## Background
Published by Ai4 on YouTube on August 24, 2025, this talk features Joshua Otten, Associate Vice President and Chief Technology Officer at Adventist HealthCare. The presentation outlines how healthcare leaders can organize leadership around AI governance, charter dedicated committees, establish usage policies, and identify targeted operational use cases.
## TL;DR
Joshua Otten presents a practical roadmap for establishing proactive AI governance in healthcare systems through a joint IT and Legal co-chair model. The framework focuses on formulating unified definitions, modernizing data privacy and vendor agreements, and cataloging AI application inventories to satisfy regulatory bodies like The Joint Commission and CMS. Once governance guardrails are set, organizations can strategically deploy AI to address clinician burnout with ambient scribes, automate revenue cycle workflows, and implement facility computer vision. Otten also emphasizes operating as a commercial off-the-shelf (COTS) 'buy shop' rather than building custom in-house AI labs.
## Atmosphere & style
The presentation is refreshingly pragmatic, candid, and approachable, opening with self-deprecating humor as Otten dispels the pretense of being an AI guru in favor of grounded operational leadership. Rather than indulging in Las Vegas tech hype, the narrative purposefully champions unglamorous governance, using relatable metaphors like bowling alley bumpers to demystify complex compliance topics. The delivery feels like an earnest, peer-to-peer briefing tailored to help healthcare leaders navigate the technological tsunami with sensible, real-world execution.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) AI governance is an urgent organizational imperative for healthcare leaders navigating a massive technological revolution, even though it lacks the glamour of frontier AI tools.
- [03:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=184s) In the absence of comprehensive federal AI regulations, health systems must establish proactive internal guardrails through a dedicated AI governance committee.
- [03:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=184s) The AI governance committee should be co-chaired by IT and Compliance/Legal, bringing together HR, Finance, Revenue Cycle, Mission Integration, and clinical leadership.
- [03:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=184s) A foundational first task for any AI governance initiative is formulating a unified, enterprise-wide definition of artificial intelligence.
- [06:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=377s) Organizations need to establish dedicated learning cadences and modernize foundational documentation, including data privacy policies and vendor Business Associate Agreements (BAAs).
- [06:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=377s) Establishing clear AI usage policies functions like bumpers in a bowling alley, keeping staff safe and preventing the risks of shadow AI.
- [09:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=557s) Maintaining a comprehensive AI application inventory is critical for meeting regulatory survey requirements from accrediting bodies like The Joint Commission and CMS.
- [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) AI initiatives should directly align with long-term strategic plans (such as institutional 2030 strategy maps) rather than adopting technology for its own sake.
- [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) Governance frameworks provide an effective lens to dust off and solve legacy operational bottlenecks that previously required extensive manual workarounds.
- [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) Ambient digital scribes represent healthcare's biggest immediate AI opportunity by curbing clinician burnout and eliminating late-night administrative charting ('pajama time').
- [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) High-value healthcare use cases span Revenue Cycle Management automation, HR candidate screening, clinical chart synthesis, and computer-vision-based patient fall detection.
- [18:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1096s) Health systems must formally define their build-versus-buy identity, with many non-tech provider organizations succeeding as COTS (commercial off-the-shelf) software buyers.
- [18:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1096s) Extending ambient patient monitoring from inpatient acute rooms to residential home health settings represents a major frontier for continuous, scalable healthcare.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) Introduction and the Healthcare Technology Tsunami
Joshua Otten opens his presentation introducing his background as the Associate Vice President and Chief Technology Officer at Adventist HealthCare, a provider-based healthcare system operating just outside Washington, D.C., in Montgomery County and Prince George's County, Maryland. The healthcare system encompasses three acute-care inpatient hospitals, multiple inpatient rehabilitation hospitals, numerous outpatient clinics, and a wide network of primary and specialty care practices across the region. Setting a candid and self-deprecating tone, Otten jokes about his preparation for the keynote: after turning to ChatGPT and Google to research how to deliver a confident presentation, every guide told him he needed to immediately establish himself as an expert.
Dispelling any pretense, Otten remarks, "Guys, I'm here to tell you I am not an expert in AI. I am far from an expert in AI." Instead, he grounds his credibility in his deep domain experience evaluating, selecting, and implementing enterprise healthcare technology. Otten observes that technology leaders across industries are currently staring down an "absolute tsunami of technological revolution." Acknowledging that the event is hosted in Las Vegas—where AI hype is "about as sexy as you can get"—he intentionally leads with governance, recognizing that while governance is inherently unglamorous and dry, it is an urgent organizational imperative rather than an optional safeguard.
- **Organization Profile**: Adventist HealthCare (Montgomery County and Prince George's County, MD).
- **Facilities**: Three acute-care inpatient hospitals, two inpatient rehabilitation hospitals, outpatient clinics, and specialty/primary care facilities.
- **Core Stance**: AI governance is not optional, even if it lacks the glamour of frontier AI tooling.
---
[03:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=184s) Assembling the Governance Team and Establishing a Baseline Charter
Because the federal government has yet to produce comprehensive AI guardrails—leaving individual states like California to draft their own regulations—organizations must proactively build their own internal guardrails. Otten jokes that while he tries to avoid using the "C-word" because healthcare organizations tend to "committee everything," establishing a dedicated governance committee is essential. At Adventist HealthCare, the committee is structured as a joint co-chair model between Information Technology and the Compliance and Legal departments (including General Counsel), ensuring technical feasibility aligns with regulatory compliance.
Beyond IT and Legal, the governance team must include representation across critical organizational functions: Human Resources, Finance, Revenue Cycle (identified as a major area for AI disruption), Mission Integration, and frontline clinical leaders including physicians and nursing leadership. Otten emphasizes that the committee's immediate first task must be chartering and formulating a single, unified definition of artificial intelligence for the enterprise. Without an agreed-upon definition, asking a room of leaders for their understanding would produce a hundred conflicting answers. Since most hospital staff and committee members are starting from the same baseline knowledge, learning together through vendor partnerships, external networking, and cross-functional discussions is essential.
```mermaid
flowchart TD
accTitle: AI Governance Committee Structure and Stakeholders
subgraph Leadership ["Committee Co-Chairs"]
IT["Information Technology"]
COMP["Compliance & General Counsel"]
end
- **Committee Co-Chairs**: Information Technology and Compliance / General Counsel.
- **Key Stakeholders**: HR, Finance, Revenue Cycle, Physician Leaders, Nursing Leaders, and Mission Integration.
- **First Step**: Formulate a clear organizational charter and establish a single operational definition of AI across all departments.
---
[06:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=377s) Educational Cadence, Policy Modernization, and Guardrails
Adventist HealthCare’s governance committee meets on a monthly cadence dedicated to continuous education and policy alignment. To build baseline competency, the organization brought in external research firms such as Gartner to provide governance frameworks, alongside outside legal counsel to conduct PESTLE analyses addressing the evolving political and legislative environment. Otten notes that the core operational responsibility of the committee lies in modernizing foundational documentation: reviewing data privacy policies, refining data handling guidelines, and updating Business Associate Agreements (BAAs) with technology vendors to account for AI-specific exposure.
The primary objective of these reviews is publishing a comprehensive, organization-wide AI policy that clearly outlines explicit "dos and don'ts." In the absence of enterprise-provided tools or clear boundaries, employees inevitably turn to free consumer solutions like personal ChatGPT accounts. Otten compares governance policies to the "bumpers in the alley for bowling," designed to keep staff safe while handling sensitive organizational and patient data. This proactive governance also ensures compliance across external regulatory bodies, including HIPAA regulations, CMS site visits, and The Joint Commission surveys.
- **Meeting Rhythm**: Monthly dedicated sessions focused on learning, regulatory assessment, and policy formation.
- **External Advisors**: Gartner (governance frameworks) and external counsel (PESTLE and legal landscape analysis).
- **Core Policy Targets**: Data privacy, data handling, vendor BAAs, and a standalone enterprise AI usage policy.
- **Safety Metaphor**: Governance functions like bowling alley bumpers, guiding safe usage and preventing shadow AI data leaks.
---
[09:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=557s) Maintaining an Application Inventory and Approaching Use Cases
A critical, practical output of AI governance is creating and maintaining an enterprise AI application inventory. While cataloging applications across a hospital system might seem administrative or trivial, regulatory oversight bodies like The Joint Commission have already begun asking hospitals during surveys and site inspections to produce inventories of AI-enabled systems and explain how they impact clinical patient care. Failing to track which vendor software or internal tools utilize AI models leaves healthcare organizations vulnerable to citations.
