Tech Execs Warn AI Headcount Cuts Backfire Without Worker Reskilling

科技高管警告：若不同步进行技能重塑，AI裁员将适得其反 企业推行AI无法单纯通过裁员来提高营业利润率，因为自动化常规认知任务会破坏组织知识储备，除非企业将节省的运营成本重新投资于员工技能重塑。 主持人Ashwin Bharath带领由企业与劳动力领域领导者组成的小组，探讨组织与政策制定者应如何应对AI引发的劳动力替代问题。讨论将传统自动化与AI在认知型白领岗位中前所未有的普及速度进行了对比，并援引了Grainger的计算机视觉库存工具以及Unum的人机协同保险运营等实际案例。小组成员强调AI带来的影响取决于有意识的机制设计，主张与一线员工共同设计工具、利用运营节省的资金支持技能重塑，并推动滞后的公共劳动力政策走向现代化。对话最后对未来的繁荣前景呈现出分歧看法，并强调广泛的经济效益需要积极的适应力、思维方式的转变以及对劳动力的持续投资。 全球科技行业每天因 AI 颠覆流失约 500 个工作岗位，凸显出宏观劳动力替代现状与职场个体乐观情绪之间的剧烈割裂。 (0:00) 本轮 AI 冲击与以往自动化浪潮存在本质不同，其核心特征表现为前所未有的落地速度、B2B 与 B2C 场景的同步扩散，以及对高技能白领知识工作领域的直接渗透。 (4:36) 工业分销商 Grainger 的落地案例表明，部署计算机视觉并未裁减现场团队，而是将员工从繁琐的库存录入转变为高价值的顾问式问题解决者。 (8:26) 以往数字化转型主要局限于面向客户的特定业务触点，而 AI 则作为贯穿各部门的智力与语境层，全方位重塑企业内部的常规运营任务。 (15:40) 企业要实现可持续的人才转型，必须将员工深度纳入 AI 协同设计体系，并将应用 AI 节省的运营成本定向反哺用于现有团队的技能重塑。 (29:42)

While enterprise AI rapidly penetrates white-collar work and drives operational restructuring, panel participants argue that widespread job displacement can be mitigated if organizations treat AI as a task-level augmentative tool, build self-funding reskilling pipelines, and intentionally preserve institutional human expertise. Macro Labor Dynamics and White-Collar Disruption (4:36) Modern AI adoption fundamentally diverges from historical automation waves in its adoption speed, simultaneous penetration of consumer and enterprise channels, and direct disruption of skilled white-collar knowledge work. Ashwin Bharath notes that the global technology sector has been shedding roughly 500 jobs per day amid AI-driven shifts. However, headline layoffs do not necessarily signal an absolute contraction in aggregate labor demand. Hardi Gokani points out that despite sustained restructuring headlines driven by capital reallocation into AI data center infrastructure, Microsoft's net global workforce expanded by 3% between 2023 and 2025. Rather than wholesale elimination of functional departments, operational roles are experiencing rapid internal skill churn and task reallocation across enterprise disciplines. Task-Level Augmentation and Retaining Domain Knowledge (8:26) Enterprise deployment yields better outcomes when organizations dissect workflows into granular tasks rather than attempting generalized autonomous replacement. At W.W. Grainger, field sales teams transitioned from hours of manual spreadsheet data entry across industrial warehouses to scanning parts in minutes via computer vision, redirecting freed-up capacity toward consultative client advisory roles. Similarly, in high-stakes environments like insurance underwriting and disability claims at Unum, repetitive cognitive logging can be handed to AI agents, but critical human empathy during moments of customer distress remains irreplaceable. Preetha Sekharan warns that evaluating AI solely as a headcount-reduction mechanism ignores the heavy economic cost of losing institutional knowledge. Replacing tenured employees with outside technical talent incurs lengthy, costly ramp-up periods. Retaining competitive performance requires uplifting existing staff with supervised AI tooling that preserves clear human escalation paths, balancing efficiency gains with output quality and turnaround speed. Financial ROI, Culture, and Self-Funding Reskilling (25:21) Despite widespread executive interest, real financial returns remain rare: while roughly 99 out of 100 corporate chief executives report using AI, only one or two can demonstrate a measurable increase in EBITDA or operating margins. Kedar Deshpande stresses that AI functions merely as an amplifier of an organization's existing culture and operating strategy rather than a remedy for flawed unit economics. In rigid, quarterly-focused corporate cultures, AI pilots produce superficial metric improvements without impacting top-line revenues. Conversely, customer-focused organizations empower employees to identify bottom-up automation opportunities independently. Reskilling initiatives frequently stall due to schedule compression and compliance restrictions on employee personal time. To overcome this resistance, enterprises can deploy structured "model office" pilots and implement self-funding change management frameworks. Under this model, a dedicated percentage of realized operational expense reductions—such as a portion of $10 million in cost savings—is routed directly back into upskilling incumbent workers rather than being absorbed entirely as margin. Institutional Policy Deficits and Human-Centered Co-Design (32:54) Public workforce policy consistently lags behind technical deployment. Alex Swartsel highlights that the United States' primary federal training framework, the Workforce Innovation and Opportunity Act (WIOA), has languished without reauthorization since 2014, leaving updated bipartisan proposals such as A Stronger Workforce for America Act pending. Effective public and private support requires lowering pragmatic barriers for adult learners by providing childcare, hardware, flexible scheduling, real-time labor market data hubs, and tax credits for employer-sponsored training. At the organizational product level, maintaining human-in-the-loop integrity relies on a structured three-phase model: co-designing tools with front-line workers as active decision-makers; deploying progressive automation where human operators retain override authority; and executing high-value redistribution, freeing staff to handle strategic edge cases as model accuracy matures. Looking three to five years forward, panelists diverge on whether AI will yield broad prosperity or concentrated advantage. Outcomes hinge on individual adaptability, near-term economic absorption of dislocation, and deliberate institutional investment in training; while blue-collar workers transitioning into augmented digital roles may gain leverage, high-earning white-collar professionals reliant on routine intellectual tasks face the sharpest exposure.

行业专家在Ai4研讨会上指出，当前人工智能引发的劳动力冲击已从传统蓝领领域蔓延至白领与认知型岗位；技术能否转化为全员繁荣，取决于企业与政策制定者是否能主动采取“以人为本”的过渡设计，将自动化节省的成本再投资于员工技能提升与人机协同体系，而非单纯追求削减人员编制。 新一轮技术冲击的独特性与广谱性 (4:36) Revature执行主席Ashwin Bharath与Jobs for the Future副总裁Alex Swartsel指出，当前AI驱动的劳动力转型与过去的工业或数字变革有着本质区别。以往的技术浪潮主要集中于遵循规则的特定操作或面向客户的电商交易界面，而现代生成式与应用型AI则是一种贯穿企业所有职能（如财务、库存、客服）的“全谱系转型”，直接冲击重复性认知任务与白领专业领域。此外，AI的高普及速度以及在B2B与B2C渠道的同步渗透，大幅压缩了劳动者与组织的适应周期。据Bharath引用数据，全球科技行业每天约有500个岗位因AI相关的颠覆而流失，这凸显了结构性阵痛的紧迫性。 从替代走向赋能：人机协同与岗位重塑 (8:26) 来自W.W. Grainger与Unum的业务实践表明，AI在企业中的核心价值更倾向于赋能而非全面取缔人力。W.W. Grainger的AI/ML产品管理总监Hardi Gokani以工业品库存管理为例，现场销售团队借助计算机视觉移动应用，将以往长达数小时的手工零件核对压缩至数分钟，这一转变并未消除岗位，而是让销售人员转向高接触面的咨询式顾问角色。Gokani还指出，尽管2023至2025年间科技行业因资本转向数据中心而频繁重组，但微软全球员工总数在此期间实际增长了3%，表明劳动需求在加速演进而非消失。 Unum副总裁Preetha Sekharan则强调，在保险等需要深度共情与复杂判断的敏感行业，企业正将价值链拆分为细分任务，将日常简单流程交给AI代理，但保留人类在重大变故及核保理赔中的关怀纽带。哈佛大学开发的过程智能监控等工具能精准识别哪些工序适于自动化、哪些可被淘汰、哪些应予以辅助增强，从而构建起以数据为依托且以人为中心的工作流。 破解企业培训阻力与政策滞后困局 (22:05) 劳动力转型目前受制于政策滞后与企业内部摩擦。Swartsel提到，美国劳工培训的核心联邦法案《劳动力创新与机会法案》（WIOA）自2014年以来未曾重新授权，虽然两党立法的《建设更强大美国劳动力法案》在推动现代化技能培训，但尚未正式落地。在企业内部，培训计划常因项目紧迫被压缩（例如由10周缩减至2至3周），且受到工时与薪酬合规的制约。 对此，专家建议构建自我造血的转型机制：企业应通过“样板办公室”（Model Office）在局部小规模试点AI工具，并明确将运营成本缩减（如1000万美元OpEx节省）中的固定比例定向重新注资到现有员工的重塑培训中。同时，公共与私营部门需通过税收抵免、弹性课时、提供笔记本电脑以及儿童托育支持，切实消除成年在职人员的学习壁垒。 三阶段落地框架与商业投资回报归因 (32:54) 为了确保人类处于运营架构的核心，研讨会提出了涵盖“共同设计（Co-Design）、渐进式自动化（Progressive Automation）以及高价值重配（High-Value Redistribution）”的三阶段实施框架。终端员工必须以决策者身份参与系统设计，确立明确的人工干预和否决权限，并随着模型精度的提高，将精力释放至复杂战略事务。Sekharan指出，企业抛弃现有员工会付出高昂的隐性代价，因为重新招聘无法替代长年积累的机构专有知识与行业洞察。 前Zappos与Groupon首席执行官Kedar Deshpande警告称，企业不应盲目陷入虚荣指标，受访的一百名高管中虽有99位声称使用AI，但仅有一至两位能展示AI对EBITDA或利润率的实际拉动。Deshpande强调，AI本身是一面放大镜，会暴露企业基础战略与组织文化的缺陷；唯有将技术应用与核心利润池、客户满意度及清晰的财务归因紧密绑定，并以自下而上的员工赋能为底座，AI投资才能真正转化为可持续的股东回报。 针对未来三至五年的前景，与会专家对能否实现“普遍繁荣”存在分歧：Kedar Deshpande与Alex Swartsel认为普遍繁荣具有可行性，但前提是社会大幅加大对技能教育与岗位质量的兜底投入，且劳动者能清晰定义自身独特价值；Preetha Sekharan与Hardi Gokani则强调短期内颠覆阵痛在所难免，其最终成果高度取决于员工的主动适应心态。Ashwin Bharath最终指出，本轮转型的利益分配或将呈现不均，技能升级的传统蓝领可能获得更大的杠杆，而高薪白领专业岗位反而面临最显著的结构性替代风险。

## Background 本内容由 Ai4 于 2025 年 8 月 28 日在 YouTube 平台发布。视频记录了一场题为《应对 AI 驱动的职业替代：设计通用转型框架》的圆桌讨论，汇集了来自政策研究智库、零售与工业供应链巨头以及保险金融领域的跨界专家，旨在探讨在 AI 颠覆劳动力市场的背景下，如何通过技能重塑、强化社会保障与公私协作来设计前瞻性的劳动力转型方案。 ## TL;DR 圆桌讨论围绕 AI 是否能做到“亲劳动者”（Pro-worker）展开，专家们指出当前 AI 正在全方位渗透传统白领与智力密集型岗位，单纯依赖滞后的公共政策无法应对急速变革。来自固安捷与 Unum 的一线落地实践表明，成功的转型依赖于“AI 增强”而非单纯裁员，应通过“协同设计、渐进式自动化与技能跃升”的三阶段框架将员工从机械流程中解放出来，专注于高价值客户沟通与共情服务。嘉宾们进一步呼吁建立“公共劳动力数据中心”与税收抵免等政策杠杆，企业端则需将降本增效带来的运营节余有意识地反哺到员工再培训中，以确保人机协作红利能够转化为普惠性繁荣。 ## Atmosphere & style 整场研讨会在主持人的幽默风趣调侃与现场民调互动中拉开序幕，随后迅速切入对科技裁员、技能脱节与 ROI 困境的冷峻审视，形成了兼具行业洞察力与人文关怀的对话张力。台上来自一线产业实践者与劳动力政策专家的思维交锋务实而接地气，既打破了关于 AI 全面自动化的神话，又直面了企业在培训时间与短期财务回报上的现实拉锯，整体氛围严谨务实且充满建设性启发。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=0s) 大众对 AI 就业影响的整体判断五五开但对个人前景普遍乐观，然而科技行业当前平均每天约有 500 个岗位因技术替代而消失。 - [04:36](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=276s) 本轮 AI 浪潮首次直面高技能白领岗位，由于公共政策存在滞后性，企业与教育机构必须主动开展自下而上的“以劳动者为中心”转型设计。 - [08:25](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=505s) 工业一线的落地实践表明 AI 的核心哲学应为“人机增强”而非简单替代，通过自动化例行工作赋能一线员工转型为高阶策略顾问。 - [10:24](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=624s) 科技行业的裁员潮本质上是资本向 AI 基础设施转移的组织再平衡，在局部重组的同时整体用工与新技能岗位仍在持续演进与扩张。 - [14:21](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=861s) 传统教育体系亟待革新，在 AI 时代除技术素养外，人际共情与人机交互沟通等复合型软技能成为必须突破的新培养盲区。 - [16:05](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=965s) 与以往仅触及局部流程的数字化转型不同，AI 转型全面侵入组织的智力层，严重依赖机械重复规则的岗位面临极高的替代风险。 - [18:20](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1100s) 尽管常规流程加速被智能体接管，但在理赔等核心痛点场景中，人类特有的深层共情与信任建立仍是任何 AI 无法取代的壁垒。 - [22:05](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1325s) 推动劳动力技能重塑需要升级公共数据基础设施与雇主税收抵免等政策杠杆，并消除托儿、硬件等阻碍普通劳动者参与培训的现实壁垒。 - [25:21](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1521s) 企业在 AI 落地中普遍面临“99:2”的 ROI 困局：绝大多数企业均在应用 AI，但仅有极少数能够将其直接转化为息税折旧摊销前利润（EBITDA）的增长。 - [27:39](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1659s) 坚持“客户第一与员工赋能”的企业文化能大幅缩短 AI 技能回报周期，让员工掌握转型的主导权比一刀切的硬性命令更有效。 - [30:46](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1846s) 企业应当通过设立试点办公室精细管理变革，并有战略自觉地将 AI 节约的运营成本资金有意识地反哺于员工的再培训与技能提升。 - [31:53](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1913s) 构建以人为本的人机协同系统需依托三阶段落地框架：前期协同设计、中期保留人类一票否决的渐进式自动化，以及后期的持续技能跃升。 - [35:02](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2102s) AI 不仅能降低成本，更能提升业务效能与交付质量；企业重新培训携带丰富“机构知识”的老员工，往往比外部招募更具长期价值。 - [36:48](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2208s) 将 AI 作为业务乘数力量必须建立对利润和收入的直接可归因性，企业能否释放技术价值根本上取决于自身核心商业模式的清晰度。 - [39:24](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2364s) 未来 AI 红利能否实现普惠取决于系统性投入与个体的学习心态，短期内劳动市场难免经历阵痛，但长期将赋能蓝领向白领技能跃升。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=0s) 开场民调：AI 究竟是在消灭工作还是创造机遇？ 主持人 Ashwin Bharath 以一种轻松幽默的基调开启了这场名为“应对 AI 引发的岗位替代：设计通用转型框架”的圆桌讨论。他打趣道，这个又长又拗口的学术标题用大白话说，其实就是探讨“我们如何获得更好的工作，或者怎样才能不被 AI 抢走饭碗”。在这个大众普遍关注技术颠覆却鲜少深入探讨应对机制的当下，Ashwin 在正式引入讨论前，先向现场观众做了一轮快速举手表决，以摸清台下的情绪基调。 当被问及“是否认为 AI 创造的新岗位会多于它所消灭的岗位”时，现场观众的反应呈现出势均力敌的五五开态势。而当问题转向个人切身利益——“你认为自己的工作会受到 AI 的积极赋能还是负面冲击”时，绝大多数观众都举手认可积极影响，仅有极少数人认为会受到负面冲击。Ashwin 赞赏了现场弥漫的乐观情绪，但同时也直面冰冷的现实数据：目前仅在全球科技行业，每天就有大约 500 个工作岗位因 AI 受到冲击或消失。他指出，阵痛期是客观存在的，而大规模的新增就业创造只是一个时间问题。 - **现场民调对比**：关于整体就业净增减，现场乐观派与悲观派各占一半；但在个人职业发展上，绝大多数观众坚信 AI 会带来正向赋能。 - **行业现实数据**：当前全球科技领域平均每天因 AI 替代而损失约 500 个工作岗位。 --- ### [01:41](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=101s) 跨界嘉宾亮相：从零售巨头、工业供应链到保险与劳动力政策 随后，四位背景跨越企业高管、一线技术落地与劳动力政策研究的专家依次亮相。Ashwin 在介绍时风趣地调侃，固安捷（W.W. Grainger）的 Hardi Gokani 可不是那种“拿着昂贵 PPT 空谈 AI 概念”的人，而是实打实在工业现场推进落地的实干家；而对于来自大型人寿及伤残保险公司 Unum 的 Preetha Sekharan，Ashwin 指出她所负责的理赔与核保业务正处于风口浪尖——有人预言这些领域将被 AI 全盘接管，也有人坚信它们会获得前所未有的赋能。为了活跃气氛，主持人还开玩笑称“深度学习之父”Geoffrey Hinton 承诺会与前三位提问的观众合影，引得台下观众纷纷回头张望。 各位嘉宾随后简要介绍了各自的实战维度与关注重点： - **Kedar Deshpande**（Zappos 与 Groupon 前 CEO）：目前深度参与多家初创企业的 AI 赋能与落地指导，对技术迭代的惊人速度有着第一视角体验。 - **Alex Swartsel**（Jobs for the Future / JFF 助理副总裁）：代表全美知名非营利组织 JFF，致力于推动教育、公共政策与劳动力体系改革，帮助普通劳动者获得高质量技能培训并实现经济阶层流动（Economic Mobility）。 - **Hardi Gokani**（W.W. Grainger AI/ML 产品管理总监）：掌舵这家工业分销巨头的 AI/ML 产品战略，聚焦于以可衡量的商业结果推进全企业范围内的 AI 采纳。 - **Preetha Sekharan**（Unum 数字化创新与应用 AI 副总裁）：全面统筹这家老牌保险服务商的 AI 顶层设计、规模化构建与业务场景部署。 --- ### [04:36](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=276s) “以劳动者为中心”的转型：应对白领岗位替代与政策滞后 讨论切入核心议题：AI 毫无疑问是“亲资本”（Pro-capitalism）的，因而吸引了千亿美元的巨额注资；它也是“亲人类”（Pro-mankind）的，在医疗健康等尖端科学领域突破不断；但问题在于，**AI 能够做到“亲劳动者”（Pro-worker）吗？** JFF 的 Alex Swartsel 对此给出了坚定的肯定回答，但她着重强调了一个前提：“我们必须主动将其设计成那样（we have to design it that way）。”她借用 Geoffrey Hinton 当天早些时候提到的“母性 AI”（Maternal AI）概念，指出当前技术浪潮与以往历次自动化的本质不同：AI 的采纳速度极快，在 B 端与 C 端几乎同时扩散，并且首次直面所谓的高技能或白领岗位（White-collar jobs）。然而，现有的教育、就业与社会保障体系极其僵化，根本无法适应这种剧烈冲击。这种脱节将加剧那些原本就面临经济壁垒人群（如无四年制本科学历者、有犯罪记录者、少数族裔及特定职业女性）的脆弱性。 针对“究竟该由谁来主导设计新体系”的追问，Alex 坦言不存在包治百病的“灵丹妙药”（Silver bullet）。虽然公众习惯寄希望于立法，但公共政策在技术变革面前往往是严重的滞后指标（Lagging indicator）。如果各方都在等待一份完美的权威宏观数据出炉再行动，世界早已将我们甩在身后。因此，拥有资源抓手的大型雇主、高校联盟和社区发展组织必须率先在局部开展沙盒式试验，自下而上地将成功经验推向制度化政策。 ```mermaid flowchart LR accTitle: AI 劳动力转型协同架构 accDescription: 展示从微观试验到宏观政策与劳动者赋能的演进路径 subgraph Micro["微观实践与试验 (先行驱动)"] direction TB E["大型雇主企业"] U["高校与培训系统"] C["社区经济组织"] end subgraph Design["核心设计准则"] direction TB H["以人类独有能力为核心"] B["创造可衡量的商业价值"] end subgraph Macro["宏观制度与阶层流动"] direction TB P["自下而上推动政策变革"] M["实现劳动者经济流动性"] end Micro --> Design Design --> Macro ``` --- ### [08:25](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=505s) 岗位并非消失而是“被增强”：固安捷的工业现场 AI 赋能实践 面对主持人和产业界关于“岗位到底是在被摧毁还是在转型”的疑问，来自 W.W. Grainger 的 Hardi Gokani 结合自身业务给出了答案：在真实的工业供应链一线，岗位的演进不是简单的替代，而是深度的“岗位转型”（Job transformation）。固安捷的业务高度依赖深入客户现场的外勤与销售团队，他们需要精准掌握客户在 MRO（维护、维修与运营，如安全防护、管道工具等维系设备运转的物资）方面的复杂需求。 固安捷推进 AI 的核心哲学是 **“AI 增强”（AI Augmentation）** 而非替代人工。Hardi 举了一个典型的业务场景：过去在开拓新客户时，销售人员必须亲自深入环境错综复杂的工业仓库，在海量无规律摆放的物资堆里手动记录规格参数，再耗费数小时将数据录入 Excel 表格进行复杂计算。如今，固安捷为外勤团队配备了集成计算机视觉（Computer Vision）的 AI 移动应用，员工只需用手机摄像头扫描盘点，AI 即可自动识别零部件型号并建立库存清单。这不仅将一线员工从枯燥繁琐的手工制表中解放出来，更将他们转型为专注于客户高阶需求与策略建议的专业顾问。 ```mermaid flowchart TD accTitle: 固安捷现场资产盘点模式对比 accDescription: 对比传统手动盘点与 AI 增强盘点在效率与员工角色上的差异 subgraph Past["传统模式 (Pre-AI)"] direction TB A1["深入复杂工业仓库现场"] --> B1["纸笔/手动收集设备物资信息"] B1 --> C1["耗费数小时在 Excel 中录入与计算"] C1 --> D1["员工角色：低效的数据搬运工"] end subgraph Present["AI 增强模式 (Augmented)"] direction TB A2["深入复杂工业仓库现场"] --> B2["计算机视觉移动端即时扫描识别"] B2 --> C2["AI 自动解析并同步系统库存清单"] C2 --> D2["员工角色：聚焦客户策略的高价值顾问"] end ``` ### [10:24](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=624s) 效率跃升背后的工作重塑与科技裁员真相 在现场作业中，原本需要耗费数小时的数据采集工作，如今通过 AI 工具仅需几分钟即可完成。这不仅带来了显著的生产力提升，更重要的是重塑了员工的实际工作方式——外勤人员不再需要把时间浪费在低头盯着 Excel 表格上，而是能够将精力投入到真正有价值的事情中，比如与客户面对面沟通、倾听需求并解决实际痛点。 谈及 AI 引发的岗位替代问题，就必须直面科技行业近期裁员潮这一“房间里的大象”（elephant in the room）。以微软（Microsoft）为例，自 2023 年以来，微软因多轮业务重组和裁员频繁登上新闻头条，其核心目的是将更多资金投向 AI 基础设施建设。然而值得注意的是，如果对比微软 2023 年至 2025 年的员工总数，其整体用工规模实际上增长了 3%。这表明岗位确实在经历结构性重构与技能升级，市场上创造出的新岗位总数甚至超过了三至五年前的水平。归根结底，核心命题在于如何实现人机协同（AI and humans working together），以创造更高、更高效的成果。 ```mermaid flowchart TD accTitle: AI引发的组织变革与劳动力结构演进 subgraph S1["传统工作模式"] A["耗费数小时手动录入数据 / Excel 报表"] end subgraph S2["AI 赋能下的业务重塑"] B["AI 自动化处理例行数据"] --> C["员工聚焦高价值任务<br/>（面对面沟通与解决复杂问题）"] end subgraph S3["组织劳动力再平衡"] D["局部重组与低效流程淘汰"] --> E["资本转向 AI 基础设施"] --> F["新技能岗位增加与整体规模增长"] end A --> B C --> D ``` --- ### [11:48](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=708s) 来自中小企业视角的发问：教育滞后与员工权益保障 一位致力于帮助中小企业构建“增强型 AI”（Augmented AI）定制平台的现场参会者提出了两个深层困惑。该机构的目标是通过人机协作协助企业平稳重组，尽量避免员工遭遇解雇。 该参会者向 Alex Swartsel 抛出第一个问题：当前教育体系在 AI 时代严重滞后。作为具备跨学科背景的人士，他深感学校现行课程早已过时，却苦于没有成熟的体系教导学生如何将 AI 融入商业实践，教育界普遍处于“知晓风暴将至却不知其究竟为何物”的迷茫状态，因此急切想了解教育系统幕后是否有革新教学大纲的具体推进计划。第二个问题则直指政策空白：当大型科技巨头（如 Google、Microsoft）依赖麦肯锡等咨询机构大刀阔斧裁员，而中小企业自身难保、政府公共资源又极度匮乏时，各州政府亦难以单独立法，未来究竟由谁来设立防线以保护普通员工的基本权益与职业尊严？ --- ### [14:21](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=861s) AI 基础素养普及与跨学科软技能的教育鸿沟 针对教育领域的提问，Alex Swartsel 阐述了当前在推广 AI 素养方面取得的实质进展。美国政府近期发布的《AI 行动计划》（AI Action Plan）以及此前的相关行政命令中，均重点强调了对 K-12 基础教育阶段的推动；例如美国教师联合会（American Federation of Teachers, AFT）已与微软展开深度合作，致力于一线教师的技能提升；同时，AI Education Project (aiEDU) 以及 AI for Education 等专业组织也在全力推进，将 AI 素养融入基础教育体系。 尽管在“AI 素养课程该如何设计”以及“如何安全审慎推进”方面有所突破，但整体进度仍远未达到理想要求。更深层的挑战在于复合型软技能的培育：在 AI 时代，劳动者不仅需要探索全新的人际沟通与共情方式，更要掌握与机器对话交互的能力。当前的传统教育体系尚未摸索出这套复合能力的培养路径，唯有依靠整个社会系统的协同发力才能打破僵局。 - **K-12 政策支撑**：白宫 AI 行动计划与行政令明确要求加大基础教育阶段的 AI 介入。 - **师资与机构协同**：AFT 联合微软开展教师培训，aiEDU 等机构加速课程方案落地。 - **技能新维度**：除技术操作外，人际共情与“人机交互沟通”成为传统教育尚未覆盖的新盲区。 --- ### [16:05](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=965s) 从“数字化转型”到“AI 转型”：无死角的智力层侵入 曾掌舵 Zappos 与 Groupon 两大数字化颠覆型企业的 Kedar Deshpande 深入剖析了“数字化转型”与当前“AI 转型”的本质差异。过去的数字化转型主要聚焦于客户触点和前端体验（如线下进店购物向线上系统迁移，改造的是局部业务流程），而 AI 转型则是一场全方位、跨领域的深层渗透（transformation across the board）。 任何岗位中的重复性工作首当其冲都会被 AI 抹平，无论是仓储库存盘点、会计记账还是客户服务皆是如此。更关键的差异在于“落地速度”与“替代机制”：昔日的数字化变革淘汰了部分旧工序，但迅速衍生出大批新职业；而在 AI 时代，由于技术已经渗透到了“智力层”（intellectual layer），若员工在组织中的核心价值仅仅建立在机械重复规则（以往算法因只能执行简单的“如果 A 则执行 B”而无法替代人类），那么这类岗位将面临极高的替代风险。AI 转型展现出了远比数字化时代更为激进的“侵入性”（invasive）。 --- ### [18:20](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1100s) 保险业前沿落地：例行流程代理与不可替代的人性共情 面对传统自动化（Automation）与情境感知智能体（Context-Aware AI Agents）之间界限的探讨，Unum 的 Preetha Sekharan 分享了保险巨头在业务一线探索中的审慎态度。她指出，目前的智能体技术尚未成熟到可以在全场景下部署通用智能体的阶段。Unum 的做法是细致拆解整个业务价值链，甄别各类任务属性： 对于高度重复、规则明确的常规事务，企业正加速引入 AI 智能体来承接；但在涉及出险理赔等核心业务时，客户往往正处于遭遇重大变故与个人痛苦的关键时刻，电话另一端所迫切需要的，是一个具备高度共情心、能够在人性层面建立深层信任的真人，这种情感连接是任何智能体都无法替代的。保险业虽为数据密集型行业，AI 必将带来深远变革，但最终呈现的形态必然是“人机协作”（human-machine collaboration），绝非自动化对所有岗位的全面接管。主持人对此补充指出，行业变革正在以惊人速度演进，例如已有哈佛大学教授创立的企业，通过在员工设备上全量捕捉键盘敲击、鼠标移动与触屏轨迹，全面建模人类与数字世界交互的底层流程，推动深层次的自动化重塑。 ### [20:55](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1255s) 流程自动化的冲击与公共政策工具的更新 主持人感叹当前 AI 工具在理解业务流程语境（process context）方面的精准度令人震惊。这些系统能够极其明确地指出：“这个环节应当被自动化，这个环节应当被淘汰，它纯粹是在浪费时间，而这个环节则应当由 Agent（智能体）来进行增强。”这种颠覆性的效率让他开玩笑说，自己甚至“由衷庆幸自己快 50 岁了，不再处于职业生涯的巅峰期”，因为这种演进确实令人感到一丝恐惧。 随后讨论转向了解决方案。主持人指出，依靠政府监管往往“说起来容易做起来难”——社会连社交媒体的监管问题都尚未理顺，更遑论 AI。同时，当前有些政策环境甚至主张不应对 AI 施加过多监管。因此，主持人向 Alex Swartsel 提问：像《劳动力创新与机会法案》（WIOA，Workforce Innovation and Opportunity Act）这类由美国劳工部主管的既有机制，是否能够被改造并升级，以专门用于应对当前的劳动力技能重塑（upskilling）需求？ ```mermaid flowchart TD accTitle: 劳动力政策与多方协同框架 subgraph Policy ["政策与公共部门 (Policy & Public Goods)"] WIOA["WIOA 法律更新 & 激励政策"] DataHub["实时劳动力市场数据中心 (Data Hub)"] TaxCredit["雇主再培训税收抵免 (Tax Credits)"] end subgraph Employers ["雇主与组织变革 (Employers)"] QualityJobs["打造高质量岗位 (薪酬/福利/灵活性)"] Culture["以客户为中心与员工赋能文化"] end subgraph Workforce ["劳动者转型支持 (Workers & Learners)"] BarrierRemoval["扫除现实障碍 (育儿支持/非工作时间培训/设备)"] StudyGroups["自发学习小组与集体知识积累"] end Policy --> Employers Employers --> Workforce Policy --> Workforce ``` ### [22:05](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1325s) 政策杠杆、公共数据基础设施与打破转型障碍 Alex Swartsel 强调，升级政策工具势在必行，因为现行的 WIOA 自 2014 年以来就未曾重新授权过。虽然去年年底通过的《美国更强劳动力法案》（Stronger Workforce for America Act）尚未正式生效，但它包含了许多优秀的改革举措，例如将资金更精准地聚焦于技能开发、积极鼓励雇主参与等。此外，正如美国劳工部的 Taylor Stockton 当天提到的，政府行动计划中还包括建立“公共研究中心”（Research Hub）。这一类公共基础设施能以统一力量收集、维护和整理高质量的“实时劳动力市场数据”，解决单个私营机构由于成本过高而无法独自承担的痛点。 在财政杠杆方面，Alex 提出政策应进一步扩大雇主为在岗员工及周边劳动力提供技能重塑培训的税收抵免（Tax Credits）。与此同时，雇主和培训机构需要共同发力，审视所创造的岗位是否属于“高质量岗位”（提供体面薪资、健全福利、灵活性与晋升路径），并思考如何通过 AI 赋能让工作变得更优质。 更重要的是，必须“深入劳动者的实际处境”（meet people where they are），打破他们获取技能的隐性壁垒： - **现实生活障碍**：例如需要养家糊口的单亲母亲，根本无暇脱产参加全日制大学课程或短期编程训练营。 - **具体支持措施**：提供灵活的非工作时间/夜间培训、配套托儿服务（child care），以及为缺乏硬件的劳动者发放笔记本电脑等设备。 ### [25:21](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1521s) 资本主义视角下的投资动力与 99:2 的 ROI 现实 主持人接着向 Zappos 和 Groupon 的前 CEO Kedar Deshpande 抛出难题：从自由市场资本主义的角度来看，“无利不起早”，在政府资金难以企及的情况下，究竟什么才能促使企业管理者自掏腰包投资员工的 AI 转型？ Kedar 分享了一个耐人寻味的真实故事：在他近期与另一位前 CEO 同行交流时，对方提到在一次有上百位 CEO 参与的闭门会议中做过现场调查。当被问及“组织内有多少人在使用 AI”时，全场近 100 人中有 99 人举手；然而当紧接着问到“有多少人真正借助 AI 提高了息税折旧摊销前利润（EBITDA）、利润率或总利润池”时，举手的却仅有区区 2 人。这揭示了当前企业 AI 落地的普遍困局——效率虽有提升，但与净利润增长之间缺乏直接的可视化关联。 - **99:2 的残酷差距**：几乎所有企业都在使用 AI，但能够转化为直接利润增长的仅占极少数。 - **技术浪潮是企业核心的试金石**：AI 就像以往的所有技术浪潮一样，既推动变革，也彻底暴露企业内部的缺陷。如果企业的底层文化缺乏变革基因、只关注单一季度的短期战术指标，AI 很难发挥价值。 ### [27:39](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1659s) 从 Zappos 的 Excel 宏到 AI 赋能的组织文化 Kedar 以自己在 Zappos 的管理经历深入阐述：如果企业真正坚持“客户第一”与“员工赋能”的文化，AI 的投资回报率（ROI）就会显现得非常迅速。五年前在 Zappos，一名一线客服员工因为厌烦了日复一日回答相同重复的客户问题，不仅没有单纯向上级抱怨，反而主动在业余时间自学 Excel 宏命令（Macros），整理出系统化的分析表格，并跨部门向其他团队演示如何为公司和客户节省时间。 Kedar 指出，当年这名员工可能需要花上整整一个月去自学该技能，而如今借助 AI，同样的学习和自动化过程“可能在一天之内就能完成”。因此，驱动企业投资 AI 的关键在于明确的核心价值观——当赋能员工解决实际业务痛点成为常态，ROI 的释放是水到渠成的。主持人对此表示高度赞同，并补充道，大型企业转型的核心在于“让员工掌握转型的主导权”，采取系统化的变革管理（Change Management）路径，而非一刀切地下达硬性命令。 ### [29:42](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1782s) 培训落地的现实拉锯战：工时、预算与意愿 谈及技能培训在企业的实际推行，主持人分享了 Revature 在企业端经历的戏剧性现实拉锯： 1. **初期热情高涨**：高管层口号喊得震天响（“这是 CEO 的直接督办项目，不惜一切代价，就算 6 个月全职培训都没问题”）； 2. **课程不断压缩**：Revature 提交了 10 周的课程大纲后，企业要求缩短到 6 周；随后项目经理介入抱怨员工业务繁忙，课程又被砍到 3 周； 3. **最终妥协与合规困局**：最终员工每天甚至连 2 小时的学习时间都抽不出来，企业转而要求员工在“非带薪时间”自学，结果又引发人力资源部门（HR）关于加班费与劳动法合规风险的阻拦。 主持人就此向 Preetha Sekharan 提问：在员工都必须维持生计的现实下，企业该如何在带薪工时与工作负荷之间合理分配 AI 培训？Preetha 幽默地回应道：“我个人一直认为，‘变革只要发生在别人身上，那就是极好的’（Change is all good when it is happening to somebody else）。”她强调，真正的解决方案始于如何让员工从一开始就具备包容性，让他们亲自去参与和塑造这场即将影响他们自身的转型流程。 ### [30:46](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1846s) 变革管理与价值实现：将运营结余反哺于员工技能重塑 在企业部署 AI 产品的过程中，必须将员工置于核心位置并分阶段推进。Preetha Sekharan 强调，在全面推广前设立“典型办公室”（Model Office）至关重要——先在一小部分精选人员中进行测试，观察需要调整的流程与痛点。这种变革管理（Change Management）不是事后补救措施，而是从项目初期就深度嵌入到各个阶段之中。 不仅在实施阶段需要精细管理，在项目落地后的价值实现（Value Realization）环节同样需要刻意规划。例如，当企业通过 AI 实现了 1000 万美元的运营成本（OpEx）节约时，绝不能将这笔收益全额直接计入利润底线（Bottom Line），而是应当有意识地将其中一部分资金重新回流，用于员工的重新培训（Reskilling）与技能提升（Upskilling）。这种良性循环不会自然发生，必须依赖管理层的战略自觉。 ### [31:53](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1913s) 年轻世代与人机协同：三阶段“以人为本”设计框架 主持人提出一个现实隐忧：年轻一代（例如自己热衷 AI 的 17 岁与 13 岁孩子）往往沉浸在未来科技的狂热中，甚至近期有如估值飞涨的 Perplexity 竞标收购 Google Chrome 这类颠覆性事件，年轻一代可能难以体会 AI 替代岗位所带来的人文阵痛。当他们未来主导 AI 叙事时，企业该如何确保“人机协同”（Human-in-the-loop / Augmentation）的理念代代相传？ ```mermaid flowchart LR accTitle: AI产品三阶段渐进式落地框架 subgraph S1["阶段一：协同设计"] A["业务人员深度嵌入"] --> B["参与交互与决策制定"] end subgraph S2["阶段二：渐进式自动化"] C["AI生成初步结果"] --> D["人类审核与一票否决"] D --> E["模型迭代精准度提升"] E --> F["人类转向高价值决策"] end subgraph S3["阶段三：技能跃升"] G["培训AI监督能力"] --> H["保留最终人类升级路径"] end S1 --> S2 --> S3 ``` 针对这一代际与技术挑战，Hardi Gokani 提出了“以人为核心”构建 AI 的三阶段闭环方法论： * **阶段一：协同设计（Co-design）**：将最终用户与员工引入设计初期，让他们不仅提供反馈，更直接参与决策，共同决定如何交互与审核 AI 生成的产出。 * **阶段二：渐进式自动化（Progressive Automation）**：坚决避免“非人即机器”的一刀切切换。初期由员工耗时审查 AI 输出并拥有“最终否决/覆盖权”；随着数据积累与模型准确率提高，员工逐渐摆脱重复性劳动，将精力转移到关乎公司营收指标的高价值事项上。 * **阶段三：技能跃升（Skill Uplifting）**：技能升级绝不能是事后诸葛亮。企业必须主动培训员工如何与 AI 互动、如何有效监督 AI，并始终保留向人类专家升级判决的通道，发挥人类独特的批判性思维与主观判断。 ### [35:02](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2102s) 超越单纯降本：AI 的综合价值主张与组织沉淀 当技术服务商向企业 CEO 兜售“AI 能够大幅省钱和压缩时间”时，如果一位兼顾人文关怀的 CEO 试图向董事会阐述非“裁员/降本”导向的过渡路径，该如何构建商业逻辑？ Preetha Sekharan 指出，如果仅仅把 AI 视为提升效率（Efficiency）的工具，视界就过于狭隘了。AI 实际上是一个“三位一体”的价值驱动引擎：除了效率，它更能显著提升工作效能（Effectiveness）、交付质量以及业务交易周转速度。此外，大型组织内部蕴藏着极其庞大的“机构知识”（Institutional Knowledge）。如果现有技能不匹配，重新培训与跨领域培养现有员工往往远优于直接从外部招聘——因为新员工建立对复杂业务背景的熟练度（Ramp to proficiency）需要漫长周期，而老员工所携带的企业历史沉淀是无可替代的资产。 ### [36:48](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2208s) 股东回报与工作转型：建立可归因的商业乘数效应 针对私募股权（PE）基金对企业 CEO “股东回报第一”的信托责任要求，Kedar Deshpande 阐述了如何平衡劳动力转型与财务回报。他回忆道，当年数字化转型浪潮中，不少团队宣称“网页 A/B 测试让转化率提升了 10 个基点”，但最终公司总营收却毫无波澜，这种缺乏真实归因的改造对股东毫无意义。 AI 应当被定位为一种“业务乘数力量”（Multiplying Force），必须具备立竿见影的直接归因能力（Direct Attribution）： - 无论是成本节约还是客户满意度提升，都必须能直接映射为收入池或利润池的实质增长。 - 相比过去需要耗时数月甚至数年的传统系统重构，AI 能够在短短数天内验证并展示成效，兼具速度与落地确定性。 - 最核心的前提依然是商业本质：如果一家公司自身连“核心业务如何赚钱、如何创造收入”都未厘清，AI 也无法凭空拯救它。 ### [39:24](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2364s) 终极对谈：AI 究竟会带来普惠繁荣还是少数人的繁荣？ 在对谈的最后三分钟，主持人抛出了核心议题：展望 3 到 5 年后，AI 带来的究竟是“普惠大众的繁荣”（Prosperity for all）还是“少数人的繁荣”（Prosperity for few）？各位嘉宾给出了各自审慎而深刻的研判： * **Kedar Deshpande**：整体上会创造**普惠繁荣**，但对于那些固步自封、不愿做出改变，或是无法清晰证明自身价值增量的人群，将会出现严峻的发展鸿沟。 * **Alex Swartsel**：倾向于**普惠繁荣**，但前提是必须进行大量系统性投入。如果维持当前现状——对教育培训投资不足、不关注 AI 如何改善工作品质、不积极扶持大众创业——局势完全可能滑向少数人受益的极端。 * **Hardi Gokani**：**视情况而定（It depends）**。这完全取决于员工个人的心态：面对 AI，你是否抱有乐观与好奇心，去主动学习如何借助技术让自己的工作更高效。 * **Preetha Sekharan**：**短期经历阵痛，长期迎来繁荣**。短期内的就业重塑必定伴随着摩擦与痛苦，长远来看才会释放普惠红利。 * **Ashwin Bharath（主持人总结）**：总体呈正向净效益。最显著的受益群体将是借助 AI 实现能力跃升、由蓝领岗位迈向白领阶层的人群；相反，当前享受高薪却从事易被替代工作的传统白领群体，可能会承受最剧烈的负面冲击。 ### [041:24] 研讨会总结与闭幕致谢 在研讨会的尾声，主持人向全场嘉宾及观众表达了诚挚的感谢，对各位专家在应对 AI 引发的职业替代问题、构建全民转型框架等核心议题上所分享的深刻见解与务实策略表示高度赞赏。 由于现场时间所限，主持人遗憾地宣布本次环节无法继续开放观众提问，并在全场掌声中正式结束了本场主题研讨。 - **研讨会主题**：应对 AI 驱动的职业替代：设计全民转型框架（Navigating AI-Driven Job Displacement: Designing Universal Transition Frameworks） - **主要参会嘉宾**： - Ashwin Bharath（Revature 联合创始人兼执行董事长） - Alex Swartsel（Jobs for the Future / JFF Labs 副总裁） - Kedar Deshpande（Zappos & Groupon 前 CEO） - Hardi Gokani（W.W. Grainger AI/ML 产品管理总监） - Preetha Sekharan（Unum 数字孵化器与应用 AI 副总裁） ## Notable quotes > [00:00](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=0s) 我们如何获得更好的工作，或者怎样才能不被 AI 抢走饭碗 — Ashwin Bharath > [04:36](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=276s) 我们必须主动将其设计成那样（we have to design it that way） — Alex Swartsel > [04:36](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=276s) 公共政策在技术变革面前往往是严重的滞后指标（Lagging indicator） — Alex Swartsel > [20:55](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1255s) 这个环节应当被自动化，这个环节应当被淘汰，它纯粹是在浪费时间，而这个环节则应当由 Agent（智能体）来进行增强。 — Ashwin Bharath > [22:05](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1325s) 深入劳动者的实际处境（meet people where they are） — Alex Swartsel > [29:42](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=1782s) 变革只要发生在别人身上，那就是极好的（Change is all good when it is happening to somebody else） — Preetha Sekharan > [36:48](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2208s) 如果一家公司自身连核心业务如何赚钱、如何创造收入都未厘清，AI 也无法凭空拯救它。 — Kedar Deshpande > [39:24](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2364s) 短期经历阵痛，长期迎来繁荣 — Preetha Sekharan > [39:24](https://www.youtube.com/watch?v=3UAhSU31Lk8&t=2364s) 最显著的受益群体将是借助 AI 实现能力跃升、由蓝领岗位迈向白领阶层的人群；相反，当前享受高薪却从事易被替代工作的传统白领群体，可能会承受最剧烈的负面冲击。 — Ashwin Bharath ## Mentioned Ashwin Bharath, Alex Swartsel, Kedar Deshpande, Hardi Gokani, Preetha Sekharan, Geoffrey Hinton, Taylor Stockton, Revature, Jobs for the Future (JFF), Zappos, Groupon, W.W. Grainger (固安捷), Unum, Microsoft (微软), Google, McKinsey (麦肯锡), American Federation of Teachers (AFT), AI Education Project (aiEDU), AI for Education, Perplexity, Google Chrome, WIOA (Workforce Innovation and Opportunity Act / 劳动力创新与机会法案), Stronger Workforce for America Act (美国更强劳动力法案) ## Video description Ashwin Bharath, Co-Founder and Executive Chairman, Revature Alex Swartsel, Associate Vice President, Jobs for the Future (JFF) Labs Kedar Deshpande, Former CEO, Zappos & Groupon Hardi Gokani, Director Product Management (AI/ ML), W.W. Grainger, Inc Preetha Sekharan, VP, Digital Incubator and Applied AI, Unum Abstracts This session will explore proactive policy solutions for managing workforce transitions amid AI disruption—highlighting strategies for reskilling, bolstering social safety nets, and building adaptable, inclusive labor markets through public-private collaboration. [Source](https://www.youtube.com/watch?v=3UAhSU31Lk8)

[0:00] Ashwin Bharath: 我们如何获得更好的工作，或者怎样才能不被 AI 抢走饭碗 [4:36] Alex Swartsel: 我们必须主动将其设计成那样（we have to design it that way） [4:36] Alex Swartsel: 公共政策在技术变革面前往往是严重的滞后指标（Lagging indicator） [20:55] Ashwin Bharath: 这个环节应当被自动化，这个环节应当被淘汰，它纯粹是在浪费时间，而这个环节则应当由 Agent（智能体）来进行增强。 [22:05] Alex Swartsel: 深入劳动者的实际处境（meet people where they are） [29:42] Preetha Sekharan: 变革只要发生在别人身上，那就是极好的（Change is all good when it is happening to somebody else） [36:48] Kedar Deshpande: 如果一家公司自身连核心业务如何赚钱、如何创造收入都未厘清，AI 也无法凭空拯救它。 [39:24] Preetha Sekharan: 短期经历阵痛，长期迎来繁荣 [39:24] Ashwin Bharath: 最显著的受益群体将是借助 AI 实现能力跃升、由蓝领岗位迈向白领阶层的人群；相反，当前享受高薪却从事易被替代工作的传统白领群体，可能会承受最剧烈的负面冲击。

[0:00] Ashwin Bharath: Navigating AI-Driven Job Displacement: Designing a Universal Transition Framework [1:46] Ashwin Bharath: Geoffrey Hinton promised he will take selfies with the first three questions [4:36] Alex Swartsel: we have to design it that way. [15:40] Kedar Deshpande: broad-spectrum transformation across the board. [17:20] Kedar Deshpande: if A happens, do B [34:13] Hardi Gokani: build your AI products around the people and not the other way around [36:48] Kedar Deshpande: show me [39:24] Hardi Gokani: It Depends

