University of Michigan Deploys Custom Campus AI Across 140,000 Users

Building AI-Ready Organizations: The Strategic Role of Chief AI Officer With Michigan University — Ai4

Ai4

Higher education institutions must mandate AI literacy for all graduates while providing custom, free-to-use tools that lower administrative burdens and provide every student with personalized tutoring.

AI-assisted summary of Building AI-Ready Organizations: The Strategic Role of Chief AI Officer With Michigan University — original source. Check the original for full context.

Source published: 2025-08-24

Detailed notes

## Background Published by Ai4 on YouTube on August 24, 2025, this presentation features Dr. Ravi Pendse, Vice President for Information Technology and Chief Information Officer at the University of Michigan. According to the page description, the session explores how organizations can prepare for AI transformation, focusing on the strategic role of Chief AI Officers in fostering innovation, aligning technology with organizational goals, and establishing robust governance frameworks.

## TL;DR Ravi Pendse shares insights from the University of Michigan's pioneering August 2023 campus-wide generative AI rollout across 140,000 community members, which established custom platforms like Maizey and LMS-integrated 24/7 tutors well before vendor solutions existed. He outlines the strategic imperatives for a Chief AI Officer, stressing that technical innovation must be paired with calculated risk-taking, rigorous data privacy, universal accessibility, and genuine empathy for workplace anxieties. Pendse argues that foundational AI literacy must become mandatory for both graduating students and the modern workforce. Finally, he underscores the need for interdisciplinary governance that actively includes ethicists, humanists, and skeptics to guide responsible transformation through a culture of 'constructive irreverence.'

## Atmosphere & style Pendse delivers an engaging, down-to-earth keynote that opens with humorous self-deprecation and Sanskrit philosophy before transitioning into high-stakes enterprise strategy. The presentation carries an optimistic, visionary momentum grounded in pragmatic risk mitigation, institutional transparency, and deep empathy for end-user hesitation. Watching the talk feels intellectually stimulating and culturally authentic, building from thoughtful institutional lessons to an inspiring showcase of campus impact and an enthusiastic closing audience rally.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) Rapid technological evolution requires organizational leaders to approach AI with a humble student mindset rather than claiming absolute expertise. - [02:22](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=142s) Responsible AI innovation thrives on 'constructive irreverence,' where stakeholders actively debate and challenge leadership on model bias, hallucinations, and sustainability. - [04:15](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=255s) The University of Michigan rolled out enterprise generative AI to 140,000 campus users in August 2023, well before commercial vendor enterprise offerings were standardized. - [04:50](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=290s) The mandate of a Chief AI Officer extends beyond technical architecture to addressing the unspoken human fears, such as job displacement, that drive institutional resistance. - [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) Deploying custom enterprise AI requires calculated risk-taking and explicit executive backing, particularly when navigating unpredictable tokenization and infrastructure costs. - [09:20](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=560s) Institutional AI deployments must be anchored by three non-negotiable principles: private and secure data architecture, universal accessibility for screen readers, and 100% free, equitable campus access. - [10:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=652s) Treating AI as the user interface and embedding it directly into existing platforms like Canvas LMS allows instructors to configure customized 24/7 course tutors in under four minutes. - [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) AI literacy is today's equivalent of 1990s internet literacy; no student should graduate and no worker should remain in the modern workforce without foundational AI proficiency. - [13:45](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=825s) AI governance cannot be left solely to technical teams; it requires multidisciplinary panels that deliberately include humanists, philosophers, ethicists, and skeptical refusers. - [15:06](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=906s) Practical generative AI applications span the entire enterprise, including virtual clinical patients for medical training, curriculum digital twins, and logistics assistants that reduce administrative email. - [17:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=1072s) Campus-wide AI scales to daily student life through personalized assistants that coordinate class schedules, live transit routing, and dining hall wait times.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) Introduction: The "Doctor Who Cannot Help Anybody" and the Student Mindset

Ravi Pendse opens his keynote by setting aside his formal title of Dr. Pendse in favor of simply "Ravi," sharing an anecdote from when his son was seven years old. While checking the mail outside his home, Pendse heard a loud crash and turned to see a neighborhood boy named Joey fall off his bicycle. As other children started screaming that Joey was gushing blood, Pendse rushed toward them in a panic, only to find Joey had a minor scratch. Before he could assist, his ever-helpful seven-year-old son rode up on his bike and announced to his friends: "Don't ask my dad for help. He's the kind of doctor that cannot help anybody." Pendse jokes that AI could never replicate a kid's innate ability to ground their parents. In full disclosure, that same son recently completed his PhD and began a faculty appointment at the University of California, San Francisco, prompting Pendse to ask him if he too was about to become "the doctor that cannot help anybody."

Moving past the joke, Pendse emphasizes that he does not claim to be an AI expert; rather, he considers himself an ongoing "student of AI." Given the breakneck speed at which technology evolves, he believes everyone must adopt a learner's mindset. He grounds this philosophy in a Sanskrit teaching passed down by his mother when he was seven: *Acharya Devo Bhava*, which teaches that a teacher is made in the image of God and should be revered accordingly. Because his mother taught him to see a teacher in everyone he interacts with, Pendse tells the audience that he treats every person in the room as his instructor.

- **Key phrase**: *Acharya Devo Bhava* ("The teacher is God") — viewing every peer and AI practitioner as an instructor. - **Perspective**: Approaching rapid technological disruption with humility and curiosity rather than claiming absolute expertise.

---

[02:22](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=142s) The University of Michigan Scale and "Constructive Irreverence"

Pendse outlines the sheer operational scale and intellectual culture of the University of Michigan. As a massive public institution, Michigan serves over 63,000 students and encompasses an Ann Arbor community of nearly 130,000 people across multiple campuses, backed by a $16 billion operating budget, a $21 billion endowment, and an annual research expenditure of $2.04 billion. Its students and faculty operate at the top of their fields, demanding cutting-edge technology alongside uncompromising transparency, ethical governance, and accountability.

```mermaid flowchart TD accTitle: University of Michigan Institutional Scale and Core Culture subgraph Scale["Institutional Footprint"] A["63,000+ Students & 130k Community"] B["$16B Operating Budget / $21B Endowment"] C["$2.04B Annual Research Enterprise"] end subgraph Culture["Campus Culture"] D["'Constructive Irreverence'"] E["Demands for Transparency & Clarity"] F["Active Debate: Bias, Hallucination & Sustainability"] end Scale --> Culture ```

To capture the campus atmosphere, Pendse coins the term "constructive irreverence." Nothing is accepted at face value; students, researchers, and clinicians constantly challenge leadership on AI sustainability, model hallucinations, algorithmic bias, and ethical frameworks. Rather than resisting this friction, leadership actively embraces the debates, disagreements, and rigorous questioning as the essential mechanism that drives the university to innovate responsibly.

- **Total student body**: 63,000+ students. - **Community reach**: ~130,000 faculty, staff, and students across Ann Arbor and regional campuses (Dearborn and Flint). - **Financial metrics**: $16 billion annual operating budget, $21 billion endowment, $2.04 billion yearly research portfolio. - **Guiding institutional ethos**: "Constructive irreverence" — continuous questioning and debate surrounding AI governance and societal impact.

---

[04:15](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=255s) Deploying Generative AI at Scale and the Mandate for a CAIO

Long before enterprise generative AI platforms became commercial standards, the University of Michigan rolled out institutional generative AI services at scale on August 21, 2023, serving a community of 140,000 individuals. Pendse describes generative AI as the most transformative technology of this century—comparable in scale to the arrival of the internet—and a potent force for positive disruption. However, realizing this potential comes with monumental caveats: the technology must be used thoughtfully, responsibly, legally, and ethically.

Navigating these massive operational "ifs" requires dedicated leadership, whether through a Chief AI Officer (CAIO) or an empowered CIO. The mandate spans admissions, student enrollment, curriculum design, clinical healthcare workflows, and enterprise administration. A leader in this space must possess visionary ideas, genuine collaborative instincts, and the emotional resilience to listen calmly during intense pushback. Pendse notes that public discourse around AI is heavily clouded by fear—ranging from dystopian *Terminator* scenarios to immediate anxieties over job losses. A successful AI executive must look beneath the surface of what people say to address the underlying fears and unstated concerns driving organizational resistance.

- **Historic launch date**: August 21, 2023 — rollout of campus-wide generative AI across 140,000 users before commercial enterprise models were widely available. - **Transformational scope**: Student enrollment, admissions, pedagogical support, research, and clinical health systems. - **Core CAIO competencies**: Empathy, strategic foresight, patience under criticism, and the ability to decode the unspoken fears fueling resistance.

---

[07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) Pioneer AI Tutors, RAG, and Navigating Calculated Risk

Engaging the audience, Pendse asks attendees to name their hardest college course, receiving shouts of "Bayesian math" and "cost accounting." Pendse reveals his own academic nemesis was thermodynamics—a class so grueling that, fifty years later, he still has recurring nightmares of failing and having his degree rescinded. He connects this to a high-profile statement made by Nvidia CEO Jensen Huang in early 2024, in which Huang argued that every student in the world should have a dedicated AI tutor. Pendse pointedly notes that Michigan students were already provided 24/7 AI tutors and Retrieval-Augmented Generation (RAG) tools back in August 2023. As a public institution, Michigan makes its foundational work, architectures, and lessons freely accessible to assist other organizations without commercial motives.