Otten advises leaders to immediately comb through their existing software portfolios and application catalogs to identify embedded AI functionality, determine how data is being processed, and document clinical use cases. Establishing this portfolio baseline prepares the health system for incoming regulatory audits from CMS, The Joint Commission, and emerging state and federal mandates. Having laid the foundation of governance, committee building, policy setting, and application tracking, Otten pivots toward the tangible applications of the technology, declaring that it is time to "bring sexy back" by exploring high-impact healthcare use cases and solving operational problems.
- **Inventory Imperative**: The Joint Commission is now actively asking for AI application inventories and clinical impact documentation during hospital survey visits.
- **Audit Preparedness**: Proactive cataloging prevents regulatory citations from CMS and healthcare accrediting bodies.
- **Next Phase**: Transitioning from foundational governance frameworks into identifying high-value clinical and operational use cases.
[10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) Aligning AI Initiatives with Strategic Goals and Legacy Pain Points
Organizations do not need to reinvent the wheel when looking for AI use cases; instead, they should start directly from their established enterprise strategy. At Adventist HealthCare, this involves referencing a laminated strategy map running through 2030, segmented across all entities, hospitals, and service lines. By taking these high-level organizational goals off the desk and decomposing them into actionable initiatives, teams can identify precise areas where AI can serve as the enabling mechanism.
```mermaid
flowchart LR
accTitle: Strategic Alignment and Problem Identification Flow
subgraph Strategy["Strategic Foundation"]
A["Strategy Map (Through 2030)"] --> B["Decompose into Initiatives"]
end
subgraph Analysis["Governance Lens"]
B --> C["Evaluate with AI Governance Lens"]
D["Dust Off Persistent Legacy Problems"] --> C
end
subgraph Execution["Actionable Outcomes"]
C --> E["Identify High-Impact AI Use Cases"]
E --> F["Deploy Targeted AI Tools"]
end
```
Beyond top-down strategic planning, AI governance frameworks serve as a diagnostic lens for persistent, unresolved operational headaches. Every organization has chronic problems that have haunted operations for decades—issues managed through 10 to 15 layered workarounds or by throwing 50 full-time equivalents (FTEs) at manual tasks. Leaders should dust off these legacy bottlenecks and re-evaluate them using the latest advancements in agentic AI. As leadership teams deepen their technical understanding through continuous learning sessions, they can view AI as "another arrow in your quiver" to drive systematic, continuous improvement across the health system.
- Use the organizational 2030 strategy map to break broad institutional priorities down into specific AI-driven roadmaps.
- Re-examine long-standing legacy bottlenecks where dozens of FTEs were historically added as workarounds.
- Treat AI solutions not as an end in themselves, but as functional tools within an ongoing continuous improvement program.
---
[13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) High-Impact Healthcare Use Cases: Scribes, Rev Cycle, and Computer Vision
Ambient digital scribes represent the single biggest AI opportunity in healthcare today. With clinicians and physicians suffering severe post-pandemic burnout and spending hours on late-night administrative charting—often referred to as "pajama time" after putting their kids to bed—ambient AI tools can passively listen to patient encounters and generatively draft clinical documentation for provider review and sign-off. Major EHR vendors are actively working to integrate these ambient capabilities natively into their clinical platforms.
```mermaid
flowchart TD
accTitle: Healthcare AI Use Case Landscape
subgraph Clinical["Clinical Support & Care"]
A["Ambient Digital Scribes<br/>(Eliminates 'Pajama Time')"]
B["Clinical LLM Assistants<br/>(Chart Summarization & Themes)"]
C["Virtual Care & Ambient Sensors<br/>(Fall Detection & Rail Alerts)"]
end
subgraph Operations["Operational & Administrative"]
D["Revenue Cycle Automation<br/>(Payer Claims & Documentation)"]
E["HR Recruitment Screening<br/>(Resume Matching to Jobs)"]
F["Hospital Computer Vision<br/>(ED Waiting Rooms & Supply Stock)"]
end
```
Significant potential exists across operations, administration, and acute inpatient care:
- **Revenue Cycle Management (RCM):** Automating the arduous, manual workflow where staff log into payer websites, pull claim statuses, re-enter details into the EHR, and retrieve clinical documentation to upload back to insurers.
- **HR & Talent Acquisition:** Leveraging AI to parse resumes against strict job requirements to accelerate hiring pipelines.
- **Virtual Care & Ambient Patient Monitoring:** Deploying in-room computer vision and sensors that integrate with smart beds to detect fall risks (such as dropped bed rails or agitated movement) and instantly alert nursing staff through mobile communications.
- **Clinical Chat Assistants:** Equipping physicians and nurses with LLM-powered virtual assistants to synthesize voluminous patient charts and surface critical clinical themes.
- **Facility Computer Vision:** Utilizing single-camera feeds to monitor emergency department (ED) waiting room throughput, track hallway traffic, and conduct automated inventory reconciliation in supply rooms against existing stock.
---
[18:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1096s) Q&A: Build vs. Buy Strategy and Expanding AI Monitoring into the Home
During the concluding Q&A session, an attendee asked how a healthcare system—not being a dedicated technology firm—can keep up with the daily, rapid evolution of AI without building a massive internal AI research lab. The speaker clarified that the organization operates strictly as a "buy shop" (COTS/commercial off-the-shelf software) rather than a custom development organization. Governance committees must explicitly define their organizational identity early on: Adventist HealthCare relies directly on trusted enterprise vendor partners rather than maintaining an in-house engineering lab.
```mermaid
flowchart TD
accTitle: Build vs Buy and Care Continuum Evolution
subgraph Governance["Strategic Approach"]
A["Healthcare Governance Decision"] --> B["COTS / Buy Shop Strategy<br/>(Rely on Vendor Partnerships)"]
end
subgraph Expansion["Care Continuum"]
C["Inpatient Hospital Monitoring"] --> D["Skilled Nursing & Rehab Facilities"]
D --> E["In-Home Residential Care<br/>(Home Health Monitoring Opportunity)"]
end
```
A final question from an attendee from Aerospike addressed the potential for expanding ambient hospital monitoring into residential patient settings. The speaker confirmed active leadership discussions with their rehabilitation service lines and home health teams. While sensor monitoring exists across skilled nursing facilities, consumer-grade, residential-level continuous monitoring remains a major unaddressed market opportunity that health systems are evaluating to support visiting home-health nurses and remote patient care.
- Health systems must formally establish whether they are a "build" or "buy" organization; Adventist HealthCare operates primarily as a commercial off-the-shelf (COTS) shop partnering with specialized tech vendors.
- Leaders should actively network and invite vendors with creative solutions to present use cases to AI governance committees for informed decision-making.
- Extending ambient monitoring from inpatient rooms and rehab centers into residential home healthcare represents a vital frontier for health systems seeking scalable, continuous patient care.
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) Guys, I'm here to tell you I am not an expert in AI. I am far from an expert in AI. — Joshua Otten
> [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) absolute tsunami of technological revolution — Joshua Otten
> [06:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=377s) bumpers in the alley for bowling — Joshua Otten
> [09:17](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=557s) bring sexy back — Joshua Otten
> [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) another arrow in your quiver — Joshua Otten
> [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) pajama time — Joshua Otten
The talk will focus on how to mobilize and organize your leadership around the governance of AI technologies. The attendees will be given insight into the journey of chartering a governance committee and how to address some of the following pervasive questions: What is our organization's stance on the use of AI? What policies do we need to put in place around AI? Who should have the decision-making authority? How do we keep pace of AI advancements and identify targeted business use cases?