[0:00] Morning. [0:05] I know we're just starting to see the [0:06] crowd in, but I think it's time it's [0:08] good for us to start. Um, I have to read [0:10] this because this name is quite [0:12] interesting. Uh, I've struggled to [0:15] memorize. Uh, welcome to navigating uh [0:18] AIdriven job displacement. Um, designing [0:22] a universal transition framework. um [0:24] which is a fancy way of saying that how [0:27] can we get better jobs or not get [0:29] impacted by jobs. Um so that's what it [0:32] is but I think it is a much more [0:34] relevant topic because nobody's talking [0:36] about this much right I think it's very [0:38] interesting um even before I come in I [0:41] know um people are coming in but I would [0:43] like to get a quick pulse so that can [0:45] set the stage for it right how many of [0:48] you here thinks that AI is going to [0:50] create um [0:53] more jobs than the number of jobs it [0:56] will kill just a quick one [0:59] I like the optimism Others means no. [1:01] Okay, good. It's equally done. Um, [1:05] let me ask this, right? How many of you [1:07] think that your jobs will be uh [1:11] positively impacted by AI? [1:14] This is a better one. Okay. How many of [1:16] you think it will be negatively impacted [1:18] by AI? [1:20] Okay, small one. So, so I think it's [1:22] good. I like it because it's most of us [1:24] it's an optimistic. I think that's the [1:26] way it should be. So um we are all [1:29] talking about uh job killings now. So [1:32] roughly uh we are losing like anywhere [1:34] like 500 jobs per day just in tech jobs [1:36] alone globally uh because of AI. I think [1:39] it's a matter of time before we start [1:40] seeing job creations. So we have a [1:43] wonderful panel to listen from today and [1:46] I will just give a very very quick uh [1:48] intro about them and I will allow them [1:50] to introduce themselves. We have Alex [1:52] Swatso. Um I think um even I'm here to [1:57] learn a lot uh because you sit literally [1:59] in the intersection of policy education [2:01] and how that can be applied in [2:03] enterprises. So I would love to see [2:05] that. Um Hardi um and um obviously each [2:10] one is going to introduce but I think uh [2:12] very curious to know is Hardi is not [2:14] just creating very expensive powerpoints [2:16] in AI she's actually implementing it. So [2:18] I like to see that. Uh and Kather I [2:21] think his background is very [2:23] fascinating. I'm going to allow him to [2:24] introduce on that. Um have been on the [2:27] hot seat of massive companies billion [2:29] dollar revenue companies are plus. So [2:31] I'm very keen to see that. And finally [2:33] Pria um u Pria's addition is going to be [2:38] very valuable because some of the areas [2:40] her firm manages like claims processing [2:43] underwriting [2:44] uh some say that it's going to be [2:46] completely taken over by AI. Some say [2:48] that it's going to be extremely [2:49] enhanced. So I also would like to get [2:50] Prita's view with that. Let's just get [2:53] your introduction. We'll start from [2:54] Kedar. [2:56] >> Hi. Uh I'm Kedar Desh Pande. I was [2:59] former CEO of Zapos and Groupon and u I [3:05] get to work with AI almost every day [3:07] because these days I'm helping a lot of [3:09] startups. Uh and so the pace of [3:12] innovation is just incredible and great [3:15] to discuss be here. [3:18] Hi everyone, I'm Alex Schwarzel. I'm an [3:20] associate vice president at Jobs for the [3:22] Future. JFF is a national nonprofit that [3:25] focuses on working with education and [3:27] workforce systems all around the country [3:30] to help make it easier for people to get [3:32] access to really highquality education [3:34] and training and ultimately a better and [3:36] economically mobile job. [3:39] >> Hi everyone, I'm Hardi Gogani. I'm a [3:41] director of product management at [3:44] Granger. Uh I lead product strategy and [3:47] vision specifically for a IML powered [3:49] products and along with that I also led [3:52] enterprisewide AI adoption initiatives [3:55] with a focus on measurement measurable [3:58] outcomes. [4:00] Hello everyone, Priita Shakran. I lead [4:02] uh the digital innovation and applied AI [4:05] team at Unun. Unum is a life and [4:08] disability insurance provider and uh me [4:11] and my team drive uh building scaling [4:14] deploying uh AI initiatives across the [4:17] organization [4:19] >> and um audience when if you have a [4:22] questions you can raise your hands we [4:23] have a probably a small section by the [4:25] end if you have in between do this so [4:27] Jeffrey Hinton promised he will take [4:29] selfie with the first three questions so [4:31] just joking [4:33] because people turn their head so either [4:34] way first let's start with Um see we all [4:37] know that AI can be pro- capitalism [4:39] that's why billions and billions of [4:41] dollars are getting in. Um we also know [4:43] that it could be pro- mankind with all [4:46] the new discoveries especially in the [4:47] healthcare side but the question is can [4:50] it be pro workers right I think that's [4:52] the big debate that we are having right [4:54] so Alex I will start with you um JFF I [4:58] think your study and research is saying [5:01] this displacement caused by AI is going [5:04] to be completely different than the past [5:06] one like to hear more about that [5:08] >> sure I mean I think the short answer to [5:10] that question can AI be proorker is that [5:12] of course it can, but we have to design [5:14] it that way. Um, Jeffrey Hinton this [5:16] morning talked about the idea of [5:18] maternal AI, which is a construct I [5:21] hadn't heard before. And I think we like [5:23] to think of it as how can AI make us all [5:25] better off. uh what what do we need to [5:27] do in how we're designing jobs and [5:29] people's pathways to livelihoods? How [5:31] we're designing learning and training [5:33] opportunities, the technology itself, [5:36] and the education and workforce systems [5:38] that we all move through, including our [5:39] employers, so that we can ensure that AI [5:43] contributes to economic mobility and [5:45] better jobs rather than the other way [5:47] around. And I do think we're in a very [5:49] different moment than we've seen in the [5:50] past for a whole variety of reasons. the [5:52] speed of adoption of AI, its rapid [5:55] diffusion in both a B2B and a B TOC [5:58] context, the fact that it is addressing [6:01] so-called higher skilled or white collar [6:03] jobs in ways that past waves of [6:05] automation haven't. And with all of [6:07] this, we're this forces are colliding [6:11] with systems that aren't built with that [6:13] level of agility and resilience. And [6:15] that's a massive challenge for all of [6:17] us, but for particularly for people all [6:20] over this country who have traditionally [6:21] experienced barriers to economic [6:23] advancement, whether those are people [6:24] without a four-year college degree, [6:26] people with criminal records, often [6:28] people of color, often women in many [6:29] occupations. And so we have to design [6:31] our systems and the technology itself to [6:35] make sure that opportunity is available [6:37] for everybody that everybody has the [6:38] skills that they need the jobs are [6:40] designed in such a way that center [6:42] uniquely human capabilities and that do [6:45] so while generating business value which [6:47] is a huge challenge but I think we can [6:48] do it. [6:49] >> Let me ask this. So um who should design [6:51] that system? Is it like like [6:53] institutions like JFF or enterprises [6:56] government? Who should design that? [6:57] Because it cannot be all three. It has [6:58] to be somebody taking the lead. [7:00] >> I mean, I think that's the exact [7:02] challenge, right? Because systems are [7:03] are systems for a reason. They are [7:05] designed by all of us. And I it can I [7:07] think be very easy to look only to [7:09] public policy as the ultimate solve of [7:12] all of these things. Well, if we only [7:13] had a law, then we could fix it. And [7:16] while on some hands that can be true, on [7:18] the other hands, policy can often be a [7:20] lagging indicator, as we're seeing in [7:21] America very, very much today. And so [7:23] the more that institutions that have big [7:27] levers that they can pull, whether they [7:29] are major employers, whether they are [7:31] major college or workforce systems, [7:33] community-led economic development [7:35] organizations that can work even in a [7:37] microcosm to design new pathways in a [7:41] place-based way to experiment with [7:42] something. I mean I ju it's so tempting [7:45] to look for the silver bullet and I [7:47] don't think there is one especially at [7:48] this moment when things are changing so [7:51] fast that we don't have good data yet. [7:54] So if the one system architect is [7:56] waiting for the one magisterial set of [7:58] data we're going to be waiting forever [8:00] and the world will pass us by. So we [8:02] have I think we have to think [8:03] differently about how we test and [8:05] experiment and inculcate innovative [8:08] approaches within all of these systems [8:11] where that can be a huge challenge. [8:13] What's that going to take? And then over [8:16] time hopefully pretty quickly that can [8:18] start to ladder up to bigger changes [8:19] that we can enact through policy or that [8:22] can be more widespread. [8:24] >> Thank you Alex. And um I will ask this [8:27] to you the sec next question. um [8:31] especially with Granger are you seeing [8:34] practically jobs getting created [8:36] disappearing or transform transforming I [8:39] would like to know more on the [8:40] transforming piece because Revit the [8:42] company I founded we are the largest [8:43] tech training company primarily helping [8:45] to address