```mermaid flowchart LR accTitle: Michigan Generative AI Implementation Pillars subgraph Foundation["Custom Private Infrastructure"] A["On-Premise / Private Cloud Instance"] B["RAG Architecture (Live Aug 2023)"] end subgraph CorePillars["Strategic Guiding Principles"] C["Strict Data Privacy & Security"] D["Universal Accessibility (Screen Readers)"] E["Equity (100% Free to Campus Community)"] end Foundation --> CorePillars ```

Building their own private instance prior to vendor enterprise offerings required taking calculated operational and financial risks. On the eve of the August 2023 launch, Pendse met with the university president, telling him: "I don't know how it's going to go tomorrow. I think it's going to be okay, but I need you to have my back." The president agreed without hesitation. The primary anxiety centered around unpredictable tokenization costs—whether 5,000 or 100,000 users would hammer the infrastructure on day one. Pendse stresses that effective AI leadership demands both calculated risk-taking and thorough risk mitigation, founded on three non-negotiable principles:

- **Data Privacy & Security**: Building a secure, isolated private environment before commercial vendor enterprise tiers existed. - **Accessibility**: Designing tools that fully integrate with screen readers so visually impaired and blind community members are never left behind, an area where many commercial tools still fall short. - **Equitable Access**: Delivering every custom AI tool, tutor, and capability 100% free of charge to all students, faculty, and staff.

[10:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=652s) Democratizing AI on Campus: Maizey and LMS Integration

University of Michigan built its own proprietary, codeless platform named Maizey to put artificial intelligence directly into the hands of its community. Operating under the conviction that "AI is the new UI," the university's strategy focused on meeting users where they already work rather than expecting them to navigate standalone, siloed tools. A prime example of this seamless deployment is that any faculty member at Michigan can configure and launch a customized, 24/7 AI course tutor in less than four minutes.