联合委员会审查系统之际，Adventist HealthCare 制定 AI 规则
各医院若不在联邦标准出台前由跨职能领导团队盘点临床软件并建立严格的使用护栏，将面临监管处罚风险。
Joshua Otten 提出了一个实用路线图，旨在通过 IT 与法务联合牵头的模式，在医疗系统中建立前瞻性的 AI 治理机制。该框架重点关注制定统一定义、更新数据隐私与供应商协议，并梳理 AI 应用清单，以满足联合委员会（The Joint Commission）和 CMS 等监管机构的要求。一旦治理护栏设立完成，机构便可开展战略性 AI 部署，通过环境语音记录工具（ambient scribes）缓解临床医生倦怠、自动化收入周期工作流，并应用设施计算机视觉。Otten 还强调，机构应作为商用现成产品（COTS）的“采购方”运作，而非在内部自建定制化 AI 实验室。
面对技术变革且缺乏联邦统一规范的现状，医疗系统必须建立由 IT 和合规/法务部门共同牵头的 AI 治理委员会，统一人工智能定义并制定内部准则。 (3:04)
治理委员会通过定期研讨更新数据隐私政策与商业合作协议（BAA），并出台明确的“使用守则”，以类似保龄球缓冲轨的机制防范影子 AI 带来的合规与安全风险。 (6:17)
随着联合委员会（The Joint Commission）与 CMS 等监管机构启动审查，医疗机构需建立并维护全面的 AI 应用资产清单，清晰梳理其对临床护理的影响以避免受罚。 (9:17)
AI 落地应紧密对齐机构至 2030 年的长期战略目标，重点聚焦于消除临床医生文书负担的环境智能听写、收入周期管理自动化以及患者跌倒防范等高价值场景。 (10:50)
非科技主导型医院在治理初期应确立“采购而非自研（COTS）”的策略，依托成熟供应商推进落地，并积极探索将院内环境传感监测扩展至家庭护理场景。 (18:16)
Reading summary
Joshua Otten, Chief Technology Officer at Adventist HealthCare, argues that healthcare systems must proactively build cross-functional governance, updated legal guardrails, and comprehensive software inventories before adopting artificial intelligence. Rather than attempting custom internal engineering, health systems should pursue commercial off-the-shelf tools to target persistent operational bottlenecks and relieve clinician documentation burdens.
Cross-Functional Governance and Enterprise Policy Guardrails (3:04)
In the absence of comprehensive federal AI regulations, Otten stresses that healthcare organizations must proactively create their own internal guardrails. Adventist HealthCare addressed this by forming an AI governance committee co-chaired by Information Technology and Compliance and Legal, including General Counsel. To ensure cross-disciplinary alignment, the committee integrates representatives from Human Resources, Finance, Revenue Cycle, Mission Integration, and frontline physician and nursing leadership. Otten emphasizes that the committee's initial priority must be defining a single, unified enterprise definition of AI to establish a shared baseline across leadership.
Meeting monthly, the committee works with external advisory partners, leveraging Gartner for governance frameworks and outside legal counsel for political and regulatory PESTLE analyses. The primary operational output involves modernizing data privacy protocols, refining data handling guidelines, updating vendor Business Associate Agreements (BAAs), and publishing an enterprise AI usage policy. Otten compares these policies to bowling alley bumpers, designed to prevent data breaches caused by employees turning to unvetted consumer tools like personal ChatGPT accounts.
AI Application Inventories for Regulatory Scrutiny (9:17)
A critical tactical responsibility of healthcare governance is assembling and maintaining an enterprise AI application inventory. Otten reports that oversight organizations such as The Joint Commission have already begun asking hospitals during routine survey visits to produce inventories of AI-enabled systems and detail their specific clinical impacts on patient care.
Failing to catalog AI components embedded in third-party or internal software leaves health systems vulnerable to survey citations. Otten advises healthcare leaders to immediately audit their software portfolios to determine how data is handled and where clinical AI operates, ensuring institutional preparedness for incoming audits from CMS, accrediting bodies, and state regulatory authorities.
Directing AI Toward Strategic Priorities and Bottlenecks (10:50)
Otten maintains that health systems should not invent novel use cases in isolation, but rather anchor AI deployments directly to existing enterprise strategy. Adventist HealthCare decomposes its organizational strategy map running through 2030 across hospitals and service lines, using AI as an enabling mechanism for predefined institutional objectives.
Governance frameworks also serve as an analytical tool to address chronic, unresolved operational bottlenecks. Healthcare organizations often manage persistent operational hurdles by layering ten to fifteen workarounds or dedicating up to 50 full-time equivalents to manual administrative tasks. Otten suggests using advancements in agentic AI as a functional tool within ongoing continuous improvement programs to systematically eliminate these legacy pain points.
Clinical Documentation and Operational Automation Opportunities (13:16)
Otten identifies ambient digital scribes as the single largest AI opportunity in healthcare today. By passively recording clinical encounters and drafting notes for provider sign-off, ambient listening tools aim to mitigate post-pandemic clinician burnout and eliminate evening documentation, commonly termed 'pajama time.' Major electronic health record vendors are currently working to integrate these capabilities natively into clinical workflows.
Beyond direct clinical care, high-impact opportunities exist across operational and administrative domains. In Revenue Cycle Management, automation can replace manual workflows where staff log into payer websites, pull claim statuses, re-enter data into EHRs, and retrieve clinical documentation for insurers. Other near-term applications include parsing resumes against rigid job criteria in HR, using computer vision and smart bed sensors to alert nurses to patient fall risks, deploying LLM assistants to synthesize complex medical charts, and utilizing single-camera feeds for emergency department throughput and supply room inventory reconciliation.
Commercial Tool Adoption and Residential Care Expansion (18:16)
Given technical and capital constraints, Otten clarifies that Adventist HealthCare operates strictly as a commercial off-the-shelf (COTS) 'buy shop' rather than maintaining an internal AI engineering lab. He recommends that health systems establish this strategic identity early and actively invite external technology vendors to present tailored use cases directly to governance committees.