skill gap sadly even though [8:47] we make good money we haven't solved [8:49] that skill gap solving itself is a major [8:51] problem I would like to hear on ground [8:54] are you seeing transformation happening [8:57] >> that's a question I reflect upon on a [8:59] lot. Um the reality as I see it on the [9:02] ground is that jobs are transforming. [9:04] I'll give an example of the work we've [9:06] done at Gringanger. Um Granger operates [9:09] in a very complex industrial [9:11] environment. We rely a lot on our field [9:14] team members to understand our customers [9:17] maintenance, repair and operations [9:19] needs. What that is is um safety [9:22] equipments, plumbing supplies, [9:24] everything that keeps a machine up and [9:26] running. And so our strategy for [9:29] implementing AI at Granger is AI [9:32] augmentation. [9:33] We work very closely with our team [9:35] members and try to identify [9:37] opportunities where AI can elevate what [9:39] they do. So for example, when we acquire [9:42] new customers, we require our sales team [9:45] members to go on field and actually [9:47] understand what our customers are using [9:49] today. [9:50] The pre-AII era the way they used to do [9:53] this is they would go manually collect [9:56] information in an Excel document [9:58] spending hours crunching data cuz the [10:02] way the products are set up are in a [10:04] very complex warehouse setting. So it's [10:06] not straightforward that you just go and [10:08] grab a list. What we did is we [10:10] implemented an AI powered mobile [10:13] application that specifically uses [10:15] computer vision that assist and augments [10:18] the work our team members do on field [10:20] for collecting that customer inventory [10:22] information. [10:24] This of course every the data collection [10:26] that used to take hours now takes only [10:28] minutes. So yes, it was a productivity [10:31] boost for us. However, what is the [10:34] interesting part to note is it was a [10:36] shift in the way they did their jobs. So [10:38] now when they're on field, they're not [10:40] just spending time staring at Excel [10:42] documents. They're actually doing work [10:44] that matters, connecting and listening [10:46] to our customers and solving their [10:48] problems. However, [10:51] um a conversation about AI and job [10:53] displacement would be incomplete if we [10:55] don't address the elephant in the room, [10:57] the recent wave of layoffs that we've [11:00] seen in the tech environment. [11:02] Let's take Microsoft for example. [11:04] Microsoft has been in the news since [11:06] 2023 for going through multiple rounds [11:09] of restructuring and layoffs so that [11:11] they can pump more investment into AI [11:14] infrastructure. [11:15] However, the interesting part to note [11:17] here is that if you compare their [11:19] workforce from 2023 to 2025, there's [11:22] actually a 3% increment in their [11:25] workforce. What this tells me is that [11:28] yes, jobs are getting reshaped and [11:31] reskilled, but there are more jobs [11:33] overall in the market versus what we you [11:37] are used to seeing maybe 3, four, 5 [11:39] years ago. So ultimately, it's about [11:41] figuring out how AI and humans can work [11:44] together to drive higher and more [11:46] efficient outcomes. [11:48] >> Thank you for that. And yes, [11:52] >> we help small and mediumsiz businesses [11:54] to make a customized um platform uh with [11:58] um augmented AI which means that we [12:01] designed AI to help you know to work [12:02] with the um humans and our goal is you [12:06] know helping them throughout the [12:07] restructuring to kind of like not [12:10] sending people home as much as we can [12:12] but few questions that we have came from [12:15] and I I'll ask just two of them that [12:17] I've been you know looking forward [12:19] answer and I couldn't find it anymore um [12:21] you know um in the past year. So my [12:24] first question goes to Alex. Schools are [12:28] behind everyone and you're working with [12:31] the educators. I personally as a person [12:33] who has an interdisciplinary skills has [12:36] been looking for opportunity to you know [12:38] teach his students about okay how you [12:40] can integrate AI into a business. I'm [12:43] not a tech person, but I have interd [12:44] discipline skills, but everyone is like, [12:48] we don't know what it is yet. We just [12:49] know the storm is coming to grab us. So, [12:52] I'm wondering if um behind the scene or [12:56] if you're informed about any plan or [12:58] anything coming up to the school and ch [13:01] all of these curriculum to me is [13:03] outdated. They have to change it. They [13:05] need to train people in a different way. [13:08] Um they need to learn how to, you know, [13:10] work with AI. the knowledge that we have [13:12] is already outdated. That's my first [13:15] question. Second question, I don't know [13:16] who can answer that is [13:20] any policy going on who is going to [13:22] protect the the rights of uh employees [13:26] large and you know large organization [13:28] like Google um or Microsoft yeah they [13:31] work with the McKenzie and Mckenzie [13:33] always is pretty harsh to send people [13:35] home. Small and mediumsiz businesses are [13:38] at really at risk. So we at least try to [13:41] help them but the governmental sources [13:43] are very short on that you need you know [13:46] the resources and also that's the [13:48] question for everyone who what are the [13:50] and you know there's no policy out there [13:52] all we know that states cannot you know [13:55] come up with their own legislations but [13:56] okay so who is going to protect and what [13:59] would be the rights of employee because [14:01] everyone advised that okay go with a [14:03] good hygiene of you know laws and [14:04] regulation until you know people put out [14:06] the the layout okay how does it look [14:09] like but I'm wondering if anyone knows [14:11] anything about um [14:12] >> so Alex you can take the first one [14:14] quickly the second one I think we will [14:16] be covering that um as part of our [14:18] subsequent question so Alex you want to [14:20] take the yeah [14:21] >> sure I think AI literacy there's a lot [14:23] of progress um so for example the [14:25] administration's AI action plan which [14:28] came out a couple of weeks ago and as [14:29] well as the previous executive order [14:31] included a huge push that's focused on K [14:34] through2 education the American [14:36] Federation of Teachers has a major [14:38] partnership with Microsoft oft for [14:39] example that's around upscaling [14:41] teachers. There are a lot of great [14:42] organizations like the AI education [14:44] project, AIEDU, AI for education, others [14:47] that are really working hard to try to [14:50] get AI literacy into the K through2 [14:52] setting. So we're nowhere near where we [14:55] need to be, but there's a lot of good [14:58] progress underway. That's both what the [14:59] curriculum should be, what AI literacy [15:01] looks like at that stage. How do we do [15:03] that in a way that's safe and [15:04] thoughtful? Progress, not enough, but [15:08] progress. The thing that I'm curious [15:10] about is something else that's embedded [15:12] in what you said, which is what about [15:13] all of the other skills? So, how are we [15:15] train just as far was talking about um [15:18] workers are going to need to communicate [15:19] in completely different ways, need to [15:21] empathize and understand interpersonal [15:24] skills in very very different ways [15:26] because it is about communicating with [15:28] humans, but also about how we [15:29] communicate with machines. Education has [15:31] not figured that out yet. that I think [15:33] we need to really focus on and that's [15:35] going to take the collective effort of [15:37] systems to help us figure out what that [15:39] looks like. [15:40] >> Thank you for that Alex and um so this [15:43] is for you. I'm I personally I'm very [15:46] much interested in this question as [15:48] well. So both the companies that you ran [15:50] Zappos and Groupon are uh huge [15:54] innovative models as a result of digital [15:57] transformation [15:58] um and you extremely succeeded in that. [16:00] How is AI different from digital [16:02] transformation? [16:05] >> Uh digital transformation was [16:09] more about customers and customer front [16:12] and so I will just answer the specific [16:15] part of it rather than taking anything [16:17] uh else into consideration. So digital [16:19] transformation you're buying something [16:21] into store you're doing something inside [16:23] the your uh systems and whatnot. AI on [16:28] the other hand is transformation across [16:31] the board in my opinion. And so when we [16:34] say across the board, what does that [16:35] mean? That means any repetitive part of [16:38] the your job first and foremost can go [16:41] away because you don't have to do that [16:44] again and again. So this example she was [16:46] talking about where you have to take the [16:48] inventory down and manually collect all [16:50] that information doesn't need to be [16:52] there. But this is not applicable to one [16:54] set of people. It's applicable to [16:56] accountants. It's applicable to customer [16:58] service. So broad spectrum that's how it [17:01] is different