This rapid deployment was made possible because Michigan integrated generative AI tools directly with Canvas back in 2023—a full two years before Instructure, the parent company of Canvas, announced its own native AI integrations. Alongside Maizey, the university rolled out the U-M GPT toolkit to deliver API-level access across the institution. Rolling out these capabilities at enterprise scale required deep campus-wide collaboration, building shared partnerships, inspiring hesitant stakeholders, and striking a careful balance between aggressive innovation and thoughtful risk mitigation. Pendse emphasizes that executing an institutional rollout of this magnitude requires dedicated leadership—whether through an appointed Chief AI Officer or a coordinated leadership group.

```mermaid flowchart TD accTitle: University of Michigan AI Ecosystem subgraph CorePlatform["Core AI Platform & APIs"] A["Maizey (Codeless Platform)"] B["U-M GPT Toolkit (APIs)"] end

subgraph InstitutionalIntegrations["Campus Workflows & Integrations"] C["Canvas LMS (Integrated 2023)"] D["Google Drive & Department Hubs"] end

subgraph UserDeployments["End-User Applications"] E["24/7 Course Tutors (< 4 min setup)"] F["Personal Student Assistants"] G["Clinical & Research Workflows"] end

CorePlatform --> InstitutionalIntegrations InstitutionalIntegrations --> UserDeployments ```

- Maizey is provided completely free of charge to the university community as in-house intellectual property. - Faculty can build and deploy interactive, course-aligned AI tutors in under four minutes. - U-M completed full generative AI integration into the Canvas learning management system in 2023. - The U-M GPT toolkit provides flexible API access across schools and colleges.

---

[12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) AI Literacy Imperative, Multidisciplinary Governance, and Trust

AI literacy has become a foundational competency for the modern era. Pendse reiterates his public stance that no student should graduate from any institution of higher education without completing at least one core course in AI or gaining substantial fluency with modern AI tools. He expands this mandate directly to the workforce: no employee in any modern organization should accept not knowing how to use AI. While organizational leaders bear the fundamental responsibility to provide training, infrastructure, and access, employees must take individual responsibility to actively learn and adopt these tools. In Pendse's view, AI literacy today occupies the exact same critical status that basic internet literacy held thirty years ago.

True institutional adoption also demands a governance model that actively embraces dissent and diverse perspectives. Acknowledging that some faculty and staff remain skeptical and vow never to touch AI, Pendse stresses that these voices must remain at the decision-making table so leadership can continue educating and collaborating with them. Speaking as an engineer, he cautions against homogenous decision-making: technical choices cannot be left to five engineers alone, but must actively include philosophers, English professors, and ethicists. Joking about using the word "diversity" in a public forum ("Uh-oh, I did use the diversity word. So, I might be in trouble now... somebody may be waiting outside"), he underlines that leading AI transformation requires balancing governance, strategy, innovation, and leadership. Above all, a Chief AI Officer cannot simply demand trust from their team—it must be earned.

- Pendse argues that graduating from higher education without AI coursework or foundational AI competency should no longer be permitted. - Workplace AI literacy is viewed as a dual obligation: leaders must provide the tooling and education, and employees must step up to learn. - AI governance requires interdisciplinary panels, deliberately combining engineering, ethics, the humanities, and philosophy. - Skeptics and AI refusers are intentionally kept at the advisory table to ensure broad institutional alignment.

---

[15:06](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=906s) Voices of Michigan: Campus-Wide AI in Action

To demonstrate the real-world impact of these systems, Pendse shares a campus video showcasing students, faculty, clinicians, and administrative staff explaining how AI transforms their everyday routines:

A faculty member describes building a digital twin trained directly on their curriculum, providing students with a 24/7 tutor capable of answering questions, working through practice assignments, and dramatically increasing classroom interactivity. Administrative staff highlight the use of AI to streamline and inform institutional procurement workflows. Students share how they use Maizey to synthesize lecture notes when studying for midterms and finals, while engineering teams embed Maizey into their student resource hubs to enable on-demand technical learning.

In healthcare, researchers and clinicians use AI to process complex health data into actionable clinical insights, expanding equitable and effective care in resource-constrained environments. In health sciences education, professors deploy AI to simulate virtual patient interactions for diagnostic practice. Another instructor demonstrates an automated assistant connected to their Google Drive that filters routine logistics—such as answering whether office hours are taking place or processing extension requests—substantially reducing daily email volume. The video concludes with a rapid montage of community members declaring "I use AI," culminating with the spirited sign-off: "And I am AI. Go Blue."

- Faculty deploy course-specific digital twins to practice assignments and answer student questions around the clock. - Procurement teams utilize AI tools to improve vendor evaluation and administrative efficiency. - Medical and public health teams transform complex clinical datasets to support equitable care in under-resourced regions. - Medical educators create dynamic virtual patients for student clinical training. - Automated assistants connected to course drives resolve routine student inquiries regarding office hours and deadlines, cutting down administrative email.

---

[17:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=1072s) Personal AI Assistants for Every Student and Closing Rally

Since February, every enrolled student at the University of Michigan has had access to a dedicated personal AI assistant. This platform coordinates directly with each student's daily academic life: it tracks their course schedules, maps out precisely which campus bus to catch to arrive on time, manages restaurant queues, and provides live wait-time updates across dining locations. Pendse concludes his talk by expressing gratitude for the opportunity to share Michigan’s AI journey, inviting the entire room to join him in a spirited chant of "Go Blue," which is met with energetic audience participation and warm applause.

- Every Michigan student is provisioned with a personalized AI assistant integrated with their class schedule and campus transit. - The assistant delivers live utility functions, including bus transit routing and campus dining queue tracking. - Pendse closes the keynote with a shared audience rally of the university's signature slogan, "Go Blue!"

## Notable quotes > [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) Don't ask my dad for help. He's the kind of doctor that cannot help anybody. — Ravi Pendse

> [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) Acharya Devo Bhava — Ravi Pendse