Looking toward future deployment frontiers, Otten highlights extending ambient monitoring into residential home healthcare as a significant unaddressed market opportunity. While sensor monitoring is established in acute and rehabilitation facilities, deploying ambient monitoring into home environments could provide continuous clinical visibility to support visiting home-health nurses and expanding remote care programs.
Adventist HealthCare's operational model demonstrates that healthcare AI adoption relies more on rigorous cross-functional governance, regulatory audit preparation, and commercial vendor partnerships than on internal software development. However, realizing long-term opportunities like residential ambient monitoring remains contingent on resolving vendor integration challenges and navigating fragmented state-level AI regulations.
## Background
该视频由人工智能行业峰会频道 Ai4 于 2025 年 8 月 24 日发布在 YouTube 平台。内容为 Adventist HealthCare 助理副总裁兼首席技术官 Joshua Otten 的主题演讲，主要探讨医疗机构如何组建领导层跨职能治理委员会、制定 AI 使用政策与安全护栏，并结合战略目标发掘切实可行的业务用例。
## TL;DR
Joshua Otten 结合一线医院运营实践，指出应对生成式 AI 浪潮的首要任务是搭建由 IT、合规、临床等多部门组成的治理委员会，统一认知并建立全院 AI 资产清单以应对监管飞行检查。在落地路径上，组织应依托长期战略地图与繁琐低效的业务顽疾作为切入点，重点推进解决医生职业倦怠的环境数字化病历记录（Ambient Scribe）、收入周期理赔自动化以及院内虚拟护理与计算机视觉监测。在战略定位上，他明确非科技类医疗机构应保持务实的“采购型组织（Buy shop）”路线，并展望了未来居家监测场景的蓝海机遇。
## Atmosphere & style
整场演讲风格务实坦诚且幽默风趣，讲者开场即幽默自嘲“并非 AI 专家”以打破行业神话，极具亲和力。叙事层层递进，将看似枯燥严肃的“治理与合规护栏”讲解得通俗易懂，随后又自然切换到充满活力与温度的一线临床场景。整场分享洋溢着医疗 IT 领袖在面对技术变革海啸时的审慎敬畏与实干探索精神，现场互动问答交流积极且极富实操参考价值。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) 面对 AI 技术浪潮，医疗机构的痛点不仅在于算法本身，更在于医疗场景下的技术评估、选型与合规落地。
- [02:44](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=164s) 在缺乏统一联邦监管的背景下，医疗机构必须自建跨职能 AI 治理委员会，由 IT 与合规部门联合牵头，涵盖临床、财务、HR 与使命整合代表。
- [04:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=290s) AI 治理委员会的首要任务是统一组织内部对 AI 的定义与语汇，并通过 Gartner 培训、外部法律研判及同行交流建立集体学习机制。
- [06:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=410s) 组织必须重构数据隐私政策与 BAA 商业合作协议，出台明确的组织级 AI 使用守则，通过提供受控的企业级环境杜绝“影子 AI”泄密风险。
- [08:47](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=527s) 联合委员会与 CMS 已开始在飞行检查中核验 AI 应用，医院需建立并动态维护全院 AI 资产清单以防范合规传票风险。
- [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) 推进 AI 落地应以组织既定战略目标为切入点，并用治理透镜重新审视那些耗费大量人工替代方案的长期运营顽疾。
- [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) 环境数字化记录员（Ambient Digital Scribe）能实时生成病历摘要，有效缓解医护人员后疫情时代的职业倦怠与“睡衣时间”。
- [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) 收入周期管理（Revenue Cycle）中跨保险公司平台的繁琐理赔核对流程，非常适合引入自动化智能体（Agentic AI）降本增效。
- [15:46](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=946s) 院内虚拟护理结合环境监测与计算机视觉，可实现跌倒高危行为预警、急诊科人流吞吐管理与耗材室动态盘点。
- [18:49](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1129s) 非科技型医疗机构需在治理初期明确“采购型”（COTS shop）定位，紧密依赖成熟供应商而非盲目自研底层系统。
- [18:49](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1129s) 针对居家护理（Home Care）的患者环境监测技术目前仍处于市场蓝海，存在显著的供给空白与落地潜力。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) 医疗体系背景与坦诚告白：面对 AI 技术海啸
作为 Adventist HealthCare 的助理副总裁兼首席技术官（AVP, CTO），Joshua Otten 首先介绍了自己所在的医疗系统。Adventist HealthCare 坐落于美国马里兰州蒙哥马利县（Montgomery County），服务范围覆盖紧邻华盛顿特区的蒙哥马利县与乔治王子县（Prince George's County）。这是一家以医院运营为核心的综合性医疗服务系统，旗下运营着三所急症住院医院、两所住院康复医院，以及遍布社区的众多门诊诊所、初级保健和专科门诊。Joshua 表示，在全行业都在探寻人工智能将如何颠覆业务流程和患者护理的当下，能够在此分享一家一线医疗机构的真实探索视角，他感到无比荣幸。
在准备这次演讲的过程中，Joshua 幽默地分享了一段插曲：他像所有人一样打开了 ChatGPT、Google，用遍了手头的各种 AI 工具去请教“如何在演讲开场树立自信且值得信赖的权威形象”。AI 给出的所有建议出奇一致——必须在开场白中将自己确立为该领域的专家。然而 Joshua 随即对全场坦白道：“各位，我必须老实告诉你们，我根本不是什么 AI 专家，甚至差得十万八千里。”但他话锋一转，指出自己真正的专长在于“医疗场景下技术的评估、选型、落地实施以及伴随的所有配套事务”。面对当前全行业瞩目的生成式 AI 浪潮，他毫不讳言，包括医疗同仁与科技领袖在内的所有人，此刻正共同直面一场扑面而来的“科技革命海啸”（tsunami of technological revolution）。
- **机构概况**：服务于马里兰州蒙哥马利县与乔治王子县（近华盛顿特区）。
- **医疗资产规模**：3 所急症住院医院、2 所住院康复医院，以及遍布社区的门诊与专科诊所。
- **讲者核心定位**：非单纯的算法研究专家，而是深耕医疗信息化评估、选型、部署与落地的实施专家。
---
### [02:44](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=164s) 破除“无趣”偏见：为什么必须搭建 AI 治理委员会
Joshua 坦言，在拉斯维加斯举办的顶尖 AI 峰会上，最吸引眼球的往往是各种炫酷前沿的黑科技，“治理（Governance）这一点都不性感”。但他之所以在幻灯片第一页就抛出治理，是因为在眼下的节点，它早已不再是可有可无的选修课。由于联邦层面尚未建立起统一且明确的行业监管护栏，除加利福尼亚等少数正自行立法的州之外，绝大多数机构必须自己动手搭建“安全边界”。他调侃道，自己在选词上刻意避开了“委员会（Committee）”这个“C 开头的脏话”，因为在医疗体系里大家习惯把任何鸡毛蒜皮的事都做成委员会，但归根结底，这就是一个不可或缺的领导治理团队。
组建这一跨职能治理团队时，必须横向穿透整个组织的关键业务职能部门。在 Adventist HealthCare 的实践中，该治理团队由 Joshua 所在的 IT 部门与合规部门（Compliance Department）联合担任主席，总法律顾问与合规领导人亲自参与把关与决策。在此基础上，委员会吸纳了人力资源（HR）、财务、营收周期管理（Revenue Cycle，AI 降本增效的核心阵地），以及至关重要的临床一线力量——主任医师、护士长等业务骨干。此外，由于机构具备教会与社区服务渊源，团队还特别引入了“使命整合（Mission Integration）”代表，以确保技术采纳与机构的道德使命和价值观保持一致。
```mermaid
flowchart TD
accTitle: AI Governance Committee Structure
subgraph CoChairs["联合领导层 (Co-Chairs)"]
IT["信息技术部 (IT)"]
COMP["合规与法律部 (Compliance & Legal)"]
end
subgraph Clinical["临床代表 (Clinical)"]
DOC["医师领导团队 (Physicians)"]
NUR["护理领导团队 (Nursing)"]
end
subgraph Ops["业务运营职能 (Operations)"]
FIN["财务与营收周期 (Revenue Cycle)"]
HR["人力资源部 (HR)"]
MIS["使命整合部 (Mission Integration)"]
end
CoChairs --> Clinical
CoChairs --> Ops
```
---
### [04:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=290s) 统一语言底座与集体学习：建立委员会章程与教育机制
组建团队之后的第一项关键任务，是为委员会订立正式章程（Charter），而其中的首要步骤就是明确“人工智能”的内部统一定义。Joshua 指出，如果在台下分发小纸条让大家写下 AI 的定义，很可能会收上来上百种五花八门的答案。在一个组织内部，如果大家不在同一张乐谱上哼唱、不在同一个认知层面上沟通，协同治理将无从谈起。绝大多数传统医疗机构内部并不具备大量原生的人工智能专家（除非自建了专门的 AI 研发实验室），绝大多数成员都站在完全相同的零起点，因此统一步调不仅是消除认知偏差的前提，更是团队凝聚共识的宝贵历程。
在此基础上，治理委员会的早期核心职责是“共同学习”（Learning together）。Adventist HealthCare 选择主动盘活内部资源与外部生态，包括邀请咨询机构 Gartner 开展治理框架搭建培训、聘请外部专业法律顾问进行针对 AI 发展宏观环境的 PESTLE 分析等。Joshua 强调，AI 治理绝非关起门来闭门造车，这也是他专程来到行业大会现场的原因——与同行积极交流医疗场景下的治理机制、评估标准与业务用例，并将前沿思路带回内部委员会的月度例会中，持续推动组织知识的更新迭代。
- **统一定义**：打破信息差，确保跨部门领导在沟通技术需求与风险时使用相同语汇。
- **外部智囊赋能**：引入 Gartner 梳理治理框架，联动外部法务开展政策与法律层面的 PESTLE 研判。
- **运作机制**：每月定期召开例会，将早期阶段的核心重心放在全员赋能、跨界交流与规则框架搭建上。
---
### [06:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=410s) 筑牢保龄球道护栏：从数据隐私到统一 AI 使用政策
谈及治理团队最繁重、最核心的“主菜”工作，Joshua 将其概括为政策的全面审查、修订与立新。伴随生成式 AI 工具的泛滥，机构原有的数据治理与隐私政策必须全盘重构，使其包含明确针对 AI 交互场景的限定条款。与此同时，医疗行业中与外部软件供应商签订的商业合作协议（Business Associate Agreements, BAA）也必须全面重新审视，确保涉及受保护健康信息（PHI）的数据流动与算法处理合规透明。而最终的落脚点，是制定并面向全员发布一套独立的、清晰明确的组织级 AI 使用政策，详细列出“可为与不可为”的红线清单。
Joshua 形象地打了个比方：如果机构不主动提供带有安全边界的企业级环境（例如企业版 Copilot 或企业版安全 ChatGPT），员工为了解决手头工作，势必会自发转向公开的免费工具。“这就像打保龄球时安装在两侧的防撞护栏（guardrails/bumpers）”，治理的目的不是扼杀创新，而是建立物理隔离，确保任何一线人员在接触和处理敏感数据时都在安全受控的航道内进行。面对监管滞后与技术狂飙的现实，机构必须先于外部强制指令为自身系好安全带。
- **政策翻新**：审查并更新既有数据处理与数据隐私政策，加入对 AI 大模型输入/输出的新约束。
- **供应商合规**：重新校准 BAA 商业合作协议，封堵第三方 AI 供应商对受保护健康信息（PHI）的隐蔽违规使用。