in my opinion. Now the [17:04] second question is adoption speed and I [17:07] think that's where the whole discussion [17:10] is about and [17:12] what I feel like is that the [17:14] displacement on these particular fronts [17:16] where digital adoption didn't uh it [17:20] created new jobs. It took away some of [17:22] the old processes and some specific [17:24] parts of it. Here if your value as an [17:28] employee is repetitative process [17:32] then there is a real good chance that [17:34] you are doing the same thing over and [17:36] over. You come in every day. You look at [17:38] that and previously you no algorithm [17:42] could do it because it was only A [17:44] happens do B or some sort of way then [17:47] you would not be displaced. But now it [17:50] is intellectual layer. So your value ad [17:53] as an employee as an organization needs [17:56] to be very clear and that's what is [18:00] going to drive [18:01] um much more again uh digital [18:05] transformation versus AI transformation. [18:07] AI transformation will be much more u [18:10] invasive in that particular sense. [18:12] >> Thank you K. And um um so I have an [18:16] interesting question. I will just wait [18:18] for this next section for you. Um so [18:20] priita let me ask this right so a lot of [18:22] people are confused between automation [18:24] and agents right so um the way I see is [18:28] automation it's context aware autom not [18:30] automation agents are context where [18:32] automation need not be especially the [18:34] field you are in at unim lot of these [18:36] things are repeated process do you think [18:38] your AI implementation are just [18:41] automation plus+ or are you genuinely [18:43] seeing a way for agents to play a role [18:44] there [18:46] >> I think for us u like where we are at [18:49] the moment And I do not believe the you [18:51] know the agent technology and everything [18:53] is at at a level where you can deploy uh [18:57] more generalized agents. So, so where we [18:59] are, we are looking at different uh you [19:03] know different parts of our value chain. [19:04] Looking at what are the jobs that people [19:06] perform, what are the tasks there are [19:08] some tasks that are a lot more um uh [19:11] conducive for uh having an AI agents [19:14] tasks that are repetitive. I think Kedar [19:16] mentioned that tasks that u are are are [19:20] more routine more much simpler uh and [19:23] those are definitely things that we will [19:25] be deploying AI agents for. But then [19:27] there are I mean if you think about uh [19:29] the the business that we are in uh [19:32] people come to us at the time of need [19:34] they're going through a personal pain [19:35] and what you need on the other side of [19:38] that of that phone call is someone who [19:41] is more empathetic who is able to [19:42] connect with that individual at an [19:44] individual level right and and that is [19:46] not something we can use agents for. So [19:49] what this I mean there is definitely a [19:52] lot of power. I mean insurance is [19:54] datari. Uh insurance is definitely a [19:56] space where um we do believe that AI [19:59] will play a big a very big role. Uh but [20:02] it is going to be human machine [20:04] collaboration. Uh the roles and and the [20:07] tasks that people perform will change [20:09] and I and I agree with you that those [20:11] many many roles will get transformed as [20:14] part of this. Um uh but it is not going [20:16] to be uh you know automation taking over [20:19] the world uh is is what we think [20:21] >> absolutely um I also believe uh priita [20:24] what you're telling is how the whole [20:25] industry is facing um at the same time I [20:28] think change is going to come fast [20:30] yesterday I saw this company it's built [20:32] by a Harvard professor so their software [20:35] is installed in everybody's machine [20:38] every employes machine every keystroke [20:40] is monitored everything using that [20:43] because there will Understanding the [20:45] process of how human beings, your [20:46] workforce interacts with the digital [20:48] world is key. The best way to track it [20:50] is keyboards and how you use the mouse [20:52] and the your touchcreen. And with that [20:55] they are creating process context like [20:58] nothing. I was really stunned to see how [21:00] accurate they are. And with that context [21:03] they're telling these should be [21:05] automated. This should be eliminated. [21:06] It's wasting your time and this should [21:07] be augmented by agent. Simple. And I was [21:10] shocked to see the results. So when I [21:12] saw that I was genuinely happy that I am [21:15] approaching my 50. So I'm not like at [21:16] the peak of my career. So but uh it's a [21:20] little scary to do that. And now let's [21:21] see get into the solution part right. Um [21:25] and like I will just start with you. [21:26] It's easier said than done. Um [21:29] especially relying on government. We [21:30] talked about governance and everything. [21:32] I genuinely believe there has to be a [21:34] proper governance but we did not even [21:35] get the social media governance right [21:38] let alone AI. And especially when we are [21:40] having a government saying there cannot [21:41] be any governance now right um can [21:45] vehicles like WA which is workforce [21:49] innovation opportunity tax I think I'm [21:50] getting the name right it's uh run by uh [21:53] department of labor division [21:56] can those initiatives be upgraded for [21:58] these solution especially for upscaling [22:00] because that's designed primarily for [22:01] this do you think we should look into [22:03] such vehicles or we should look for [22:04] something else [22:05] >> I mean of course we should right wea has [22:07] which is the law that you referred to [22:09] hasn't been reauthorized since 2014. Um, [22:12] one or two things have happened since [22:14] then. And there is great legislation. [22:16] Um, the stronger, this one I had to [22:18] write down, the stronger workforce uh [22:19] for America act, which was passed at the [22:21] end of last year, hasn't been enacted [22:22] yet. Um, that has a lot of really great [22:25] changes to making training that it [22:27] supports more focused on skill [22:28] development, encouraging employer [22:30] engagement. Um, a lot of really, really [22:33] good changes. It needs to actually [22:35] become law and that's one piece of the [22:37] puzzle. Um, there are some other great [22:39] proposals in the administration's action [22:41] plan. For example, like the creation of [22:43] a research hub. Taylor Stockton from the [22:45] Department of Labor mentioned this [22:46] earlier today. That would essentially [22:49] use public structures to to create [22:52] public goods that would be much too [22:55] costly or inefficient for any [22:57] constellation of private actors to do on [22:58] their own. So the the collection and [23:01] maintenance and curation of much much [23:03] better real-time labor market data for [23:06] example which is just a really profound [23:08] need right now. Um and you know there is [23:12] still more that we can do that also is [23:14] about how we use the levers of policy to [23:16] invest in workers directly or to [23:18] encourage or incentivize behavior. for [23:20] example, what would it look like to [23:22] expand tax credits for employers to [23:24] offer reskilling training, whether [23:25] that's to incumbent workers or to to [23:27] other workers that are in their orbit. [23:29] Um, so there is a lot that we can do on [23:31] a policy level. Um, and I think we've [23:35] been talking about this a bit. How then [23:37] do we prepare our broader systems and [23:39] prepare workers and learners themselves [23:42] to be able to step into those [23:43] opportunities? So, there is a lot that [23:46] policymakers and governments can do at [23:48] both the federal and state level. But [23:50] there is also a lot and we've heard [23:51] about it from Hardy and Priita that [23:52] employers can do whether that is taking [23:55] a really hard look at the jobs that [23:57] they're creating to say are these [23:59] quality jobs do they pay people well do [24:01] they offer benefits do they offer [24:03] flexibility and opportunities for [24:05] advancement for example and how if they [24:07] don't how can we use AI integration to [24:10] make them quality jobs for example um [24:13] how can we if we're a training provider [24:15] or a workforce organization how do we [24:18] meet people where They are for example [24:20] to provide training at off hours so that [24:23] somebody can work and then still go to [24:25] school and be upskilled in some ways or [24:27] provide things like child care, provide [24:29] laptops for people that don't have them. [24:31] There is a lot about how do we meet [24:34] people where they are and understand [24:36] what are the real barriers that often [24:39] exist to people being able to develop [24:41] their skills and access opportunities. [24:43] You know, those barriers often are are [24:45] things that you don't necessarily [24:47] expect, but they are very very real [24:49] because if you know, a single mom has to [24:52] work to support her family and can't [24:54] take time away to go to a short-term [24:57] boot camp, much less get a four-year [24:59] college degree, then