O'Keeffe Warns AI Retraining Risks Sorting Out Vulnerable Workers

Panel 6 - 1: Personalized, Powerful, and Equitable: The Role of AI in Workforce Retraining and — Ai4

Ai4

Unchecked AI retraining turns into a discriminatory sorting mechanism unless employers test models on marginalized workers and mandate human oversight for career decisions.

AI-assisted summary of Panel 6 - 1: Personalized, Powerful, and Equitable: The Role of AI in Workforce Retraining and — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel in August 2026, this presentation features Suzanne O'Keeffe, a professor of education at Sonoma State University. According to the description, the session explores how AI can personalize workforce retraining for diverse employees—including multilingual, neurodivergent, and disabled workers—while outlining a human-centered framework to address algorithmic bias and surveillance risks.

## TL;DR Suzanne O'Keeffe presents a human-centered model for leveraging artificial intelligence in organizational workforce retraining without displacing human judgment. She highlights opportunities for AI to support marginalized and vulnerable learners through personalized pacing, multimodal materials, and real-time language translation. To prevent AI from acting as a discriminatory sorting mechanism, she cautions against intrusive productivity tracking and algorithmic bias, proposing a six-pillar framework grounded in transparency, equity audits, and worker co-design. Leaders are encouraged to actively map baseline learning barriers and test new tools with tech-hesitant or edge-case employees first.

## Atmosphere & style The presentation maintains an earnest, structured, and empathetic tone typical of an academic speaking at an industry conference. O'Keeffe delivers her insights with steady, clear conviction, balancing enthusiasm for accessible educational tech with firm, protective warnings against invasive workplace surveillance. The pacing feels measured and pragmatic, culminating in memorable, human-focused maxims that keep the audience anchored on inclusion over technological hype.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) Artificial intelligence in workforce development should serve as an empowering support system for human learning rather than a replacement for human workers. - [00:52](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=52s) Without intentional equity, rapid AI adoption risks acting as a sorting mechanism that leaves multilingual, neurodivergent, disabled, and digitally emerging workers behind. - [01:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=115s) Organizations must integrate common AI tools—like adaptive learning, speech-to-text, and coaching—without surrendering human mentorship or organizational responsibility. - [03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) Shifting from rigid, one-size-fits-all training to personalized pacing and multimodal access removes physical and cognitive barriers for diverse learning styles. - [03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) Real-time translation and plain-language AI tools expand workforce inclusion by allowing employees to master complex concepts in their native languages. - [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) Algorithmic bias poses a major ethical threat, as historical training data can perpetuate past discrimination in career advancement and performance evaluations. - [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) Invasive tracking metrics like keystrokes and speed fail to account for contextual nuances, such as an employee taking extra processing time due to a disability. - [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) Overreliance on automation degrades workplace trust, strips away human empathy, and undermines the productive cognitive struggle needed for true learning. - [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) A human-centered framework dictates that AI should deliver raw data, while humans provide critical interpretation, context, and empathy. - [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) Responsible workforce AI adoption relies on six core pillars: transparency, human oversight, accessibility, equity audits, worker voice, and a strong learning culture. - [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) High-stakes career decisions, such as hiring, promotions, and terminations, must always require mandatory human oversight rather than automated outputs. - [12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) Leadership should pilot new AI retraining systems with marginalized and tech-hesitant employees first to ensure tools work effectively for the entire organization. - [12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) Continuous equity audits are essential guardrails to guarantee that retraining opportunities and career mobility are not placed behind locked doors. - [15:56](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=956s) Intentional, inclusive design choices determine whether AI will bridge workforce opportunity gaps or deepen existing societal inequities.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) Introduction and Research Purpose

Suzanne O'Keeffe, a professor of education at Sonoma State University whose research focuses on equitable learning opportunities and inclusion, welcomes the audience following lunch. She introduces her research topic, titled *"Personalized, Powerful, and Equitable: The Role of AI in Workforce Retraining."* The presentation explores how organizations can leverage artificial intelligence within the workforce to expand equitable opportunities without allowing technology to take over positions. O'Keeffe emphasizes that AI should serve as a support system that enables human workers to learn, adapt, and grow.