- **行为指引**：出台全院统一的 AI 使用守则，明确免费公共 AI 工具与企业合规环境的界限，杜绝“影子 AI”泄密风险。
---
### [08:47](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=527s) 应对飞行检查：摸清家底与维护应用资产清单（Inventory）
随着政策层面的合规要求不断下沉，医疗行业所面临的外部监管压力正急剧放大。除了必须遵守联邦层面的 HIPAA 隐私法案，各大医院还要接受美国医疗保险和医疗补助服务中心（CMS）以及联合委员会（The Joint Commission）的定期现场飞行调查。这些监管机构正在将目光转向医院内部的 AI 落地现状。对此，Joshua 提出了一个看似简单、甚至有些微不足道，但实则至关重要的行动指南：建立并动态维护一份全院“AI 资产清单”（AI Application Inventory）。
绝大多数医院在不知不觉中，其采购的系统或日常业务流里已深度嵌入了各类 AI 组件。IT 与治理团队必须全面排查应用资产目录，将所有涉及 AI 技术的应用逐一剥离、梳理并记录在案——明确每一项 AI 工具的部署位置、实际用途及其对患者护理产生的具体影响。Joshua 警示道，在联合委员会近期的实地审查中，审查员已经开始主动索要 AI 清单，若机构因底账不清、一问三不知而招致合规传票（Citation），后果将极其严重。建立清单不仅是应对审计的“防火墙”，也是未来组织推行一切高阶用例的前提。在理清了治理、政策与合规底座之后，Joshua 兴奋地宣布告别枯燥的合规部分，正式带领全场重回充满吸引力的业务现场：“接下来是真正有趣的部分，我们要把性感的创新带回来了——探讨那些亟待解决的业务用例与现实痛点！”
- **监管风向转变**：联合委员会（Joint Commission）与 CMS 在医院现场审查中已开始核验 AI 工具的应用情况。
- **清单核心要素**：覆盖现有软件库中所有 AI 属性产品，标记其技术归属、应用部门、在患者诊疗链条中的介入程度。
- **防范违规罚单**：无论治理体系是否完备，摸清软件资产底账是必须立刻推行的避坑底线。
### [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) 借助战略目标与顽疾问题确立 AI 落地切入点
推进 AI 落地并不意味着需要“重新发明轮子”（reinvent the wheel），最核心的出发点应当是企业自身的战略目标。Joshua Otten 分享了自己组织的实际做法：他们拥有一套持续更新至 2030 年的战略地图（strategy map），细分到了系统内的所有实体医院与业务线（service lines）。这本塑封精美的战略手册常年摆放在他的办公桌上，随时可以翻阅。在实际推行技术时，治理团队所做的就是从中拆解出具体的战略目标，并将其转化为可落地的行动举措，评估如何利用 AI 技术作为解决方案来达成这些既定目标。
此前建立的治理框架与学习成果，正是审视业务的“全新镜头”（lens）。除了对齐宏观战略目标外，更关键的是用这一视角去重新审视组织内部“长期存在的顽疾”（persistent problems）。任何组织内部都存在一些困扰多年的老问题，可能在过去 20 年里衍生出了 5 到 15 个临时替代方案（workarounds），甚至专门为此堆叠了 50 名全职员工（FTEs）来进行人工修补。管理者应该“拂去这些陈旧问题上的灰尘”，结合智能体 AI（Agentic AI）等前沿技术演进重新评估解决途径。归根结底，AI 只是工具箱中的“另一支箭”（another arrow in your quiver），组织需要做的是保持持续改进与常态化审查。
```mermaid
flowchart TD
accTitle: AI 落地切入点选择与落地路径
subgraph S1["输入层：业务与战略源头"]
A["2030 战略地图<br/>(分解为行动举措)"]
B["组织长期顽疾<br/>(耗费大量 FTE 的繁琐流程)"]
end
subgraph S2["评估层：AI 治理与技术透镜"]
C["AI 治理委员会评估框架"]
D["Agentic AI / 生成式 AI 认知"]
end
subgraph S3["输出层：高价值落地场景"]
E["环境数字化记录<br/>(Ambient Scribe)"]
F["收入周期与理赔处理<br/>(Rev Cycle)"]
G["院内虚拟护理与监测<br/>(Virtual Care)"]
end
A --> C
B --> C
D --> C
C --> E
C --> F
C --> G
```
### [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) 医疗核心应用场景：环境数字化记录与收入周期自动化
在医疗行业中，当前热度最高、落地最为迫切的场景当属“环境数字化记录员”（Ambient Digital Scribe）。经历过 2020 年疫情的冲击，医护人员在后疫情时代正面临严重的职业倦怠（burnout）。很多医生和临床工作人员不得不经历所谓的“睡衣时间”（pajama time）——即在忙碌一整天下班回家、哄睡孩子之后，半夜坐在电脑前补录电子病历系统（EHR）中白天来不及记录的接诊信息。Adventist HealthCare 正在试点多种数字化记录方案，并与主要 EHR 供应商密切沟通原生集成能力。通过引入能够实时聆听医患沟通并自动生成病历摘要的生成式工具，医生只需在确认无误后签字确认，极大地释放了生产力，缓解了临床倦怠。
另一个极具潜力的场景是收入周期（Revenue Cycle）中的理赔处理。收入周期管理是医疗机构运转的生命线，但传统流程中往往需要投入数百人去执行极其繁琐的机械操作：反复登录各大保险公司（Payer）网站核对理赔信息、手工录入至 EHR、再从 EHR 导出相关证明材料上传回支付方平台。这种长链路、高耗人力的流程，非常适合引入自动化智能体（Agentic AI）来端到端自主执行，从而节约大量人力成本。
- **HR 招聘与简历筛选**：利用 AI 快速解析候选人简历，并与岗位需求进行自动化匹配，大幅提升招聘效率。
### [15:46](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=946s) 虚拟护理、计算机视觉与院内监测的全面融合
针对医院围墙内部（within the four walls of the hospital）的运作，AI 同样展现出广泛的渗透力。在护理资源有限的背景下，“环境患者监测”（Ambient Patient Monitoring）与虚拟护理（Virtual Care）发挥着关键作用。通过在病房部署传感器和摄像头，系统能及时捕捉患者的烦躁情绪或试图自行起床等跌倒高危行为（fall risk），甚至联动智能病床检测床栏是否滑落，并在危急时刻即时向护士通讯终端推送警报，指导护士立即介入干预。
此外，类似 ChatGPT 的临床助手（Clinical Assistants）正逐步应用于临床支持，辅助医生和护士快速提炼病历重点，提供更高质量的照护方案。计算机视觉（Computer Vision）也在医院运营效率提升中大放异彩：
- **急诊科吞吐量管理**：仅需在急诊大厅安装单台视觉摄像头，即可实时统计候诊人数、走廊人流密度与拥堵情况。
- **耗材室动态监控**：通过摄像头识别耗材领用与消耗状况，自动与现有库存系统对账盘点。
Joshua 在演讲尾声展示了个人 LinkedIn 二维码，呼吁同行与有创新方案的供应商积极交流，共同为治理委员会的未来决策收集前沿思路。
### [18:49](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1129s) 现场问答：自研与采购抉择，以及居家监测的未来蓝海
在问答环节，现场观众围绕非科技公司的技术应对策略及院外场景延伸提出了针对性问题：
**Q1：作为非科技公司的医疗机构，面对日新月异的 AI 技术，是自建内部实验室还是依赖供应商？**
Joshua Otten 明确回应：Adventist HealthCare 属于坚定的“采购型组织”（Buy shop / COTS shop），而非“自研型组织”（Build shop）。这是 AI 治理委员会在初期就必须明确的核心战略抉择之一。组织本身不从事大量底层开发，而是紧密依靠供应商与合作伙伴来获取成熟方案或开箱即用的产品。
**Q2：院内的环境监测模式能否进一步规模化复制到患者居家（Home Care）环境中？**
Joshua 表示高度认同。他指出 Adventist HealthCare 旗下拥有康复医院及居家护理业务（Home Health），护士需要日常上门走访，居家监测具有极其广阔的应用场景。目前他已与康复业务负责人及集团 CEO 展开过专项探讨，甚至在评估组织自主推动该领域落地的可行性。他直言，虽然面向养老院（Nursing homes）和专业护理机构（SNFs）的监测技术相对多见，但目前市面上针对“真正家庭住宅场景”的消费级监测技术仍处于蓝海阶段，存在明显的供给空白。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) 各位，我必须老实告诉你们，我根本不是什么 AI 专家，甚至差得十万八千里。 — Joshua Otten
> [00:04](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=4s) 科技革命海啸 — Joshua Otten
> [02:44](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=164s) 治理（Governance）这一点都不性感 — Joshua Otten
> [02:44](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=164s) C 开头的脏话 — Joshua Otten
> [04:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=290s) 共同学习 — Joshua Otten
> [06:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=410s) 这就像打保龄球时安装在两侧的防撞护栏（guardrails/bumpers） — Joshua Otten
> [08:47](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=527s) 接下来是真正有趣的部分，我们要把性感的创新带回来了——探讨那些亟待解决的业务用例与现实痛点！ — Joshua Otten
> [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) 重新发明轮子 — Joshua Otten
> [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) 拂去这些陈旧问题上的灰尘 — Joshua Otten
> [10:50](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=650s) another arrow in your quiver — Joshua Otten
> [13:16](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=796s) 睡衣时间 — Joshua Otten
> [15:46](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=946s) within the four walls of the hospital — Joshua Otten
> [18:49](https://www.youtube.com/watch?v=Mjtv8uc0cWI&t=1129s) Adventist HealthCare 属于坚定的“采购型组织”（Buy shop / COTS shop），而非“自研型组织”（Build shop）。 — Joshua Otten
## Mentioned
Joshua Otten, Adventist HealthCare, ChatGPT, Google, Gartner, CMS (Centers for Medicare & Medicaid Services), The Joint Commission, HIPAA, Copilot, LinkedIn
## Video description
Joshua Otten, AVP, Chief Technology Officer, Adventist HealthCare
Abstracts
The talk will focus on how to mobilize and organize your leadership around the governance of AI technologies. The attendees will be given insight into the journey of chartering a governance committee and how to address some of the following pervasive questions: What is our organization's stance on the use of AI? What policies do we need to put in place around AI? Who should have the decision-making authority? How do we keep pace of AI advancements and identify targeted business use cases?
