuals, [4:22] but it's hard for them to come in to a facility.  Number one. Number two, because of all of these [4:26] various challenges, they need a lot of different  kinds of medical support, which means they need [4:31] to go to 10 different facilities to be able to  get access to what they need. Uh, and it just [4:35] doesn't happen. And if they do show up, they're  challenging subjects to manage. And so they take [4:39] a lot of time for a particular physician and the  physician can't really bill for it. So it's not a [4:43] very sustainable way to provide care for this  population that needs it the most. What they [4:48] do is they frequent emergency departments when  something becomes catastrophic. So in our world, [4:54] in Aikido's world, we're able to send people into  the field. Uh these people are case workers or [5:01] care managers, often times people that have lived  experience that may have been homeless themselves [5:05] at some point in time. They're armed with a  backpack and they're armed with our technology, [5:10] Scope AI, which is the world's most advanced  AI physician that is on their iPads or that [5:15] is on their phone. And powered by Scope, they  can have a full dialogue with this individual, [5:21] with this person experiencing homelessness.  And as that individual informs them what's [5:25] going on with them, scope is asking the next best  question, just like a doctor might in a a visit, [5:30] but instead of it being, you know, one area of  questioning, as that individual starts talking [5:35] about a different area of concern, scope starts  investigating that area of concern. So what that [5:39] uh case worker or that care manager is left  with is an abundance of diagnostic information [5:44] about what's happening to that individual purses  allows them to kind of care for mult. It's almost [5:50] multipety care in the field from a field worker.  Exactly. It's like Dr. House. Okay. Um and so you [5:57] have, you know, eight different diagnoses  that might emerge just from that one hour, [6:01] hour and a half long conversation that's happening  at a shelter or that's happening at a person's [6:06] tent. So in a place where that individual  might be comfortable and be open to to having [6:11] a conversation about what's going on with them. So  what if a diagnosis is raised that actually needs [6:17] um a physician interaction? How does that work?  Yeah, great question. So, um, as a abundance of [6:24] diagnoses are raised, some of them are going to  be needed to be dealt with immediately. Some, [6:28] uh, might be able to be handled by that person  in the field. I think a great example would be, [6:33] you know, uh, often times you're going to have a  person experiencing homelessness that might have [6:37] a wound that has been dealt with and it's clear  that, uh, based on this diagnostic process that [6:42] they need antibiotics immediately. Now we can have  a standing order put in place that's has oversight [6:49] by a clinical governance council that says if  a person is is exhibiting XYZ symptoms they can [6:55] get a broadspectctrum antibiotic immediately and  that person can hand that antibiotic out of their [7:00] backpack to that individual but that individual  might also be suffering from a meth addiction and [7:06] they might be ready at that moment in time to get  on a medication assisted treatment program enabled [7:11] by say suboxone. Now, Suboxone is a controlled  substance. You can't have that handed out without [7:16] having physician oversight. So, in that moment,  Scopes as a dispatch system orchestrates pulling [7:22] in a doctor into the loop so that that individual  case worker can literally with a phone have a [7:28] virtual visit done right there, the point of care,  have that physician provide the proper oversight [7:33] and sign off on that Suboxone prescription and  then give that patient the suboxin they need. [7:38] And this is really powerful because as you might  imagine uh when a person has a moment they're [7:43] willing to get on a treatment program you have to  capture that immediately right without this point [7:48] of care it takes like weeks to get somebody  a suboxone treatment program going now we [7:53] can do it in a matter of hours. Yeah I think it's  remarkable I read the white paper the reduction in [7:58] um emergency room visits the um increase in  treatment the retention of this population is just [8:05] remarkable compared to what has happened before.  And this is a large population in Los Angeles. Um, [8:11] so I imagine that if you can treat a vulnerable  population living in a very difficult environment, [8:20] how that must translate into a clinic, how that  translates into a almost a national part of the [8:25] health system. So Aikido is serving a million  lives. Yeah. Over a hundred clinics, right? [8:32] So brick and mortar, also virtual care, street  program. we talked about what are you what are [8:39] you most proud of that you're building and that  you built and maybe take me a little bit into what [8:46] the plans are for the future of Aikido. So I am  proud of a lot of things. Our team has done some [8:53] incredible work in an incredibly tough industry.  Um I think you know one thing you mentioned is [8:59] that we have scaled Aikido to now take care of  over a million patients from startup right I [9:04] mean from a startup and you know just to put that  in context Kaiser Permanente which is one of the [9:08] biggest health systems in the country takes care  of about 11 million people now we're not Kaiser [9:12] size yet but a million people uh is going from a  startup to being part of the American healthcare [9:18] tapestry so for instance in Rhode Island we take  care of 40% of the primary care in that state in [9:23] Southern California if you're living in Pomona  alley, there's a very very good chance you're [9:28] coming to one of our cardiology services. So  being able to be uh vet heft and being this [9:34] forward thinking is I think a true accomplishment.  