> [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) I don't know how it's going to go tomorrow. I think it's going to be okay, but I need you to have my back. — Ravi Pendse

> [10:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=652s) AI is the new UI — Ravi Pendse

> [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) Uh-oh, I did use the diversity word. So, I might be in trouble now... somebody may be waiting outside — Ravi Pendse

> [15:06](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=906s) I use AI — Video Montage

> [15:06](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=906s) And I am AI. Go Blue. — Video Montage

> [17:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=1072s) Go Blue — Ravi Pendse

## Mentioned Ai4, Ravi Pendse, University of Michigan, University of California, San Francisco, Jensen Huang, Nvidia, Maizey, Canvas, Instructure, U-M GPT, Google Drive

## Video description Ravi Pendse, Vice President for Information Technology and Chief Information Officer, University of Michigan

Abstracts

In this session, we will explore how organizations can effectively prepare for AI-driven transformation. Discussions will focus on the expanding role of Chief AI Officers in aligning emerging technologies with business goals, fostering innovation, and establishing robust governance and ethical frameworks for responsible AI deployment.

[Source](https://www.youtube.com/watch?v=8BtaxwaAnmE)

中文简报

密歇根大学面向14万名用户部署定制校园AI 高等教育机构必须对所有毕业生强制推行AI素养要求，同时提供免费定制工具以减轻行政负担，并为每位学生提供个性化辅导。 Ravi Pendse分享了密歇根大学于2023年8月在全校14万名师生员工中率先推广生成式AI的经验见解。早在供应商方案成熟之前，该校便建立了Maizey等定制平台以及与LMS集成的24/7全天候导师系统。他概述了首席AI官的战略要务，强调技术创新必须与审慎的风险承担、严格的数据隐私、通用的无障碍访问以及对职场焦虑的真切同理心相结合。Pendse主张，基础AI素养必须成为毕业生和现代劳动力的强制要求。最后，他强调了跨学科治理的必要性，主张积极吸纳伦理学家、人文学者和怀疑论者，以“建设性批判”文化引导负责任的变革。密歇根大学早在2023年8月商业版本普及之前，就面向全校14万名师生全面推出了企业级生成式AI服务。 (4:15) 学校设立了机构AI落地的三大不可妥协原则：保障数据隐私与安全、全面支持读屏等无障碍功能，以及向全体师生提供100%免费平等的访问权限。 (7:44) 基于“AI即新UI”的理念，密歇根大学开发了自研无代码平台Maizey，并将其深度嵌入Canvas等教学系统，让教师在4分钟内即可配置出定制化的24/7全天候课程AI助教。 (10:52) AI素养被视为当今时代的“互联网素养”，高等教育毕业生与现代职场员工都必须掌握AI基础能力，同时AI治理需由工程、哲学、伦理及反对者跨学科共同参与。 (12:40) 全校学生均已配备个人AI专属助手，能联动课程表、规划校园巴士接驳，并实时追踪食堂排队动态，全方位融入学术与校园生活。 (17:52)

Reading summary

University of Michigan Chief Information Officer Ravi Pendse argues that enterprise AI adoption requires bold operational risk-taking paired with equitable, responsible governance. Rather than waiting for vendor products, institutional leaders must build accessible, privacy-first infrastructure while embedding broad cross-disciplinary oversight. Early Enterprise Rollout and Core Architectural Pillars (4:15) On August 21, 2023, the University of Michigan launched an institutional generative AI environment across its 140,000-person community, deploying Retrieval-Augmented Generation (RAG) and course tutors ahead of standardized commercial enterprise offerings. Taking on unpredictable initial tokenization costs and technical risks required explicit support from executive leadership. Pendse emphasizes that this rollout rested on three strict, non-negotiable commitments: isolated data privacy and security, complete screen-reader accessibility for visually impaired users, and zero-cost equitable access for all enrolled students, faculty, and staff. Democratized Workflows and Campus-Wide Deployments (10:52) To integrate AI into existing daily workflows rather than standalone silos, Michigan developed Maizey, an in-house codeless platform, alongside the U-M GPT toolkit. By directly integrating generative AI into the Canvas learning management system in 2023—two years ahead of native vendor integrations—the university enabled instructors to build course-specific AI tutors and digital twins in under four minutes. Operational deployments span multiple disciplines: automated assistants link to course drives to answer routine logistical queries, medical educators run virtual patient diagnostic simulations, and health researchers convert complex clinical data for under-resourced settings. Additionally, enrolled students receive a personal assistant that syncs academic schedules with campus transit and real-time dining queue data. AI Literacy, Institutional Culture, and Governance (2:22) Michigan’s operational scale—encompassing 63,000 students, a $16 billion operating budget, and $2.04 billion in annual research expenditures—operates under an ethos Pendse terms 'constructive irreverence.' Rather than suppressing campus friction regarding algorithmic bias, sustainability, and model hallucinations, leadership treats continuous questioning as a safeguard for responsible deployment. Pendse advocates for mandatory AI literacy, asserting that higher education should not graduate students who lack core AI competencies, nor should corporate employees opt out of training. To manage the human anxiety and skepticism surrounding technology, he argues that AI governance cannot be left exclusively to engineers; it must deliberately seat ethicists, humanities faculty, and vocal AI skeptics at the decision-making table. Pendse maintains that technological transformation succeeds only when technical execution is matched by human trust and institutional empathy. While Michigan established an early operational template, the enduring challenge for Chief AI Officers is balancing rapid technical deployment with uncompromised ethics, multidisciplinary oversight, and universal access.

精炼阅读摘要

密歇根大学首席信息官兼技术副校长 Ravi Pendse 指出，生成式人工智能是本世纪极具颠覆性的战略力量，但其在大型组织中的成功落地并非单纯依赖算法，而取决于具有前瞻性的风险承担、严谨的跨学科治理以及普惠平等的底线原则。Pendse 强调，首席人工智能官（CAIO）必须超越技术部署本身，在全机构范围内普及 AI 素养，并以包容质疑的文化化解组织内部的深层焦虑。机构体量与“建设性质疑”的治理文化 (2:22) 密歇根大学具备庞大的运行体量，涵盖超过 63,000 名学生、分布于安娜堡及分校的近 130,000 名社区成员，并由 160 亿美元的年运营预算、210 亿美元的捐赠基金及 20.4 亿美元的年度科研经费提供支撑。Pendse 指出，顶尖公立学府的师生与临床科研人员在追求前沿技术的同时，对透明度、伦理合规与机构问责有着极高标准，绝不会盲目接受任何技术方案，而是围绕算法偏见、模型幻觉、环境可持续性及伦理框架持续向领导层发问。 Pendse 将这种校园文化定义为“建设性质疑”（constructive irreverence）。校方管理层并未视这种学术摩擦为阻力，反而将其确立为驱动负责任创新的核心机制。Pendse 认为，面对技术的迅猛演进，管理者必须放下技术权威的傲慢，保持谦逊的学习者心态（遵循印度梵文格言 Acharya Devo Bhava，将每位共事者与提问者视作导师），以同理心倾听并吸收尖锐的反对声音，从而确保技术演进始终贴合公共机构的使命。先行部署的技术架构与三大核心原则 (7:44) 在商业化企业级生成式 AI 方案普及之前，密歇根大学已于 2023 年 8 月 21 日面向全校约 140,000 名用户推出了全校范围的生成式 AI 服务，并提供了全天候 AI 导师与检索增强生成（RAG）工具。Pendse 特别指出，英伟达首席执行官黄仁勋在 2024 年初公开倡导全球学生应配备专属 AI 导师，而密歇根大学早在数月前就已实现这一构想。作为公立机构，学校选择将其技术架构与实践经验免费向社会公开，不带任何商业盈利目的。在供应商企业级服务尚未成熟时自建隔离环境，必然伴随着巨大的运营与财务风险，其中最主要的顾虑在于不可预测的 Token 消耗成本。为此，管理层在推进部署时确立了三条不可妥协的核心原则：其一是严格的数据隐私与安全防护；其二是通用无障碍设计，确保系统原生兼容屏幕阅读器，保障视障群体的平等使用权利；其三是完全免费与普惠，面向全校所有师生员工无偿提供自主研发的工具与导师服务。融入既有工作流的平台生态与系统整合 (10:52) 密歇根大学秉持“AI 即全新用户界面”（AI is the new UI）的核心策略，强调技术必须主动融入用户现有的工作与学习场景，而非强迫用户适应割裂的独立软件。为此，学校自主构建了无代码平台 Maizey，并推出 U-M GPT 工具包以实现全校级别的 API 接入。在教学系统整合方面，密歇根大学早在 2023 年便将生成式 AI 工具无缝嵌入 Canvas 学习管理系统，比 Canvas 母公司 Instructure 宣布官方原生 AI 集成提前了整整两年。这一深度集成的平台架构使技术应用门槛大幅降低，任何教职人员均可在 4 分钟之内配置并上线一个与其教学大纲紧密绑定的专属 24/7 AI 课程导师。此外，自 2024 年 2 月起，全校每名在读学生均获得了专属个人 AI 助手。该系统直接联动学生的学术日程与校园生活服务，不仅能追踪课表、规划精准对齐上下课时间的校园巴士换乘，还能提供校内餐饮站点的实时排队时间预测，实现了技术对日常事务的深度赋能。全民 AI 素养要求与跨学科治理模型 (12:40) Pendse 提出，现代 AI 素养具备与三十年前基础互联网素养同等的关键战略意义。他明确主张，任何高等院校的学生都不应在未修读至少一门 AI 核心课程或未掌握现代 AI 工具应用能力的情况下毕业；这一要求同样适用于职场，现代组织中的员工均应掌握 AI 基本应用技能。他强调，组织管理层有责任提供必要的基础设施与培训支持，而个体员工也需承担主动学习与适应的责任。在推进组织采纳的过程中，技术领袖必须具备穿透表面阻力、体察深层焦虑的能力，理性化解公众对失业或极端科幻场景的无形恐慌。为了建立坚实的组织信任，密歇根大学推行跨学科决策机制。Pendse 告诫称，底层技术选择绝不能仅交由少数工程师决定，而必须主动纳入哲学家、伦理学家和人文学者的视角，并始终让坚决抵制 AI 的教职员工参与顾问委员会，通过持续对话达成全校范围的治理共识。教学、行政与临床医疗的多场景落地 (15:06) 现场披露的多项应用实例展示了生成式 AI 在科研、教学和管理多维度的实际成效。在教学一线，教师基于自身课程大纲构建“数字分身”（digital twins），全天候解答学生疑问并辅助课后练习；亦有教师开发了挂载于个人云端硬盘的自动化助理，专门处理答疑时间确认与作业延期申请等常规后勤咨询，显著减少了日常行政邮件积压。学生则广泛使用 Maizey 平台提炼与梳理期中、期末复习笔记，工程部门也将其嵌入资源中心提供即时技术支持。在专业实践与行政管理领域，学校采购团队运用 AI 工具提升供应商评估效率与采购透明度；医疗健康科研人员与临床医生借助 AI 解析复杂的临床数据集，推动资源匮乏地区的普惠医疗发展；医学教育部门更进一步利用 AI 构建动态虚拟病人交互系统，供医学生开展问诊逻辑与临床诊断模拟演练，有效强化了专业实操技能。密歇根大学的实践表明，构建 AI-Ready 组织是一场深度依赖制度信任、风险共担与全员教育的系统工程，远超出单纯的技术引入范畴。然而，随着基础模型调用成本的累积与算法伦理议题的深化，如何在长期维系完全免费的高校公共技术底座的同时，保持技术架构的财务可持续性并持续消化组织内部的抵触声音，依然是公立学府与大型机构管理层需要持续应对的现实命题。