[Source](https://www.youtube.com/watch?v=Mjtv8uc0cWI)
Source quotes
[0:04] Joshua Otten: 各位，我必须老实告诉你们，我根本不是什么 AI 专家，甚至差得十万八千里。
[0:04] Joshua Otten: 科技革命海啸
[2:44] Joshua Otten: 治理（Governance）这一点都不性感
[2:44] Joshua Otten: C 开头的脏话
[4:50] Joshua Otten: 共同学习
[6:50] Joshua Otten: 这就像打保龄球时安装在两侧的防撞护栏（guardrails/bumpers）
[8:47] Joshua Otten: 接下来是真正有趣的部分，我们要把性感的创新带回来了——探讨那些亟待解决的业务用例与现实痛点！
[10:50] Joshua Otten: 重新发明轮子
[10:50] Joshua Otten: 拂去这些陈旧问题上的灰尘
[10:50] Joshua Otten: another arrow in your quiver
[13:16] Joshua Otten: 睡衣时间
[15:46] Joshua Otten: within the four walls of the hospital
[18:49] Joshua Otten: Adventist HealthCare 属于坚定的“采购型组织”（Buy shop / COTS shop），而非“自研型组织”（Build shop）。
Source quotes
[0:04] Joshua Otten: Guys, I'm here to tell you I am not an expert in AI. I am far from an expert in AI.
[0:04] Joshua Otten: absolute tsunami of technological revolution
[6:17] Joshua Otten: bumpers in the alley for bowling
[9:17] Joshua Otten: bring sexy back
[10:50] Joshua Otten: another arrow in your quiver
[13:16] Joshua Otten: pajama time
[18:16] Joshua Otten: buy shop
[3:04] Joshua Otten: committee everything
Original transcript
[0:04] My name is Josh Uton. I serve as the
[0:07] associate vice president, chief
[0:08] technology officer for Adventist
[0:10] Healthcare. Uh we're located in
[0:12] Montgomery County, Maryland. We serve
[0:14] the Montgomery County and Prince
[0:16] George's County communities uh of of
[0:19] Maryland u just outside Washington DC.
[0:22] And we are a providerbased health care
[0:25] system. So we operate hospitals. Um, we
[0:29] have three acute care inpatient
[0:31] hospitals in Montgomery County and
[0:33] Prince George's County. We've got a
[0:35] couple of rehab inpatient hospitals,
[0:37] number of outpatient clinics, and a
[0:39] number of just outpatient services, uh,
[0:42] physician, primary care, especially
[0:44] specialy care practices uh, all
[0:46] throughout that community. And I am
[0:50] absolutely honored to be standing in
[0:52] front of you and speaking. And I
[0:54] appreciate everyone here uh coming and
[0:56] listening to my my talk and what I have
[0:58] to say and sharing my perspective uh of
[1:01] a healthcare organization that is trying
[1:03] to understand the landscape of
[1:05] artificial intelligence and how that is
[1:08] going to impact uh how we do business
[1:10] and how we care for our patients going
[1:12] forward. So as I was preparing for this
[1:16] talk, I um you know I went on ChatgPT, I
[1:20] went on Google, I I used all the tools,
[1:22] right? And said how how do you uh you
[1:25] know how do I come off as a as a
[1:29] confident and uh someone that uh you
[1:32] should be listening to and and um
[1:35] everything said that in your
[1:36] introduction you have to establish
[1:39] yourself as an expert.
[1:42] Guys, I'm here to tell you I am not an
[1:44] expert in AI. I am far from an expert in
[1:47] AI. But what I do consider myself to be
[1:50] an expert in is the evaluation, the
[1:54] selection and the implementation of
[1:56] technology in a healthcare setting. Uh
[1:58] and everything that goes along with
[2:00] that. Um and I like many of you uh and
[2:05] many of my colleagues in healthcare and
[2:07] other industries and leaders in the
[2:09] technology space. I am staring at a an
[2:14] absolute tsunami of technological
[2:17] revolution. Um, all hyperbole aside,
[2:22] everyone knows what AI means uh and and
[2:26] and how impactful uh it is. And you
[2:29] heard that from the keynote speaker uh
[2:31] around education. Uh Randy. Um, so I'm
[2:35] here to just talk to you about how do
[2:38] you in your organization
[2:41] prepare for uh handling this tsunami,
[2:44] right? Uh, and I get it. I know we're in
[2:48] Vegas. This is an AI conference. It's
[2:51] about as sexy as you can get. Governance
[2:53] is not sexy. I know that the first slide
[2:56] I put up here, I'm putting governance
[2:58] right in front of you. And it's because
[3:01] you need it. It's not something that is
[3:04] optional at this point. Um, you heard
[3:07] Randy talk about uh the fact that the
[3:10] federal government does has not given us
[3:12] the guard rails. You have to develop
[3:14] your own guard rails at this point
[3:16] unless you're in a state that's starting
[3:18] like California that's starting to
[3:19] develop out their own guard rails. Um,
[3:22] so how do you do this? Well, you build
[3:24] your team. And I've I've shied away from
[3:27] using the Cword here, committee, right?
[3:30] Because in healthcare we committee
[3:32] everything and uh so I'm going to call
[3:35] it a team but at the end of the day this
[3:36] is truly a committee and it is uh you
[3:40] need to build your team of leaders uh
[3:42] throughout from from your organization.
[3:45] Um these leaders need to represent your
[3:47] critical business functions. Um we at
[3:51] Adventist Healthcare we have decided
[3:53] that we will chair uh the IT department
[3:56] um which I am a part of will chair this
[3:58] committee and in conjunction with our
[4:01] compliance departments. Uh so our
[4:04] general counsel our compliance leaders
[4:07] we really kind of are the o overseeing
[4:10] really the uh the formation and the
[4:12] management of this uh of this committee
[4:15] and the decisions that come out of it.
[4:17] And then apart from that you want to
[4:19] bring in your leaders that can represent
[4:21] again those critical business functions
[4:23] your HR your uh your finance department
[4:27] revenue cycle uh is huge for AI and then
[4:31] ultimately your physician leaders your
[4:33] your clinical leaders so your physicians
[4:35] your nursing leaders uh and um everyone
[4:39] else in the organization that will be
[4:40] impacted uh from the uh the advent of
[4:43] AI. We have mission integration as a
[4:45] part of it. I'll talk about touch on
[4:47] that in just a moment. Um, and then next
[4:50] you need to charter that that committee.