that's a huge [25:01] barrier for her in in seeking [25:03] opportunity. And we have a constellation [25:05] of actors. Sure, there, you know, policy [25:08] can help, but her future employer can [25:11] help, training systems can help create a [25:14] pathway for her that allows her to live [25:16] into her potential and to build a career [25:19] that allows her to support herself and [25:21] her family. [25:21] >> Thank you, Alex. And you're right. And [25:23] one more thing which I'm seeing pockets [25:25] of success is forming like learning [25:27] group or study group among people [25:29] because the biggest challenge workers [25:31] are facing is the unknown is unknown. So [25:33] collective knowledge is helpful and and [25:35] Kend let me uh ask this question to you. [25:38] Um with your background I assume you are [25:40] a free market capitalist. Uh I am one. [25:43] Um and again let's be honest um you [25:48] don't do anything unless there's a [25:49] benefit especially the capitalism right [25:51] that's the way it works. What will [25:53] motivate you? Let's say that you are [25:54] sitting in an organization running the [25:56] company with lot of workforce. You know [25:58] the transition is imminent. What will [26:00] motivate you to invest money to make the [26:03] transition to AI? Because ultimately [26:05] without money nothing is possible and [26:07] getting government money is not easy [26:08] either. So what will motivate you [26:10] actually? [26:12] >> Yeah. So um just give you an anecdote um [26:15] I was talking to one of my other former [26:18] CEO colleague and he was telling me they [26:21] run also [26:23] uh very big conferences uh between CEOs [26:26] and one of the things they asked in that [26:28] particular conference like at least 100 [26:31] attendees is that how many of you are [26:33] using AI within your organization and [26:35] almost 99 people raised their hand [26:38] except one and then the then the next [26:41] question asked is how many of you [26:43] actually increase your IIDA or margins [26:47] or total profit pool using AI and [26:50] probably two went up and so that's the [26:53] challenge we have as overall um [26:56] ecosystem and so I think that direct [26:58] visibility is not there and there are a [27:00] lot of organizations which are [27:01] struggling with this particular thing [27:02] which say I can eliminate but I can add [27:05] their AI and I can help employees I can [27:09] increase the efficiency but there is not [27:11] direct direct correlation in terms of my [27:14] profit pool increasing and I think there [27:16] are two different so if you step back [27:18] there are two different uh challenges [27:20] organizations haven't solved so AI is [27:22] just like any other wave in my opinion [27:25] because [27:26] waves transform the organization but [27:28] they also expose the deficiencies inside [27:31] the organization and so if your core is [27:34] not solid then it's time to reflect on [27:37] that core so when I say that for example [27:39] if your organization Culture is not [27:42] about transformation. It's not about [27:44] customer first. It's all about making [27:47] the tactical decisions for that specific [27:49] quarter and that's what you have been [27:51] showing with your axe. Not preaching but [27:53] showing with your axe then it's very [27:55] difficult. But if your organization is [27:57] all about hey we are customer first and [28:00] whatever we are going to do AI actually [28:03] empowers and this is the example I'm [28:04] going to get into is where at Zapos we [28:06] always cared about culture with the [28:10] customer and employee happiness first [28:13] empowerment first and so I had an [28:15] employee and this was 5 years back she [28:18] was looking at customer service calls [28:20] and taking the repetitive customer calls [28:23] and she was like I'm I'm very sick of it [28:26] to tell the same problem over and over. [28:29] So she didn't just escalate it. She [28:31] actually put the spreadsheet learn Excel [28:34] micros and then presented to the all [28:38] other groups to say look this is what we [28:40] can say for you as well as for us [28:43] just initiativedriven. And so my point [28:46] here is that now somebody she might it [28:49] might have took her one month to learn [28:51] that particular skill. Now with AI she [28:54] can do that probably in a day if you ask [28:56] it. So this is the kind of thing which [28:59] is if your organization is clear about [29:02] its culture it's about what the [29:04] priorities are then you can see the ROI [29:06] really fast and that's what is the core [29:10] to your organization. So yes, there is a [29:12] lot more ROI that you can see through [29:14] but as long as you understand your core [29:17] >> that's that's a very good see what [29:19] you're saying quite recently I say a few [29:21] weeks ago a CEO of a large company said [29:24] the same thing the best way to get the [29:26] transition is empowering your employees [29:28] that they are in charge of this [29:30] transition and also taking a change [29:32] management approach rather than telling [29:33] that oh by turn off a switch you have to [29:35] be an AI no just give them a structured [29:37] path for a change management approach so [29:39] which is very valid And um a priita this [29:42] question is for you right so when from [29:44] revature when we go to firms asking for [29:46] hey train my people in AI the initial [29:49] conversation is like rah right like oh [29:51] it's a CEO direct you whatever it takes [29:53] 6 months training whatever it is great [29:55] we give them a 10 weeks curriculum come [29:57] back ah can you make it to 6 weeks then [29:59] we say yes then comes back oh the [30:01] project manager said they're busy with [30:02] their work it cannot be done it has to [30:04] be like 3 weeks in reality ultimately if [30:06] I get 2 hours per day that's a challenge [30:09] and then then they say can the workers [30:11] ask within their non-paid hours then the [30:14] HR comes and says that oh wo we won't be [30:16] obligated to pay too many confusions [30:18] right so how are you allocating your [30:20] employees training because they [30:22] everybody has to make a living how are [30:24] we allocating that is it within the [30:26] company's paid hours like partial [30:28] part-time how do you how what's your [30:30] approach on it [30:31] >> so I I personally believe change is all [30:34] good when it is happening to somebody [30:35] else uh and so it it pretty much starts [30:38] with right when like how do you be more [30:41] inclusive in having them shape what is [30:44] actually going to happen to the rest of [30:46] their organization. So including them in [30:48] shaping uh these AI products I think [30:50] that's a very very critical first step [30:53] uh as we deploy this uh you know taking [30:56] it through phases. So we we we are very [30:58] structured about having a model office [31:00] where you actually test it out with few [31:02] people see you know what needs to change [31:05] within within that select group of [31:07] individuals before you roll that out. Uh [31:09] so change management in most of the [31:11] initiatives that we have been deploying [31:13] in our organization we think of change [31:15] management right from the beginning uh [31:17] it is pretty much built as part of the [31:20] initiative through the various stages uh [31:22] and then on the back side right so once [31:26] uh the initiative is completed the value [31:28] realization as well right so we're [31:31] intentional about you know like if you [31:34] have if you realize 10 million of $10 [31:36] million of opex savings uh Not all of [31:39] that goes to the bottom line. There's [31:41] some that has flowed back into [31:43] reskilling individuals and and [31:45] upskilling them. So that's that's kind [31:47] of you know I think Alex you mentioned [31:49] you have to be intentional about these [31:50] things they just don't happen. Thank you [31:53] Priita and I have this question for you [31:56] because this is based on what you said [31:57] before which is you are taking an [31:59] augmentation approach. Um to me it's the [32:02] right approach. Um because that's going [32:05] to be a massive value especially in your [32:07] ability to transition. Um the problem I [32:11] at least I can foresee that I hope I'm [32:13] wrong is uh the younger kids I have two [32:15] boys 17 and 13. My elder one 17 like so [32:19] much into AI. I don't think so he will [32:22] see the human elements of job killing [32:24] and everything. He's always talking [32:25] about how futuristic it it is and [32:28] whether we want it or not, they are the [32:30] one going to set the narration of AI in [32:32] the future. Um I just read the news in [32:34] CNBC that perplexity which is valued at [32:37] 151 18 billion has offered to take over [32:39] Google Chrome for 35 billion. Right? So [32:41] I thought it'll be the other way around, [32:43] right? That's what these young kids do [32:44] and they are going to rule it. How do [32:46] you make sure that from our generation [32:49] to the next generation, the human in the [32:51] loop that augmentation mode can be [32:52] transitioned? [32:54] For me, that principle is very simple. [32:57] Your users and your employees cannot be [32:59] an