Detailed notes — zh

## Background 该视频由 Ai4 于 2025 年 8 月 24 日发布在 YouTube 平台，由密歇根大学副校长兼首席信息官（CIO）拉维·彭德斯（Ravi Pendse）主讲。简介表明，本场演讲探讨组织如何有效准备迎接 AI 驱动的转型，重点阐述首席 AI 官（CAIO）在协调新兴技术与业务目标、促进创新以及建立负责任 AI 治理和伦理框架中的战略作用。 ## TL;DR 密歇根大学副校长兼 CIO 拉维·彭德斯分享了该校率先在全校 14 万师生中规模化落地生成式 AI 的战略历程与治理实践。密歇根大学早在 2023 年 8 月便自主搭建了私有化大模型底座与无代码平台 Maisy，以领先行业的节奏实现了与 Canvas 教学系统的集成和 24/7 专属 AI 导师覆盖，同时严格坚守数据隐私、无障碍支持与教育公平三大支柱。拉维指出，首席 AI 官等技术领袖必须具备卓越的同理心与承担“经过计算的风险”的魄力，并倡导打破纯工程思维、引入人文哲学等多元视角参与决策。最后，他强调 AI 素养已成为当代毕业生和职场人的必备基本功，并通过教学、医疗和日常管理等丰富案例展现了 AI 对高等教育生态的全面赋能。 ## Atmosphere & style 拉维的演讲风格幽默亲切且极富谦逊智慧，开场通过调侃自己是“帮不上忙的博士”与援引梵文古训，瞬间拉近了与现场听众的距离。整场分享节奏明快而真诚，既有就大学挂科噩梦与台下展开的欢快互动，也有直面算力成本风险与校园思辨文化的坦率剖白，兼具高瞻远瞩的战略视野与务实落地的执行力。随着生动展现师生日常 AI 实践的宣传片播出以及最后全场齐声高呼“Go Blue”，现场气氛达到高潮，充满科技变革的激昂与感染力。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) 拉维·彭德斯强调在 AI 时代应始终保持谦逊的“终身学习”心态，并将每个人视为师者以应对快速迭代的技术变革。 - [02:22](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=142s) 密歇根大学依托“建设性质疑”（constructive irreverence）的校园思辨文化，不回避争议，通过公开审视与辩论探索 AI 演进路径。 - [04:15](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=255s) 密歇根大学早在 2023 年 8 月 21 日便面向全校 14 万名师生规模化部署了生成式 AI 平台服务，领先于主流商业科技公司。 - [04:15](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=255s) 首席 AI 官（CAIO）的核心使命在于兼顾战略统筹与同理心倾听，化解公众对失业和技术失控的恐慌，洞察未言明的真实诉求。 - [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) 密歇根大学领先业界自建私有化实例与 RAG 架构，在商业企业级方案普及前为全校师生打造了 24/7 在线个性化 AI 导师。 - [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) 密歇根大学的 AI 底座坚定捍卫三大支柱：杜绝外部抓取的数据隐私与安全、针对视障群体的无障碍访问，以及全员免费的教育普惠。 - [10:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=652s) 自主研发的无代码平台 Maisy 践行“AI 即新 UI”理念，支持教职工在 4 分钟内配置上线专属 AI 导师，并早在 2023 年即深度集成 Canvas LMS。 - [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) 拉维公开呼吁当代高校毕业生必须具备 AI 核心课程背景，企业员工也须具备 AI 工具素养，领导者与员工均有学习责任。 - [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) 推动 AI 治理需打破纯工程思维，引入哲学、文学等多学科学者及持怀疑态度的教职员工，构建多元包容的决策团队。 - [14:56](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=896s) AI 在密歇根大学已全方位落地，涵盖教授数字分身答疑、医学临床“虚拟患者”模拟、科研攻坚与行政采购流程优化。 - [17:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=1072s) 自 2024 年 2 月起，学校为每位学生上线专属个人 AI 助手，无缝连接课表跟踪、公交路线指引及食堂排队预测等日常校园场景。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) 幽默开场与终身学习：从“帮不上忙的博士”到敬畏每一位师长密歇根大学副校长兼首席信息官（CIO）拉维·彭德斯（Ravi Pendse）在登台时首先幽默地纠正了主持人的称呼，表示大家直接叫他“拉维”即可。他回忆起儿子 7 岁时的一桩趣事：当时邻居小孩乔伊骑自行车摔倒擦伤，其他小朋友慌乱大喊“彭德斯博士快来，乔伊流血不止”，而拉维闻声狂奔过去时，他那个热心的儿子却骑着小车对同伴喊道：“别找我爸帮忙，他是那种什么忙都帮不上的博士！”拉维打趣道，这种来自亲生骨肉的当头棒喝，即便是现在的 AI 也做不到。如今，这位儿子已经拿到博士学位并在加利福尼亚大学旧金山分校（UCSF）走上教职，拉维还不忘反问他是否也要做“什么忙都帮不上的博士”。玩笑过后，拉维严肃地声明自己绝非无所不知的 AI 专家，而是一名“AI 的学生”，并呼吁在技术日新月异的当下，每个人都应保持求知若渴的谦卑心态。他引用母亲在他 7 岁时传授的一句梵文古训——“Acharya Devo Bhava”（师者如神）：教师如神明般值得敬重，且应当在遇到的每一个人身上发现师者的光芒。因此，与会专家和听众皆可成为他的老师，这种不断学习的心态正是应对前沿科技浪潮的根本。 ### [02:22](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=142s) 密歇根大学的基石：超大规模与“建设性质疑”文化拉维介绍了密歇根大学（University of Michigan）庞大的办学规模与学术生态。作为一所世界一流的大型公办学府，学校不仅拥有庞大的师生社群和雄厚的科研财力，更形成了一种极具辨识度的校园思辨氛围。拉维将这种校园文化风趣地概括为“建设性质疑”（constructive irreverence）——学生要求透明公开与严格问责，教师要求清晰界限与最前沿技术的赋能，所有人每天都在提出尖锐的挑战。正是在这种允许质疑与激辩的环境中，密歇根大学没有回避 AI 治理、算力可持续性影响、大模型幻觉以及算法偏见等敏感争议，反而张开双臂拥抱分歧。拉维指出，唯有经历严苛的审视与公开辩论，组织才能在颠覆性技术浪潮中探索出正确的演进路径。 - **庞大师生与社区规模**：在校学生超过 63,000 人，教职员工数千人，安娜堡（Ann Arbor）校区周边社区近 130,000 人，另有两个分校区。 - **全球顶尖学术实力**：位列全美/全球前十的学术项目数量超过世界上任何其他高等教育机构。 - **雄厚资金体量**：年度运营预算高达 160 亿美元，捐赠基金（Endowment）规模达 210 亿美元，年度科研经费支出约为 20.4 亿美元。 - **全面触达的业务范围**：从本科招生、录取流程、教学实践，到全美规模领先的临床医疗体系，均在广泛探索 AI 的深度融入。 ### [04:15](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=255s) 领跑生成式 AI 浪潮：首席 AI 官（CAIO）的战略使命拉维强调，密歇根大学早在生成式 AI 形成行业狂潮之前便已抢先布局。2023 年 8 月 21 日，密歇根大学面向全校 14 万名师生及社区成员全面上线并规模化部署了生成式 AI 平台服务，这一步甚至早于绝大多数商业科技公司在市场宣传中铺天盖地推销 AI 的时点。拉维坚信，生成式 AI 将是本世纪最具深远影响力的核心技术，其变革威力堪比当年的互联网；如果能够在合规、道德、负责任且深思熟虑的框架下运用，它将成为推动社会积极变革的强大力量。然而，如何穿越技术落地的层层假设与潜在伦理雷区？这催生了对“首席 AI 官”（Chief AI Officer, CAIO）这一角色的迫切需求。拉维指出，无论组织是专设 CAIO 还是由既有的 CIO 等高管兼任，这一领袖必须具备跨界统筹与卓越的同理心。当公众脑海中普遍充斥着《终结者》式的末日危机感、失业恐慌以及世界被 AI 接管的焦虑时，AI 领导者必须能够从容面对误解乃至正面言语攻击，保持冷静倾听，精准洞察人们“说了什么”以及“隐藏在恐惧背后没有说出的真实关切”。 ```mermaid flowchart TD accTitle: 首席 AI 官的核心战略职责与落地架构 subgraph Mindset["战略与沟通素养"] A["战略远见与跨部门协作"] B["化解恐慌与敏锐倾听<br/>(洞察言外之意)"] C["驾驭算力成本与技术未知风险"] end subgraph Execution["技术底座与治理机制"] D["私有化部署实例<br/>(早于商业 Enterprise 模型)"] E["RAG 检索增强生成架构"] F["全天候 24/7 AI 导师"] end subgraph Principles["核心伦理与合规原则"] G["隐私与数据安全 (Privacy & Security)"] H["无障碍访问支持 (Accessibility)"] I["全员普惠零门槛 (Equity)"] end Mindset --> Execution Execution --> Principles ``` ### [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) 领先英伟达的实践：2023 年构建全天候 AI 导师与私有化底座为了说明 AI 在教育辅导中的迫切价值，拉维在现场与观众互动，询问大家大学时期觉得最难啃的课程。台下观众纷纷喊出“贝叶斯数学”、“成本会计”等科目，而拉维则透露自己 50 年前的噩梦是“热力学”——甚至至今仍偶尔会梦见自己热力学挂科导致学位被收回。正因复杂学科对课外辅导有着极高需求，拉维特别提到英伟达（NVIDIA）CEO 黄仁勋（Jensen Huang）在 2024 年 2 月曾公开发言称“未来每个人都应当拥有一位专属 AI 导师”；拉维幽默地表示，自己早已向黄仁勋指出：早在 2023 年 8 月，密歇根大学就已为全校每一位学子提供了 7×24 小时在线的个性化 AI 导师。密歇根大学早在 2023 年便自主运用了检索增强生成（RAG）技术来定制知识库工具，且作为公共学术机构，他们将相关成果与架构经验无偿开放给外界借鉴。拉维回顾道，在 OpenAI 等厂商推出成熟的企业级安全模型之前，密歇根大学便冒着巨大的未知风险搭建了校内私有化实例。上线前一天，他向大学校长直言“我无法预料明天会发生什么，但我需要你做我的后盾”，校长毫不犹豫地给予了信任。当时最大的隐忧在于成本不可控：面对从 5,000 人暴增至 10 万人的潜在并发，Token 计费与资源调度的未知数极大。拉维指出，卓越的 AI 领导者必须具备勇于承担“经过计算的风险”（calculated risk）的决断力与风险缓释能力。 - **领先业界的上线节点**：2023 年 8 月 21 日正式面向 14 万人发布自主搭建的私有生成式 AI 套件。 - **全方位前置 RAG 技术**：在生成式工具中集成检索增强生成，实现课业辅导材料的安全检索与问答。 - **坚定捍卫三大支柱**： - **隐私与安全性（Privacy & Security）**：自建私有化隔离环境，杜绝师生数据被外部厂商模型非法抓取训练。 - **信息无障碍（Accessibility）**：底层适配针对盲人及视障群体的屏幕阅读器，填补商业厂商早期产品的无障碍盲区。 - **教育公平（Equity）**：平台对校内全员永久免费开放，消弭因购买高价商业 AI 工具而产生的数字鸿沟。以下是逐字稿第二部分的详细内容整理： ### [10:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=652s) 密歇根大学的 AI 工具生态与跨校园协作密歇根大学为全校师生免费提供生成式 AI 工具。其中，**Maisy** 是大学自主研发且拥有知识产权的无代码 AI 平台。其核心理念是“AI 即新 UI”，旨在直接将 AI 送到用户手中，而不是让用户来适应技术。早在 2023 年，学校就将生成式 AI 工具与 Canvas 学习管理系统进行了深度整合——相比之下，Canvas 的母公司 Instructure 直到两年后才宣布类似的整合计划。借助 Maisy，教职人员只需不到 4 分钟即可配置并上线一个 24/7 全天候运行的 AI 助教。此外，学校还提供了 **UMGPT Toolkit**，为有开发需求的用户提供 API 级别的访问能力。要在全校范围内规模化推广这些工具，必须依赖全校性质的跨部门协作与伙伴关系。推行过程需要激发他人的热情，同时在承担创新风险与做好风险规避之间取得平衡。无论是设立首席 AI 官（CAIO）还是通过团队协作的方式，领导层都需要具备这种在风险与创新之间进行把控的能力。 - **Maisy 平台**：无代码平台，支持教职员工在 4 分钟内配置完成 24/7 AI 导师。 - **Canvas 深度集成**：密歇根大学在 2023 年即完成了 AI 与 Canvas LMS 的集成，提前行业主流数年。 - **UMGPT Toolkit**：提供 API 层面的开发工具接口，满足定制化需求。 --- ### [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) AI 素养与多元包容的决策机制演讲者强调，AI 素养在当今的重要性不亚于 30 年前的互联网素养。他公开呼吁：**当代任何高等教育机构的毕业生，都必须修读至少一门 AI 核心课程或具备扎实的 AI 工具使用背景**；同样，任何企业的员工都不应以“不会使用 AI”为借口。领导者有责任提供学习资源，而员工自身也有义务主动学习。在推动 AI 的过程中，必须确保所有声音都能被听见。密歇根大学校园内同样存在反对使用 AI 的声音，但学校选择持续与他们对话和沟通。作为一名工程师，演讲者特别提到了多元化视角的重要性：“我不希望由五名工程师决定所有的事，我希望哲学教授、文学教授都能参与到讨论中。”他甚至打趣道，自己公开放弃了纯工程思维并提及“多元化”可能会引来麻烦，现场气氛轻松幽默。谈及首席 AI 官（CAIO）的角色，他指出该岗位涵盖治理、创新、战略与领导力等多重职责，一个人很难面面俱到，因此必须建立一支互相信任、由领导者亲自赢取信任的优秀团队。 ```mermaid flowchart TD accTitle: AI 决策与推广体系 subgraph Inclusion["多元化决策层"] A["工程技术专家"] B["人文哲学学者"] C["持怀疑态度的教职员"] end subgraph Core["CAIO 领导力与团队"] D["战略与治理 (CAIO)"] E["跨学科执行团队"] end subgraph Education["AI 素养普及"] F["学生：必修 AI 核心课程"] G["员工：全员 AI 工具赋能"] end Inclusion --> D D <--> E E --> Education ``` --- ### [14:56](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=896s) 密歇根大学 AI 实践展示视频演讲者展示了一段 2 分钟的宣介视频，生动展现了密歇根大学师生在日常教学、科研和管理中应用 AI 的真实场景： - **教学与虚拟助教**：教授通过 AI 构建自身的“数字分身”（Digital Twin），基于课程资料进行训练，为学生提供 24/7 的答疑与作业练习；另有教师利用 Google Drive 资料创建虚拟助手，自动回答诸如“今天有 Office Hours 吗”、“能否延期交作业”等常见邮件问题。 - **医疗与临床模拟**：医学教育中，AI 被用来生成“虚拟患者”与学生互动；在公共卫生领域，AI 用于处理复杂的健康数据，以在资源有限的环境下实现更高效、公平的医疗护理。 - **科研与学生创新**：工程团队将 Maisy 整合进学生工程团队资源中心，实现按需学习；科研人员借助 AI 帮助学生克服技术瓶颈，加速研究进展。 - **日常学习与管理**：学生使用 Maisy 整理笔记、复习期中与期末考试；行政部门则利用 AI 优化采购流程，提高决策效率。 --- ### [17:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=1072s) 个人 AI 助手与“Go Blue”结语自 2024 年 2 月起，密歇根大学为每位学生配备了专属的个人 AI 助手。该助手能够无缝融入学生的日常校园生活：跟踪课程表、指引前往教室的公交线路、实时监控校园餐厅的排队情况并预测等待时间等。演讲者在演讲最后对能够与大家分享密歇根大学的 AI 经验表示荣幸，并邀请全场听众一同喊出密歇根大学经典的口号 **“Go Blue”**，在热烈的掌声与互动中结束了本次演讲。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) 别找我爸帮忙，他是那种什么忙都帮不上的博士！ — 拉维·彭德斯之子 > [00:04](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=4s) Acharya Devo Bhava — 拉维·彭德斯之母 > [02:22](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=142s) 建设性质疑 — 拉维·彭德斯 > [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) 未来每个人都应当拥有一位专属 AI 导师 — 黄仁勋 > [07:44](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=464s) 我无法预料明天会发生什么，但我需要你做我的后盾 — 拉维·彭德斯 > [10:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=652s) AI 即新 UI — 拉维·彭德斯 > [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) 当代任何高等教育机构的毕业生，都必须修读至少一门 AI 核心课程或具备扎实的 AI 工具使用背景 — 拉维·彭德斯 > [12:40](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=760s) 我不希望由五名工程师决定所有的事，我希望哲学教授、文学教授都能参与到讨论中。 — 拉维·彭德斯 > [17:52](https://www.youtube.com/watch?v=8BtaxwaAnmE&t=1072s) Go Blue — 拉维·彭德斯 ## Mentioned 拉维·彭德斯 (Ravi Pendse), 密歇根大学 (University of Michigan), 加利福尼亚大学旧金山分校 (UCSF), 安娜堡 (Ann Arbor), 黄仁勋 (Jensen Huang), 英伟达 (NVIDIA), OpenAI, Maisy, UMGPT Toolkit, Canvas LMS, Instructure, Google Drive ## Video description Ravi Pendse, Vice President for Information Technology and Chief Information Officer, University of Michigan Abstracts In this session, we will explore how organizations can effectively prepare for AI-driven transformation. Discussions will focus on the expanding role of Chief AI Officers in aligning emerging technologies with business goals, fostering innovation, and establishing robust governance and ethical frameworks for responsible AI deployment. [Source](https://www.youtube.com/watch?v=8BtaxwaAnmE)

Source quotes

[0:04] 拉维·彭德斯之子: 别找我爸帮忙，他是那种什么忙都帮不上的博士！ [0:04] 拉维·彭德斯之母: Acharya Devo Bhava [2:22] 拉维·彭德斯: 建设性质疑 [7:44] 黄仁勋: 未来每个人都应当拥有一位专属 AI 导师 [7:44] 拉维·彭德斯: 我无法预料明天会发生什么，但我需要你做我的后盾 [10:52] 拉维·彭德斯: AI 即新 UI [12:40] 拉维·彭德斯: 当代任何高等教育机构的毕业生，都必须修读至少一门 AI 核心课程或具备扎实的 AI 工具使用背景 [12:40] 拉维·彭德斯: 我不希望由五名工程师决定所有的事，我希望哲学教授、文学教授都能参与到讨论中。 [17:52] 拉维·彭德斯: Go Blue

Source quotes

[0:04] Ravi Pendse: Don't ask my dad for help. He's the kind of doctor that cannot help anybody. [0:04] Ravi Pendse: Acharya Devo Bhava [7:44] Ravi Pendse: I don't know how it's going to go tomorrow. I think it's going to be okay, but I need you to have my back. [10:52] Ravi Pendse: AI is the new UI [12:40] Ravi Pendse: Uh-oh, I did use the diversity word. So, I might be in trouble now... somebody may be waiting outside [15:06] Video Montage: I use AI [15:06] Video Montage: And I am AI. Go Blue. [17:52] Ravi Pendse: Go Blue

Original transcript

[0:04] So my name is Ravi Pensei and I know she [0:07] did introduce me as Dr. Ravi Pence but I [0:09] prefer to go by Ravi and there's a quick [0:11] story behind it that I want to share [0:13] with you. This was when I was when my [0:15] son was about 7 years old and kids were [0:18] playing in the neighborhood on their [0:19] bicycle and as they're riding bicycles I [0:22] was out in the neighborhood checking [0:24] snail mail and I heard this loud sound. [0:27] So I turned around and I saw one of the [0:30] kids Joey had fallen from his bike. So [0:32] as a concerned parent I already started [0:34] walking towards Joey and that time rest [0:37] of the kids were saying Dr. Pen say come [0:39] over quick Joey is gushing blood. So, I [0:43] really got scared and I started running [0:44] now. And Joey wasn't gushing blood. He [0:47] was totally fine. There was just a [0:48] little scratch. And my everpresent son [0:50] who wanted to be helpful, he rode his [0:53] bike over to his friends and said to [0:55] them, "Don't ask my dad for help. He's [0:59] the kind of doctor that cannot help [1:00] anybody." [1:05] Now, tell me, AI can do that. AI cannot. [1:08] Okay. [1:10] Grounding your parents. Now full [1:13] disclosure of that story that same son [1:15] finished his PhD last year and starting [1:18] a faculty position now in few weeks at [1:21] University of California at San [1:22] Francisco and I told him are you going [1:25] to be the doctor that cannot help [1:26] anybody [1:29] but jokes aside uh I am not an AI expert [1:32] and I want to make sure I make that [1:34] clear. I consider myself a student of [1:36] AI. I think everybody should be a [1:40] student of AI in my humble opinion [1:41] because the rate at which things are [1:44] moving rapidly, we all should be [1:46] students of AI. I begin every [1:48] presentation with a recognition of what [1:51] my mother taught me when I was 7 years [1:52] old. She taught me this Sanskrit phrase [1:55] that goes something like this. Acharia [1:58] Devhava Achari is teacher. Teacher is [2:01] made in the image of God and should be [2:04] respected as one would respect God. She [2:07] further taught me to see a teacher in [2:10] everyone I interact with. So if you are [2:13] an AI expert, you are going to be my [2:15] teacher because I'm going to be your [2:17] student as we have this conversation or [2:19] as we talk later. So what about [2:22] University of Michigan? Uh if you're a [2:24] Ohio State grad, you definitely know [2:26] about Michigan, right? [2:29] But jokes aside, uh it's a very large [2:32] public institution. We have 63,000 plus [2:35] students, thousands of faculty members [2:38] and staff, a community of almost 130,000 [2:41] people in Ann Arbor, Michigan, two other [2:44] campuses, more programs in top 10 than [2:47] any other institution in the world. We [2:50] have a budget of $16 billion, one of the [2:53] largest health care system, an endowment [2:55] of $21 billion, and I could and yearly [2:59] research of about $2.04 billion. now so [3:03] it's a large public institution and we [3:06] have a community that is very active [3:10] brilliant individuals I get to work with [3:11] across campus our student demand [3:14] transparency our students demand [3:16] accountability our faculty demand [3:18] clarity they want state-of-the-art [3:20] technology deployed they want to use it [3:23] they challenge us every single day I [3:25] jokingly say on Michigan campus we [3:28] practice constructive irreverence [3:31] let that sink constructive irreverence [3:34] everything is questioned which is great [3:36] and the reason that's relevant is when [3:39] you're having the previous panel [3:40] discussion [3:42] they did not get chance to talk a little [3:43] bit about AI governance and when you [3:46] talk bring AI conversation in all of [3:49] these questions come up people have [3:51] questioned us about sustainability [3:53] impact on AI people have questions us [3:56] questioned us of course about [3:57] hallucinations people have questioned us [3:59] about bias and that's okay that's part [4:02] of being what happens at Michigan. [4:05] That's what we enjoy at Michigan. We [4:07] love the debate. We love the [4:09] disagreements. We embrace those because [4:11] that's how we get better. But that's [4:12] quick introduction to Michigan. So when [4:15] you