[4:52] Um, and as part of that, I recommend
[4:54] defining artificial intelligence in this
[4:57] room right now. If I were to go around
[4:59] and pass out little slips of paper and
[5:01] tell you guys to write the definition of
[5:02] AI, it would get like a hundred
[5:03] different definitions of AI. You guys
[5:05] need to be working on the same sheet of
[5:07] music, the same page, the same
[5:09] definition of what AI is. Once you can
[5:12] start it because you're all everyone for
[5:14] the most part is starting at the same
[5:16] place. There's not a lot of true experts
[5:19] that are going to be in your
[5:20] organization. There will be some
[5:22] hobbyists and there's some people that
[5:24] are that you know if you're building out
[5:26] your own uh dev shop in AI, yes, you're
[5:29] going to have some experts there, but
[5:31] for the most part, it is this is going
[5:33] to be an exercise that's going to be
[5:34] very valuable for everyone that's a part
[5:35] of your committee. And then it's about
[5:37] learning together. We are again all
[5:40] starting at the same place. So leverage
[5:43] your existing resources whether they be
[5:45] internal your vendor relationships guys.
[5:49] That's why I'm here in Vegas with you
[5:51] guys. I want to make network. I want to
[5:53] make those connections. I want to uh I
[5:55] want to be able to take some of those
[5:57] connections back to my committee and and
[5:59] have conversations and bring back ideas.
[6:02] So I just putting a plug out there. I
[6:04] want to talk to you guys. I want to talk
[6:05] to you about uh you know how you're if
[6:08] you're in healthcare, how are you
[6:09] looking at uh at at AI, how are you
[6:12] looking at governance, what are your use
[6:14] cases, those types of things.
[6:17] And then ultimately you want to hold
[6:18] education s uh sessions. So our
[6:20] committee we meet monthly and um we've
[6:23] been doing this for the the better part
[6:24] of uh all of this year. Uh and we've
[6:28] been um we've been doing education,
[6:31] right? We've been we've brought in
[6:32] Gartner to help us talk about you know
[6:35] some some framework around developing
[6:36] out of governance. We have outside
[6:38] council that has brought uh that has
[6:40] come to us and provide us you know some
[6:42] pestal analysis around AI where is where
[6:45] is the political landscape those types
[6:47] of things and then this last bullet
[6:50] point I could spend a lot of time where
[6:52] this is really the meat of what you're
[6:54] doing in your committee right you are
[6:55] looking at your policies what are your
[6:57] existing policies you need to review
[6:59] them update them as accordingly with uh
[7:03] with language addressing artificial
[7:05] intelligence you're you need to uh make
[7:08] sure that your data handling your data
[7:10] data privacy policies if they don't
[7:12] exist they h they they must exist as
[7:15] part of this going forward uh and be
[7:17] updated and relevant to AI if you have
[7:20] BAA uh that BAAS that you work with with
[7:23] vendors
[7:25] um making sure that those are are indeed
[7:27] updated and um uh
[7:32] ultimately what you want to get to is a
[7:34] specific AI policy
[7:37] that you can publish to your
[7:38] organization so that everyone knows what
[7:41] the do's and don'ts are right now you
[7:43] probably your organization everyone is
[7:46] probably using chatbt right and they're
[7:49] in some cases it's maybe the free
[7:51] version unless if you're if you're not
[7:53] offering an enterprise version if you
[7:55] don't have like a co-pilot enterprise
[7:56] license those types of things people are
[7:58] going to go they're going to use the
[8:00] tools that are available to them before
[8:02] you give them the guardrails to work in
[8:04] so this is about building out those
[8:06] guard guardrails, the bumpers in the in
[8:08] the alley for bowling, right? Making
[8:10] sure that your your
[8:12] end users, the people that are in uh
[8:15] that are handling your data are doing it
[8:18] in a safe uh in a safe manner. Um I
[8:21] believe there's even a there's a
[8:22] conversation or a talk happening right
[8:24] now um at the same time that's competing
[8:27] with this uh from a a member of the
[8:29] House of Representatives giving his uh
[8:32] take with the United States House of
[8:33] Representatives giving his take on what
[8:35] the political landscape is going forward
[8:37] for AI. So I appreciate you guys coming
[8:39] visiting me today. So again, I could
[8:41] spend a lot of time here, but this is
[8:43] really as far as when you're talking
[8:45] about your teams. Um, you also want to
[8:47] talk about, you know, ethically,
[8:50] morally, uh, regul from a regul
[8:53] regulatory perspective from HIPPA laws
[8:57] down to joint commission um of, uh, CMS
[9:01] that come and and do site visits and
[9:03] site surveys of your of your hospitals
[9:05] and your environments. They are all
[9:07] going to have their own individual
[9:09] governance and and and requirements
[9:12] around AI. One of those being
[9:17] one more time. There we go. Um in
[9:19] inventory. So this might seem simple. It
[9:22] might seem trivial, but it's actually
[9:24] becoming very uh important to what you
[9:26] need to do. And this is uh I just took a
[9:29] snippet of what we're doing as far as
[9:31] maintaining an inventory. Um, chances
[9:33] are again your your organization is
[9:36] probably using artificial intelligence
[9:38] in some capacity. Uh, so going and
[9:41] looking through your catalog or your
[9:44] portfolio of of applications and trying
[9:47] to pull out what is um what falls into
[9:51] the domain of AI and and how uh and then
[9:54] how are you using it, how is it being
[9:57] applied. This is very I have this my
[9:59] snippet is very high level. You can
[10:01] build it out any way that you want, but
[10:03] you're going to need to have this. I've
[10:05] heard um we've had joint commission
[10:07] surveys recently where they're starting
[10:09] to ask for this. Um so
[10:13] start on it now. Start building this out
[10:15] whether or not you um you have a a a
[10:18] governance framework or not. Um this is
[10:20] going to be valuable as your compliance
[10:23] departments come to you and say, "Hey,
[10:25] Joint Commission's here." and they're
[10:27] asking for all of our AI applications
[10:29] and how they're being how it affects our
[10:31] patient care and and everything else.
[10:33] You don't want to get that citation.
[10:34] This is also going to be uh utilized in
[10:37] in in future regulatory capacities as
[10:40] well.
[10:43] All right. And now the fun part. We're
[10:45] bringing sexy back. Use cases and
[10:48] problems that need to be solved.
[10:50] This is where we can really go into the
[10:52] technology and try to understand how it
[10:55] can serve your business. Uh you don't
[10:59] need to reinvent the wheel. What are
[11:01] your your organization's specific
[11:04] strategic goals? Uh my organization has
[11:08] a strategy map that we we update. Uh it
[11:11] goes through 2030 and it's divided out
[11:14] by all of our entities, all of our
[11:16] hospitals and service lines. Um, and
[11:18] it's a nice nice binder laminated. It
[11:22] sits on my off sits right on my office
[11:24] desk. I can go to it and I can look in
[11:26] there and I can pull out specific
[11:28] strategic goals and then I can have a an
[11:32] exercise to decompose those goals into
[11:35] actionable initiatives that can be that
[11:38] uh AI can be used uh to solve for uh
[11:43] that that goal uh to reach that goal. So
[11:46] what you're doing is you're what you're
[11:48] you're doing is you're taking that
[11:49] governance framework and everything that
[11:51] you've put together previously and
[11:53] you're using that as a lens to look look
[11:56] at one the strategic goals but also your
[12:00] persistent problems. Every organization
[12:03] has problems that have been haunting
[12:06] their organization forever. uh and you
[12:09] have maybe five, 10, 15 workarounds in
[12:13] place that have developed over the last
[12:15] 20 years. You've thrown 50 FTEEs at it
[12:18] and it's something that has just been a
[12:20] burdensome.
[12:22] Dust that off. Dust those problems off
[12:24] and take a fresh look at them through
[12:26] the lens of AI and and you know things
[12:29] around you know agentic AI. This goes
[12:31] back to your learning sessions, knowing
[12:34] more and more about the technology and
[12:36] where it's going and how you can
[12:38] leverage it. When you can start to
[12:40] develop that lens, you can start to look
[12:43] at your organization and start to see
[12:46] where there's going to be an
[12:48] opportunity. Um,
[12:51] AI at the end of the day, it's a tool,
[12:54] right? AI solutions are tools. They're
[12:55] another arrow in your quiver. So what
[12:57] you what you need to do is just
[12:59] continually it's the the continuous
[13:02] improvement the continuous review of
[13:05] your organization uh again through that
[13:08] lens. So what in healthcare are some of
[13:11] your your uh use cases that we're
[13:14] seeing?