[00:52](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=52s) The Need for Equitable AI-Supported Workforce Development

Artificial intelligence is rapidly altering how work gets done, compelling organizations to fundamentally rethink how they deliver workforce learning. A major risk during this transition is job displacement coupled with unequal AI adoption, which leaves certain groups of employees particularly vulnerable. O'Keeffe specifically identifies the populations most at risk:

- Multilingual learners - Neurodivergent learners - Disabled learners - Emerging digital literacy learners

If organizational AI training accelerates faster than the support provided to employees, AI functions as a "sorting mechanism" rather than an empowering tool, stranding vulnerable workers who cannot keep pace with shifting job expectations.

[01:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=115s) Scope and Commonly Used AI Tools

Rather than presenting an overly theory-heavy or research-heavy paper, O'Keeffe focuses on concrete next steps to provide workforce opportunities across all learning styles. She examines how organizations can ethically deploy AI to personalize retraining and professional development without surrendering human judgment, mentorship, or organizational responsibility. Her conceptual framework synthesizes research across responsible AI, accessibility, organizational learning, workforce development, and inclusive technology adoption.

The analysis evaluates common AI tools present in professional learning environments: - Adaptive learning platforms - Speech-to-text tools - Real-time translation tools - Predictive analytics - AI-driven coaching systems

[03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) Opportunities for Personalized and Responsive Retraining

To foster growth, workforce training must shift away from rigid, "one-size-fits-all" formats toward flexible, multimodal, multilingual, and responsive models. Used in tandem with human workers, AI can strengthen inclusion through tailored design features:

```mermaid flowchart TD accTitle: Personalized AI Learning Supports subgraph Personalization["Personalized AI Learning Supports"] P["Personalized Pacing"] M["Multimodal Access"] L["Language Support"] C["Timely Coaching"] end

P --> |Identifies gaps| P_Out["Targeted Strategic Modules"] M --> |Combines formats| M_Out["Text, Audio & Hands-on Options"] L --> |Translates & simplifies| L_Out["Native Language Engagement"] C --> |Low-stakes feedback| C_Out["Safe Practice Environment"] ```

- **Personalized Pacing**: AI pinpoints specific concepts where a worker struggles, allowing organizations to deliver targeted, strategic training tailored to individual needs. - **Multimodal Access**: Information is delivered using text, audio, captions, summaries, and concrete examples, removing cognitive or physical barriers for workers who require extra processing time. - **Language Support**: Translation and plain-language tools broaden participation by allowing employees to learn complex concepts in their native languages. - **Timely Coaching**: Automated systems provide low-stakes feedback, offering safe spaces for practice without replacing human mentors.

[05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) Ethical Risks: Bias, Privacy, Surveillance, and Overreliance

While AI offers massive personalization benefits, uncritical or unethical deployment can intensify organizational inequities:

- **Algorithmic Bias**: Historical workplace data used to train AI models can reproduce past discrimination. For instance, if historical promotion metrics disproportionately favored native English speakers without disabilities over neurodivergent or multilingual workers, predictive career algorithms will unfairly penalize marginalized groups. - **Privacy Invasive Metrics**: Systems that track keystrokes, completion speeds, or interaction times fail to capture real-world contextual nuances, such as an employee taking extra processing time due to a disability or an unexpected interruption. - **Surveillance and Distrust**: Continuous monitoring erodes workplace trust and suppresses active participation. - **Automation Dependency**: Overreliance on automated judgment eliminates human empathy and undermines the natural cognitive "tension" required for true learning breakthroughs.

[09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) A Human-Centered Model for AI Workforce Inclusion

O'Keeffe proposes a foundational principle: *"AI belongs inside a human-centered workforce learning system and not to replace one."* AI should deliver raw data, while humans provide critical interpretation and empathy.