I think the other thing that I'm very proud of is [9:39] that who we bring care to. Um you know a lot  of technological innovation uh is reserved for [9:46] people that have meats and we have been able to  build a truly inclusive program. We take care of, [9:52] you know, more people experiencing homelessness  in California than I think anybody else in the [9:55] country, uh, or anybody else in California rather.  And, you know, I think that that inclusiveness, [10:02] uh, being able to bring this democratizing  technology, artificial intelligence, uh, [10:07] to people, uh, to create a truly concierge level  experience for those that would never be able to [10:13] afford it, I think is is is something that I'm  incredibly proud of. I think there's a story that [10:18] comes to mind. Um it it's it's you know it's like  tear inducing. Uh one of our uh case workers as I [10:26] described the case workers that go into the field  to take care of people experiencing homelessness [10:29] um they actually were experiencing homelessness  themselves and uh our team met them uh got them [10:37] well got them off of their addiction um and they  were so inspired that they went to become a case [10:43] worker and then joined our team. So now they  are part of the Aikido journey themselves. [10:48] And I think I don't know where else can you do  that, man. Yeah, that's remarkable. And I know [10:52] that you're earlier when we were talking you were  teasing out some major stories and major um just [10:58] news that you're going to be breaking and talking  about in the near future and it's very exciting to [11:02] hear what's going on and what you're doing. So I  really appreciate the time to talk to you and No, [11:07] thank you time to talk to you as well. Appreciate  you making the time and uh taking interest in what [11:11] we're up to. There's a lot of news coming out.  Uh so, you know, that should be announced pretty [11:15] shortly. Got some exciting new folks joining,  exciting new pay places we're launching in. [11:19] We're operating in New York City. We're operating  in the the Bay Area as well. And so, we'll talk [11:23] more about that soon. That's great. So, this is  exciting, you know. So, I'm a physician. I'm a [11:29] great uh US population is getting older and older,  but we don't have enough geratrician specialists. [11:35] Okay. So, my cousin is a geriatrician who lives  in Vegas actually. So, so my question is so how do [11:41] you evaluate you know the uh accuracy relevant all  this stuff you know you you you know the AI sorry [11:50] so AI is the tools right you know so humor must be  in the loop so I say in your company so how do you [11:57] you know make sure the quality you know accuracy  all those you know recommendation great question [12:03] so there's a few different dimensions to it um one  of the things we didn't get into is how we built [12:08] uh for the first uh six years of the company's  history, we spent a lot of time compiling a [12:13] massive collection of clinical case studies and  longitudinal information where we could see how [12:18] patients were treated and the diagnosis that  emerged from different sorts of of investigative [12:23] processes. So we have the largest medical  textbook assembled in the country uh I can say [12:31] uh about 10.2 2 million case studies. And so we  evaluate our technologies accuracy against those [12:38] real life visits and not just the diagnosis  accuracy but the investigation accuracy. It's [12:44] a funny thing that diagnosis if you have a  full case study is actually pretty easy. The [12:48] hard thing is what you do which is how do you go  through the investigative process to understand [12:52] because the way that a disease manifests can  be in so many different ways, right? Is it [12:57] something serious or is it something light? That  is a question and answer process you have to to [13:01] navigate. So that's one. The second thing is uh  we have physicians in the loop for this purpose [13:07] to provide a final set of oversight to any order  that gets submitted. Um and that provides us not [13:13] only a learning opportunity if a physician is  going to be doing something different but also [13:17] compliance oversight and regulatory coverage  around the diagnosis that are emerging from [13:22] our technology. Does that answer your question?  Yes. Who pays for it? Insurance. [laughter] The [13:30] goal of our system was let me start over.  Our perspective is that autonomous health [13:38] is going to be an inevitable future. You cannot  train enough doctors to solve the problem that [13:44] you're referring to. It's just not possible. We  also feel strongly that autonomous health needs [13:48] to be brought into the world in today's payment  environment, in today's regulatory environment, [13:54] not some extra burden that a patient has to  take on just to get access to something better. [13:59] So we've designed this system so it works within  the existing insurance world. Medicare, Medicaid, [14:05] commercial insurance pays for is the objective to  replace clinicians? Totally. [laughter] There's [14:13] always the AI question, right? The objective is  to most effectively optimize where physicians, [14:21] providers, and other workers that are handling  these really challenging jobs need to spend their [14:26] time. Right now, we aren't able to do that. I  think that there's uh replacing we're far off [14:32] from replacing because we just don't have enough  at the moment. So, we have to uh we you know we're [14:38] trying to multiply that capacity and make sure  that you know all of the physicians that we have, [14:44] all the other providers we have can uh spend  their time on where it matters the most.