[13:16] Ambient digital scribe it's probably
[13:18] number one right now. We're hearing
[13:20] about it. Um we uh we're we're we're
[13:23] piloting a number a couple of different
[13:25] solutions. Uh we're having a lot of um
[13:28] discussions with our EHR vendors that
[13:30] are you know looking to integrate and
[13:32] make uh native capabilities for ambient
[13:35] digital scribes. Uh here at we're coming
[13:39] out of the the pandemic. We are five
[13:41] years removed from 2020 and only really
[13:44] two or three years removed from kind of
[13:47] the end of the pandemic. Right? Our
[13:50] health care workers are burnt out. Our
[13:53] physicians are burnt out. Our clinicians
[13:55] are burnt out. They are spending time.
[13:58] They call it pajama time where they go
[14:00] home at the end of their day and they
[14:03] sit down and at the end of the day after
[14:05] they're putting their kids to bed,
[14:06] they're going and they're charting in
[14:08] the EHR for the encounters that they
[14:10] didn't have time to chart on earlier in
[14:12] the day. giving them a tool that can
[14:15] that can that can listen in on the
[14:16] encounter and and generatively develop
[14:19] out a note that they can just sign off
[14:22] on is a is a major major productivity
[14:25] and really relief of of burnout. um
[14:28] revenue cycle claims processing
[14:31] um that is another huge huge opportunity
[14:34] for AI. Rev cycle is one of those
[14:37] things. I mean, it's the lifeblood of uh
[14:39] an organization, a healthcare
[14:41] organization's revenue. Um and uh it is
[14:45] often, you know, you're throwing
[14:48] hundreds and hundreds of people at uh
[14:51] just menial tasks of going on to payer
[14:55] websites, checking for claims, going
[14:58] back and rewriting that to the EHR,
[15:02] you know, going and finding the
[15:04] documentation from the EHR to then
[15:06] upload back into the website of the of
[15:10] the payer. It is a long arduous process
[15:13] and it takes people it takes bodies to
[15:16] do it. This is another opportunity for
[15:20] that level of automation those agentic
[15:22] AIs that can go in and do this for you.
[15:27] Recruiting I think there's a lot of
[15:29] people that uh you know similarly right
[15:32] I I think this really kind of goes
[15:34] without saying uh AI is is great for uh
[15:37] HR recruiting. It's great at SC at
[15:39] looking at resumeumés uh and being able
[15:42] to um apply that against the
[15:43] requirements of of the job positions. Uh
[15:46] patient monitoring this I could do a
[15:49] 20-minute talk on just virtual care and
[15:52] where things are going uh in the four
[15:55] walls of the hospital. um ambient
[15:58] patient monitoring um putting putting
[16:01] sensors and cameras in patient rooms um
[16:04] to actually help uh help the clinician
[16:09] keep an eye on that patient. We only
[16:11] have so many nurses, right? Virtual care
[16:13] being able to um you know have an
[16:15] ambient sensor that can alert a nursing
[16:19] a centralized nurse that maybe a patient
[16:22] is getting agitated or they're starting
[16:24] to get up and maybe they're at a fall
[16:26] risk. looks like they're getting up.
[16:28] Maybe that's integrated with your beds
[16:29] in some way to tell you that a rail is
[16:31] down. There's a risk there. And then you
[16:33] can you have the ability to uh alert the
[16:37] nurse through their nursing
[16:39] communications that hey, you have you
[16:41] have to get to this room and and
[16:42] intervene immediately or you have a fall
[16:44] on your hands. Um other, you know,
[16:47] there's a ton of technology and these
[16:50] sensors um that, you know, don't
[16:52] necessarily go into the world of of AI.
[16:55] uh but everything is kind of uh
[16:57] consolidating back to that platform of
[16:59] AI uh in in being able to help monitor
[17:02] patients and then um clinical assistance
[17:06] similar to your your chat GPTs of the
[17:08] world. You're starting to see that in um
[17:11] in in the clinical space and giving you
[17:14] know physicians and nurses uh a a a
[17:17] virtual assistant to help you know
[17:20] review the uh the the chart uh and pull
[17:24] out some themes um to allow them to give
[17:27] better care of their patients. Similarly
[17:29] also I mean I I see a lot of
[17:31] opportunities in the space of um
[17:33] computer vision in hospitals having more
[17:36] cameras in hospitals that can allow for
[17:39] thing uh to to be able to allow uh
[17:41] review things like throughput um you
[17:44] know how many people do we have in the
[17:45] waiting room of the ED um based off of a
[17:48] computer vision uh camera just one
[17:50] camera looking in that waiting room um
[17:52] you know the traffic how how how busy is
[17:55] the how many people do you have in the
[17:57] hallways right that all these types of
[17:59] things. What are uh putting a camera in
[18:01] the supply room to see how many supplies
[18:03] are being and then denigrating that
[18:04] against your existing uh stock. There is
[18:07] a lot a lot of opportunities and use
[18:10] cases for AI in uh within the four walls
[18:14] of the hospital.
[18:16] So with that I have I think only about
[18:19] two minutes or so left. I do want to
[18:22] thank you guys for coming today. Uh I
[18:23] put up my QR code. Please connect with
[18:25] me me with uh with me in LinkedIn. Um I
[18:29] I want to talk to you guys about
[18:30] healthcare use cases. So if you see me,
[18:33] please come up to me. I want to hear
[18:34] what you guys are doing. I want to hear
[18:36] uh if you are, you know, a vendor and
[18:38] you have a creative solution, I want to
[18:39] talk to you. I want to be able to bring
[18:42] this back to my committee so that we can
[18:44] make informed decisions going forward. I
[18:46] think maybe I have one minute for
[18:48] questions.
[18:49] >> You do indeed. Just first a round of
[18:51] applause for Mr. Joshua. Thank you so
[18:54] much. Thank you very much. Thank you.
[18:55] >> He does have time for maybe one to two
[18:58] very quick questions as our sessions are
[19:00] closely scheduled, one behind the other.
[19:02] If you have a question, do raise your
[19:04] hand and step out into this lane here
[19:06] and I will assist you with the mic. Yes,
[19:09] sir.
[19:12] >> Oh, thank you very much, Josh. Oh,
[19:13] sorry. I just short question because you
[19:16] are not a IT company. Yeah. You're
[19:17] healthcare organizations, right?
[19:19] >> And at the moment AI is changing
[19:21] rapidly. I mean like every day you have
[19:23] to follow it. Yeah. you have to build
[19:25] kind of huge AI team to be able to
[19:26] follow it. So my question to you is how
[19:29] you manage it. You're building your own
[19:30] AI internal lab to manage it or you rely
[19:33] on vendors or
[19:34] >> No, we're relying on vendors. We are not
[19:36] a we're not a build organization. That's
[19:38] that's one part they really get a chance
[19:40] your committee. One of the things you're
[19:41] going to you're you're going to
[19:43] decisions you're going to make are you a
[19:44] build or a buy shop, right? Are you
[19:46] going to be le leveraging your vendors
[19:48] to help you build solutions or or have
[19:50] off-the-shelf uh solutions for you or
[19:52] going to build your own? Us personally,
[19:54] we are b we rely we're K's uh we're K's
[19:57] shop for the most part. We don't do a
[19:59] lot of our own development. So um we are
[20:02] very much uh leaning on our vendors
[20:04] >> partners.
[20:05] >> Thank you. And our last question.
[20:08] >> Hi uh great presentation. Thank you. My
[20:10] name is Swini from Aerospike. My
[20:12] question is very simple. So you are
[20:15] monitoring you know you talked about uh
[20:17] making things uh which may not be
[20:19] working that great now better
[20:21] >> but what about the opportunities like
[20:23] the kind of monitoring you're doing
[20:24] within the hospital you could monitor
[20:26] people at homes I mean you could scale
[20:28] this really well make it uh much more
[20:30] effective do you have any thoughts on
[20:32] that
[20:32] >> yeah yes absolutely and that's something
[20:34] uh you know those are conversations that
[20:36] we've had uh our organization has a a
[20:39] rehab hospital that we mentioned a rehab
[20:41] service line as part of that we have
[20:42] home home
[20:44] as well. So, we have nurses going into
[20:46] homes. That is absolutely there's an
[20:48] absolute use case for that. And there
[20:50] that is something that I've actually
[20:51] been talking to one of the presidents of
[20:53] of uh the rehab uh organization and our
[20:56] CEO about if there is there
[20:58] opportunities for developing out that
[21:00] our ourselves because I see a really a
[21:02] lack of of that type of technology that
[21:05] is consumerbased. You see it in nursing
[21:08] homes, skilled nursing facilities,
[21:09] things like that. but actually taking it
[21:11] to the home and making it residential.
[21:13] Not a lot of opportunities today.