```mermaid flowchart LR accTitle: Six Pillars of Human-Centered AI Retraining subgraph Framework["Human-Centered AI Framework"] direction TD A["1. Transparency"] B["2. Human Oversight"] C["3. Accessibility"] D["4. Equity Audits"] E["5. Worker Voice"] F["6. Learning Culture"] end ```

The model rests on six core pillars: 1. **Transparency**: Clear policies around data usage, operational limits, and explicit worker consent to build organizational trust. 2. **Human Oversight**: Mandatory human intervention on high-stakes career decisions, including hiring, termination, and promotions. 3. **Accessibility**: Flexible, multimodal learning systems that equalize opportunity for all background types. 4. **Equity Audits**: Continuous evaluations to identify which populations benefit from a tool and who might be excluded. 5. **Worker Voice**: Co-designing AI tools alongside employees prior to deployment to build genuine buy-in and practical utility. 6. **Learning Culture**: Creating environments rich in feedback, safe practice, and trust where AI is viewed as an empowering tool rather than a replacement.

[12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) Practical Action Steps for Leadership

O'Keeffe provides three concrete strategies organizational leaders can implement immediately:

1. **Map Workforce Learning Barriers**: Identify pre-existing friction points in language, technology literacy, time, and confidence before introducing new tools. For instance, heavy text-based manuals can be supplemented with AI translation or summary tools to support multilingual staff. 2. **Pilot Within the Margins**: Test new AI systems with marginalized or tech-hesitant employees rather than confident early adopters. Resolving barriers for edge cases ensures the tool works effectively for the entire organization. 3. **Establish Guardrails**: Conduct continuous equity audits to guarantee equal access across all tiers. O'Keeffe summarizes this imperative with the warning: *"Do not put the ladder behind a locked door."*

[15:56](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=956s) Final Takeaways

O'Keeffe concludes by reiterating that while AI fundamentally changes the nature of work, organizations must make intentional, human-centered design choices. By prioritizing inclusion and equitable learning supports, technology can expand workforce access rather than deepen existing inequalities.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) Personalized, Powerful, and Equitable: The Role of AI in Workforce Retraining — Suzanne O'Keeffe

> [00:52](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=52s) sorting mechanism — Suzanne O'Keeffe

> [03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) one-size-fits-all — Suzanne O'Keeffe

> [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) AI belongs inside a human-centered workforce learning system and not to replace one. — Suzanne O'Keeffe

> [12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) Do not put the ladder behind a locked door. — Suzanne O'Keeffe

## Mentioned Suzanne O'Keeffe, Sonoma State University, Ai4

## Video description Suzanne O'Keeffe, Professor, Sonoma State University

Artificial intelligence is transforming the workplace, but will it create new opportunities for everyone, or leave some workers behind? As organizations increasingly rely on AI, I believe we have a responsibility to ensure that workforce retraining is accessible, equitable, and responsive to the diverse needs of today's employees. In this session, I will explore how AI can personalize learning and support workers with disabilities, multilingual backgrounds, and varying levels of digital literacy while also addressing the ethical challenges of algorithmic bias, data privacy, and overreliance on automation. I will share a human-centered framework for implementing AI that balances innovation with inclusion and keeps people, not technology, at the center of workforce development. Whether you work in education, industry, or organizational leadership, you will leave with practical ideas for using AI to build a workforce where everyone has the opportunity to learn, adapt, and thrive.

[Source](https://www.youtube.com/watch?v=PuGpoOQb9n4)

中文简报

奥基夫警告：AI再培训恐将弱势员工淘汰出局 缺乏约束的AI再培训将演变为歧视性筛选机制，除非雇主针对边缘化员工测试模型，并对职业发展决策强制实行人工监督。 Suzanne O'Keeffe提出了一个以人为本的模型，主张在利用人工智能开展组织员工再培训的同时不取代人类判断。她指出，AI有望通过个性化进度、多模态材料和实时语言翻译为边缘化及弱势学习者提供支持。为防止AI沦为歧视性的筛选机制，她告诫要防范侵入式生产力追踪与算法偏见，并提出了一个立足于透明度、公平性审查和员工协同设计的六支柱框架。她建议管理者积极排查基准学习障碍，并率先在对技术持观望态度或处于边缘情形的员工群体中测试新工具。 索诺马州立大学教授 Suzanne O'Keeffe 指出，AI 在劳动力再培训中应定位为赋能人类学习的支持系统；若缺乏公平性考量，AI 的加速应用可能沦为边缘化多语言、神经多样性及残障员工的筛选机制。 (0:52) 企业需从“一刀切”的培训模式转向基于 AI 的个性化支持，结合针对性进度、多模态呈现与母语实时翻译，消除员工的认知与物理障碍并提供低风险的练习环境。 (3:07) 警惕历史训练数据引发的算法偏见以及击键、速度等侵入式监控指标，过度依赖自动化不仅忽视了残障等现实情境差异，还会削弱职场信任与深度学习所需的认知张力。 (5:55) O'Keeffe 提出涵盖透明度、人类监督与公平审计等六大支柱的以人为本框架，并建议领导层优先在边缘群体和技术迟疑员工中试点 AI 工具，确保高风险决策始终有人类把关。 (9:22)