think about Michigan, [4:17] we believe at Michigan, we started our [4:21] AI journey respectfully way before [4:24] anybody even was thinking about this in [4:26] terms of generative AI. Our services at [4:28] scale at Michigan went live two years [4:32] ago. August 21st, 2023. And I'll tell [4:35] you more story about that more. August [4:37] 21st, 2023 is when University of [4:39] Michigan deployed generative AI services [4:42] at scale to a community of 140,000 [4:46] individuals. August 21st, 2023. Way [4:49] before any other organizations. I'm not [4:51] talking higher education institutions. [4:53] Even companies were thinking about that. [4:56] Overnight many companies did have AI in [4:58] all their marketing material after [5:00] October announcement but that's what we [5:02] did. We believe that generative AI is [5:05] going to be the most impactful [5:08] technology of this century and a force [5:10] for positive disruption if used [5:13] thoughtfully, if used responsibly, if [5:16] used legally, if used ethically. Now [5:19] these are all big ifs. So who is going [5:22] to help you navigate through big these [5:24] big ifs? You're going to need a person, [5:27] a leader, a group of leaders who are [5:29] thinking about this every day. And [5:31] that's where you need someone like a [5:33] CIO. You have to have somebody either a [5:37] designated person or somebody who is [5:39] already in your organization. But that [5:41] person needs to have lot of different [5:43] competencies to think about because if [5:45] you look at this chart here we are [5:47] talking about student enrollment, [5:49] admissions, admission process, education [5:52] and I could go on and on and I don't [5:54] even show all the things we are doing in [5:56] the clinical area with AI right now. [5:58] Okay, I don't even show that. But all of [6:00] these are touched by what's happening [6:02] with AI. I believe that it's as [6:05] impactful as internet has been. it will [6:08] be and we can't bury our head in the [6:11] sand. We have disagreements on our [6:13] campus. We don't make it mandatory but [6:16] we have debate. We have dialogue. We [6:18] encourage people to seriously consider [6:20] leveraging this technology. So taking [6:22] one step further what I'm going to do is [6:25] share with you what we have learned on [6:28] our journey while commenting on CIO [6:32] which I already started talking about. [6:33] This is a person who needs to have the [6:35] ability to work with people across have [6:37] to have visionary ideas. Can they [6:40] collaborate with people? Are they able [6:42] to listen? Are they comfortable in a [6:44] room without losing their patience even [6:46] though people may be attacking you? [6:48] Happens. It's normal. That's that's the [6:51] way it is in academia and beyond because [6:54] frankly whenever [6:57] AI comes in the mind of people maybe not [7:00] this group but people who are not in [7:01] this room typically there is fear there [7:04] is fear of those old Terminator movies [7:07] there is fear of job losses there's a [7:09] fear of you know the world you know [7:12] being taken over by AI and that fear [7:16] then reflects in variety of different [7:19] opinions that are being expressed. And [7:21] so a chief AI officer has to have the [7:23] ability to look under the hood to see [7:25] what people are actually saying and what [7:27] they are not saying. What they're not [7:29] saying sometimes is more important than [7:31] what they are saying. And what happened [7:33] at Michigan was I was able to lead that [7:37] journey even though I serve as the vice [7:39] president and CIO. I have a large [7:41] organization but we saw an opportunity [7:44] and as I before I share that story I [7:46] would love to hear from you. Shout it [7:48] out. What was your hardest class in [7:51] college? [7:53] >> Sorry. [7:55] >> Say that again. Basian math. Okay. [7:57] Basian. Okay. Anything else? [7:59] >> Cost accounting. [8:00] >> Cost accounting. Okay. In my case, my [8:02] hardest class was uh thermodynamics when [8:04] I was in college. [8:07] Even today, even today, 50 years later, [8:11] I get dreams that I failed my [8:14] thermodynamics class and that somebody's [8:17] coming to take my degree away. And I, [8:19] this is crazy, but I don't know if it [8:21] happens to you, but it happens to me [8:22] even today. This is But you know that [8:25] class gives me nightmares. Why am I [8:27] asking that question? I'm asking that [8:29] question because back in February, early [8:31] this year, Jensen Huang, there I [8:32] mentioned his name, so I should be cool [8:34] now. Okay. [8:36] Uh Jensen Huang stated that u every [8:41] person should have an AI tutor. I did [8:43] let Jensen Huang know that at University [8:45] of Michigan, every one of our students [8:48] had access to 24/7 AI tutors in August [8:51] of 2023. [8:53] In August of 2023. Think about that. Had [8:56] access to AI tutors in August of 2023. [8:59] Previous panel talked about rag [9:01] retrieval augmented generation rag. We [9:03] leveraged rack technology back in 2023. [9:07] Okay, so I just want you to know that. [9:08] So we've done a lot of interesting work [9:10] and by the way all of that work is [9:11] available for anybody to leverage if if [9:13] it's helpful to you because we are a [9:15] public institution. We want to support [9:18] all organizations. We are not in here to [9:20] make money on that. But our first tools [9:23] as I shared with you went live in August [9:25] of 2023 [9:27] and we prioritize privacy, security and [9:32] equity. So this is our own private [9:34] instance. This was even before the likes [9:37] of open AI and others had their [9:39] enterprise models. There was nothing [9:41] there. Things didn't exist. We took a [9:44] huge risk. It was a risk because there [9:46] was so many things we didn't know. I [9:48] remember that day when we went live I [9:50] one day before I went to our president [9:51] and I said my boss I said you know hey [9:53] you know I don't know how it's going to [9:55] go tomorrow. I think it's going to be [9:56] okay but I need to need you to have my [9:59] back. He never asked me what that meant [10:01] and he said sure I got your back. Okay. [10:04] Um and what I was worried about was I [10:06] don't know how much it was going to cost [10:08] us candidly because will 5,000 people [10:11] use it or will 100,000 people use it. Uh [10:14] the tokenization and cost we hadn't [10:16] worked things out. So there was some [10:18] risk takingaking involved. It was a [10:19] calculated risk but we took risk. So [10:22] when you think about a chief AI officer [10:24] risk takingaking and risk mitigation [10:26] both are necessary. As we think about [10:28] this, we also focused on, you know, [10:31] accessibility. All of our tools, we [10:34] wanted to make sure those in our [10:35] community who are visually impaired or [10:37] blind can use them by supporting all the [10:40] screen readers. Many of the vendor tools [10:42] even today do not necessarily do that. [10:45] So something to think about and of [10:47] course equity because all of these tools [10:49] are completely free to use for our [10:51] community. They can all use it [10:52] completely free. Uh there's no cost to [10:54] them. We also built um Maisy. Um Mazy [10:58] was is our codeless platform, our own [11:00] intellectual property. But our goal was [11:02] if AI is the new UI, we