afterthought when you're designing a [33:01] product. They have to be at the [33:03] forefront of anything that you're [33:05] designing. So, there's a three um phased [33:09] loop approach that I like to use quite a [33:11] bit where phase one is code design. You [33:14] invite your users of whatever you're [33:16] building into the design phase. Embed [33:18] them not just for feedback but for [33:21] decision- making. For example, at one of [33:23] our product portfolios, what we did was [33:25] we invited our users and put them into a [33:28] code design group where they were making [33:31] decisions on how they would interact and [33:33] review AI generated outputs. Then we go [33:37] to phase two. Um phase two is [33:40] progressive automation. What that means [33:43] is we don't turn a switch on and off [33:45] from human to AI. Initial phases of [33:49] deployment would look like where our [33:50] team members are spending time reviewing [33:53] AI outputs. So they always have the [33:55] liberty to override a decision that AI [33:58] made. However, as the AI model gets more [34:01] and more data, it's going to get more [34:03] accurate. So what will happen is our [34:06] team members will do less and less of [34:08] those repetitive tasks and would be [34:10] spending more time on high value [34:13] initiatives that matter to the company [34:15] topline metrics. And then phase three is [34:18] skill uplifting. Skill uplifting as Alex [34:21] mentioned cannot be an afterthought. [34:24] Companies and employees should start [34:26] thinking about it when you're building a [34:29] transition. And so what we do is we [34:31] educate our team members and users on [34:34] how to interact with AI, how to [34:36] supervise AI and always have an [34:38] escalation path back to the human to [34:40] make a decision on using their judgment [34:44] and critical thinking. So all this to [34:46] say that do not build [34:50] build your AI products around the people [34:52] and not the other way around. [34:54] So if I have to summarize you're saying [34:56] it is augmented workforce with the [34:58] capability of augmented decision-m right [35:00] I think that's what it is good very good [35:02] so now just let's get into the economics [35:04] of it right and ultimately we need to [35:06] make sure that money matters so pa I'll [35:08] start with you so firms like ours tech [35:11] services we go to your CEOs and say that [35:14] hey by AI we can save you so and so [35:17] things so and so time all those things [35:18] we do it and um sadly we can be [35:21] convincing right so How if your um CEO [35:25] let's say a kind-hearted person comes to [35:27] you and asks hey what will I tell the [35:29] board um that I don't need to use AI for [35:34] saving cost or job killing but rather [35:37] this is what if I do a proper transition [35:40] from the current way of working to AI [35:42] way of working this is how you would [35:43] benefit what advice would you give your [35:45] CEO [35:47] >> I think looking at any any AI initiative [35:50] and saying I mean it's just driving [35:52] efficiency you're probably being [35:54] short-sighted. What we have found is [35:56] like invariably most AI initiatives [35:59] would kind of not just improve like [36:02] efficiency, it will improve [36:03] effectiveness, the quality of your work, [36:05] the turnaround times for uh you know for [36:08] for that particular transaction. So it [36:09] is pretty much uh it's a trifecta of a [36:13] lot of other value drivers other than [36:15] efficiency. Uh I I would also say that [36:18] like in organizations like ours uh [36:20] there's a lot of institutional [36:22] knowledge, right? So bringing in like if [36:24] if you find that the skills are not [36:27] appropriate, you're probably better off [36:30] upskilling or cross-skilling you know [36:32] the individuals that you already have [36:34] because the ramp to proficiency is very [36:36] long. It takes a long time for people to [36:38] kind of bring in new people and and [36:40] train them on on things that they should [36:42] know about the organization because [36:44] there's a lot of institutional knowledge [36:46] uh that that comes with it. [36:48] >> I like that institutional knowledge [36:50] should be key. Um K let me ask this [36:52] question. So the last 10 years my life [36:54] has been uh like managing companies for [36:58] two private equities. So they literally [37:00] make all of us sign our number one [37:03] fiduciary duty as a CEO is to give [37:06] shareholders in returns. Simple, right? [37:08] And they keep reminding us every month, [37:10] every quarter, right? Um how do you [37:13] balance between [37:16] uh AI transition, workforce transition [37:18] return? Yes, AI transition can give you [37:20] returns but how can you balance between [37:22] AI workforce transition along with [37:23] shareholder return actually. [37:27] >> Yeah. So I think AI um can be actually [37:32] multiplying force and that's why I was [37:34] saying that there is a direct has to be [37:36] correlation between um not just a there [37:40] is some savings here and there like this [37:45] it reminds me of one of those particular [37:47] things where like back in the days when [37:49] the digital transformation happened like [37:51] a lot of websites they used to come back [37:52] and say I did this particular experiment [37:55] and my conversion went up by 10% basis [37:57] points and then you look at the sales [37:59] and sales are same. [38:02] So your value is like oh I did something [38:05] it was great everybody said right but [38:08] then it didn't matter to shareholders in [38:11] general and so this has to be show [38:14] moment and AI has the capability to show [38:16] that directly attribution and so if it [38:18] is all about savings or it is all about [38:22] customer uh happiness that can transform [38:24] into direct revenue or profit pool and I [38:27] think that has to be there and so it [38:29] needs to come with a pragmatic lens and [38:31] so as other participants are saying [38:34] mostly if if it is just like hey trust [38:38] me this will happen that's not going to [38:40] happen you have to have the direct [38:42] visibility into that from a shareholder [38:44] perspective and I think AI makes it much [38:47] more easier to get that there uh and [38:51] it's faster as well so for example if [38:54] you did whatever I was talking about [38:55] website redesign and whatnot it will [38:58] take months and months and months and [38:59] then you are looking at [39:01] AI you can do that within days and you [39:03] can cease that sort of transformation. [39:05] So both acceleration as well as adoption [39:08] is important but again I will be the [39:11] broken tune in there is like if your [39:14] cores are not clear how do you make more [39:16] money how do you generate more revenue [39:18] if that is not clear to your [39:19] organization then AI is not going to [39:22] help you there so we just have 3 minutes [39:24] left so I'm just going to um ask last [39:27] question please respond each one of them [39:30] give a quick respond uh like uh and [39:32] we'll try to use that 3 minutes the [39:34] question is let's say that 3 years, 5 [39:35] years down the line, we are having the [39:37] same panel, same team. [39:39] Will AI create prosperity for all or [39:43] prosperity for few? And let's be honest [39:45] for it. I will give my take too. Let's [39:47] just [39:49] uh I think AI will create prosperity for [39:52] all and not for few uh overall. But I [39:56] think it will create uh gaps for [40:00] somebody who is not ready to change or [40:02] is not clear um about how they add the [40:06] value. [40:09] >> I think all but there's a lot that we [40:11] have to do to get there um that ladder [40:15] into the skill development that we've [40:17] been talking about. If I think if we [40:19] proceed on the current course where [40:22] we're not investing enough yet in [40:23] education, we're not asking questions [40:26] about how AI is making jobs better, [40:28] we're not helping people become [40:29] entrepreneurs where they have the [40:31] capability to do that and we're seeing [40:32] early interest, then I think it goes the [40:35] other way. [40:37] I've been a consultant. So I'll say it [40:38] depends. Uh it depends because it [40:40] depends on your mindset, how optimistic [40:43] and curious you are to learn about how [40:46] AI can make your life easy in the job [40:48] that you do. So it depends on how you as [40:51] an employee are going to approach it. [40:55] >> I think it'll be painful in the short [40:57] term. There will be prosperity but [40:59] that's probably a much longer term. [41:01] >> Got it. I'm same belief. uh but uh with [41:05] one caveat I think let's call it as like [41:08] uh like to be white collar jobs people [41:10] who are elevating themselves from blue [41:11] collar they will benefit the most but I [41:15] believe the current high paid white [41:17] collar job they will get negatively [41:18] impacted the most so that's my take but [41:20] in net I the number of people who will [41:22] benefit in AI will be good with that I [41:24] really want to thank you all of you for [41:26] this and um we don't have time if I [41:28] question no we don't have thank you so [41:30] much for that okay thank [41:32] [Applause]