Reading summary

Suzanne O'Keeffe argues that artificial intelligence must serve as an assistive, human-centered support system in workforce retraining rather than an automated replacement. Without intentional, equity-focused deployment, accelerated AI adoption risks functioning as a sorting mechanism that strands vulnerable workers. Risks of Unequal AI Workforce Adoption (0:52) Rapid technological transitions risk displacing employees when enterprise AI adoption outpaces internal training and employee support. O'Keeffe identifies multilingual learners, neurodivergent workers, disabled employees, and individuals with emerging digital literacy as the populations facing the highest risk during this transition. When organizations accelerate AI implementation without tailoring support, the technology operates as a sorting mechanism rather than an equalizer. In this scenario, workers who cannot instantly adapt to shifting job expectations are excluded from emerging career opportunities. Opportunities in Personalized, Multimodal Retraining (3:07) To counter workforce exclusion, O'Keeffe advocates moving away from rigid, one-size-fits-all training toward flexible, multimodal, and responsive educational models. Integrating tools such as adaptive platforms, speech-to-text, real-time translation, predictive analytics, and automated coaching systems allows organizations to accommodate varied learning styles without eliminating human mentorship. These systems enhance inclusion by personalizing pacing to address specific skill gaps and offering content across text, audio, and hands-on formats. Additionally, translation and plain-language tools allow employees to process complex topics in their native languages, while automated coaching offers a low-stakes environment for safe practice. Algorithmic Bias and Surveillance Risks (5:55) Uncritical deployment of AI in professional development introduces severe ethical hazards. Algorithmic bias can reproduce historical discrimination if training models rely on past workplace metrics that disproportionately favored native English speakers without disabilities over neurodivergent or multilingual employees. Invasive surveillance metrics present an additional danger. Tracking completion speeds, interaction times, or keystrokes ignores legitimate contextual factors, such as an employee taking extra processing time due to a disability. O'Keeffe warns that continuous monitoring damages workplace trust, while overreliance on automation removes the human empathy and cognitive effort necessary for authentic learning breakthroughs. Human-Centered Framework and Implementation (9:22) O'Keeffe asserts that AI belongs inside a human-centered workforce learning system rather than replacing one, noting that AI should supply data while humans provide critical interpretation. Her model outlines six core pillars: transparency, mandatory human oversight on high-stakes career decisions (including hiring, promotion, and termination), universal accessibility, continuous equity audits, worker voice in co-designing tools, and a trust-based learning culture. For immediate leadership implementation, O'Keeffe recommends three actions: mapping pre-existing friction points such as language and literacy before deploying software; piloting systems with marginalized or tech-hesitant staff instead of confident early adopters; and conducting rigorous equity audits to ensure access is not locked away from the employees who need it most. O'Keeffe concludes that the transformative impact of workplace AI depends on intentional organizational design. Ensuring that algorithms remain governed by human judgment and accessibility safeguards dictates whether workforce retraining broadens career mobility or reinforces existing structural inequalities.