wanted to bring [11:05] AI to the users, not we are over here, [11:08] you come use it. We wanted AI to be in [11:10] your hands. Do you know that at Michigan [11:12] a faculty member can configure a 24/7 AI [11:16] tutor in less than 4 minutes? It takes [11:19] them four minutes to configure it and [11:20] have it up and running because we built [11:22] integration between Genai tools and [11:26] canvas back in 2023. If you remember in [11:29] structure the canvas founder just [11:30] announced some integration recently. [11:32] Okay, two years ago we did that. So want [11:34] you to be aware and of course we have [11:35] UMGPD toolkit. This is the API level [11:38] access for people to use and we have [11:41] many many different tools available and [11:43] as I indicated to you our journey began [11:46] back in 2023. Now when we began this [11:49] journey and I'm making sure I'm keeping [11:50] track of time here as we began this [11:52] journey it was important for you to know [11:55] that this required campuswide [11:58] collaboration. This required people [12:00] coming together. This required [12:02] partnerships. This required inspiring [12:05] others. this required making sure that [12:08] you understand risk and we take some [12:10] risk but we also focus on risk [12:12] mitigation. So all of these qualities [12:15] are critical as you think about who is [12:18] your CIO. Do you have one? Is somebody [12:21] playing that role? Should you have one? [12:22] In my opinion, you should consider [12:24] having one and if not maybe have a group [12:26] of people doing this work. But all of [12:28] this is needed if you really want to [12:30] roll something out at scale. What I want [12:32] to do is since I'm running out of time, [12:34] we'll skip this and move forward a [12:37] little bit. And [12:40] I have talked publicly a lot about AI [12:42] literacy. I'm on public record saying [12:44] that no student from any institution of [12:47] higher education today should graduate [12:49] without at least one core course in AI [12:52] or significant background to AI tools. [12:54] No student should graduate. If you take [12:57] that further depending on where you are [12:58] from each one of your organizations, no [13:02] employee in your organization, you know, [13:04] should be there and say I don't know how [13:07] to use AI. It is your responsibility as [13:09] leaders. [13:11] If you are an employee, it is your [13:12] responsibility as an employee to make [13:14] sure you learn. There is no excuse. [13:16] There is there are tools available. [13:18] There's information available. We have [13:20] to just step up and just do it. You just [13:23] have to do it because there's no option. [13:25] AI literacy is as critical today as [13:28] internet literacy was 30 plus years ago. [13:30] It's the same thing. You have to know [13:32] it. There's no option not knowing. You [13:34] know, we talked about all voices at the [13:36] table. We need to make sure that all [13:39] voices are heard. As I told you at on my [13:42] campus, we often have disagreements. We [13:44] have people who say that they will never [13:46] use AI. They do not believe that. And [13:48] that's okay. We will continue to have [13:50] conversations with them and educate them [13:52] and work with them. Those voices need to [13:55] be at the table. As a engineer, I always [13:57] say I do not want five engineers making [13:59] all the decisions. I do want that [14:01] philosopher at the table. I do want that [14:03] English person at the professor at the [14:05] table. I want diversity of views at the [14:08] table. Uh-oh, I did use the diversity [14:11] word. So, uh, I might be in trouble now. [14:14] So, [14:16] somebody may be waiting outside. You [14:18] never know. [14:20] I'm just messing around. But uh I have [14:24] already talked about all of this. As you [14:26] think about CIO role, think about all [14:29] the attributes this person needs to [14:30] have. It's about governance. It's about [14:32] innovation. It's about inspiring people. [14:34] It's about strategy. It's about [14:36] leadership. It's about lot of different [14:38] attributes. That's a lot to ask of one [14:41] person. But that's where the team comes [14:43] in. The person has to have a capable [14:45] team, a team that trusts that person. So [14:48] the person has to earn the trust of the [14:50] team. You can't just demand trust. You [14:52] have to earn it. And so what I want to [14:54] do was [14:56] wrap this up by showing this two-minute [14:58] video that list shows you all the things [15:01] that are happening at Michigan. And I [15:03] can talk to you more about that later. [15:04] Let's see if this video will work. [15:06] >> I use AI. [15:18] I use AI because it's important to me to [15:20] work with the tools that are going to [15:22] shape our future. [15:23] >> I use all the great AI tools here at [15:25] Michigan to create a digital twin of [15:27] myself, train on the courses that I [15:30] teach, which I make available to [15:32] students to answer their questions 24/7, [15:35] practice assignments with them, and [15:37] inject more interactivity into the [15:39] classroom than ever before. I use AI for [15:42] a more informed and efficient [15:44] procurement process. [15:45] >> I've used U of Maisy to compile my notes [15:48] uh and use them to study for midterms [15:51] and finals. [15:52] >> I use AI when I integrate Mazing into [15:55] the student engineering team resource [15:58] hub to enable ondemand learning. I use [16:02] AI to transform complex health data for [16:05] actionable insight, fostering meaningful [16:08] impact in the communities we serve, [16:11] enabling efficient, effective, and [16:13] equitable care, especially when [16:15] resources are limited globally. [16:18] >> I use AI a lot to help make complex [16:21] ideas simpler when I especially when I'm [16:23] learning new concepts. I use AI in the [16:26] classroom to allow students to have an [16:29] interaction with a virtual patient. [16:32] >> I use AI to create a virtual assistant [16:34] from my Google Drive. It answers [16:36] students questions like, "Are there [16:38] office hours today?" and "Can I have an [16:40] extension?" It cuts down on a lot of [16:42] emails. [16:43] >> I use AI to help inspire my creativity. [16:47] >> AI empowers my students to work through [16:49] technical challenges and rapidly advance [16:51] their research. [17:17] I use AI. [17:19] >> I use AI. [17:20] >> I use AI. [17:22] I use AI. I use AI. [17:26] >> I use AI. [17:28] >> I use AI. [17:30] >> I use AI. [17:31] >> I use AI. [17:33] >> I use AI. I use AI. [17:37] >> I [17:40] use AI. [17:41] >> I use AI. [17:44] >> And I am AI. Go Blue. [17:50] [Music] [17:52] So, I want to conclude by letting you [17:54] all know that every student at Michigan, [17:56] uh, we released this since February, has [17:58] a personal AI assistant that works with [18:00] them, follows their class schedule, tell [18:03] them which bus they can get on to get to [18:04] their class in time, manages restaurant [18:07] cues, tells them how much is the wait [18:08] time, and I could go on and on, but it's [18:11] been a privilege to share these ideas [18:13] with you. So, if you'll honor me by [18:14] simply saying go blue with me, that'll [18:16] be great. Can you all say go blue? [18:18] >> Thank you. Thank you. There you go. Well [18:19] done. I appreciate it. Thank you. [18:21] [Applause]