精炼阅读摘要

索诺马州立大学（Sonoma State University）教育学教授苏珊·奥基夫（Suzanne O'Keeffe）指出，组织在利用人工智能进行劳动力再培训时，必须将其置于以人为本的框架内作为辅助工具，防止技术演化为淘汰边缘员工的“分流机制”，从而实现公平且个性化的技能提升。 劳动力转型中的弱势群体与分流风险 (0:52) 奥基夫指出，人工智能正快速重塑工作流程，迫使组织重新审视员工学习模式的构建方式。在此转型过程中，岗位流失叠加不平等的AI采纳风险，对特定员工群体构成了尤为严重的威胁，包括多语言学习者、神经多样性（neurodivergent）学习者、残障学习者以及处于数字素养起步阶段的员工。奥基夫警告，如果企业内部的AI培训推进节奏快于向员工提供的基础支持，人工智能就可能沦为一种“分流机制”（sorting mechanism），将无法及时适应工作预期变化的弱势群体抛在身后。 驱动自适应与多模态培训的技术潜能 (3:07) 为促进员工成长，劳动力培训必须从僵化的一刀切模式转向灵活、多模态、多语言及响应及时的体系。奥基夫分析了自适应学习平台、语音转文本、实时翻译、预测分析与AI辅导等工具的效能。她指出，AI能精准定位员工知识断层，使企业能够提供个性化进度的定向培训；多模态支持结合了文本、音频、字幕、摘要与实操案例，消除了需要额外信息处理时间的员工的认知与物理障碍；本地语言翻译工具降低了复杂概念的理解门槛；而AI系统则能提供低风险的即时反馈，营造出无需取代人类导师的安全练习环境。 算法偏见、监控入侵与自动化依赖风险 (5:55) 奥基夫强调，不加审视或缺乏伦理考量的技术部署会进一步固化并放大组织现存的不平等。在算法层面，若采用包含历史歧视的数据训练模型，系统将重现过往偏差，例如历史晋升指标若偏向非残障的英语母语者，职业预测算法就会对边缘群体造成系统性惩罚。在评估层面，追踪按键记录、完成时长或交互速度等侵入式指标无法反映残障员工额外处理时间或意外干扰等现实情境。此外，持续监控会破坏组织信任度，过度依赖自动化裁决更会剥离人类同理心，抹杀学习突破所必需的认知张力。 六大支柱框架与管理层的实践路径 (9:22) 奥基夫提出“AI属于以人为本的劳动力学习系统，而非取而代之”的核心准则，并建立了由六大支柱组成的治理框架：明确数据用途与员工知情同意的透明度、在招聘与晋升等重大决策中保留强制性的人工监管、消除技术壁垒的多模态可访问性、定期评估受益与排斥群体的公平性审计、让员工参与协同设计的发声机制，以及营造充满反馈与信任的学习文化。针对企业领导层，她提出三项落地策略：引入工具前先摸排语言、技能和信心壁垒；优先在边缘群体或对技术持疑的员工中开展试点，通过化解极端痛点确保普适性；并建立常态化公平护栏，确保不把向上的发展阶梯锁在门后。 奥基夫总结认为，尽管人工智能从根本上改变了工作性质，但组织必须做出有意图且以人为中心的设计抉择。只有优先保障包容性与公平的学习支持，技术才能拓展劳动力的发展机遇，避免既有不平等的进一步加剧。

Detailed notes — zh

## Background 本内容由 Ai4 频道于 2026 年 8 月 18 日发布在 YouTube 平台上。视频记录了索诺马州立大学教授 Suzanne O'Keeffe 的主题演讲，探讨了在人工智能重塑职场的背景下，组织应如何利用 AI 实现无障碍、公平且个性化的员工再培训，并兼顾算法偏见与数据隐私等伦理挑战。 ## TL;DR Suzanne O'Keeffe 教授提出，AI 在员工再培训中应充当赋能人类学习的辅助工具，而非将弱势群体边缘化的“人员筛选机制”。通过自适应进度、多模态访问、语言支持及即时辅导等机制，AI 能够打破传统“一刀切”培训模式，但在落地过程中必须防范算法偏见、过度监视与自动化依赖等风险。为此，她提出了涵盖透明度、人类把关、公平性审计等六大支柱的以人为中心框架，并建议管理者从排查现有学习壁垒和开展边缘群体试点切入，构建真正包容普惠的职业发展生态。 ## Atmosphere & style 整场演讲呈现出严谨、专业且极富人文关怀的基调，主讲人条理清晰地将前沿技术与教育公平理念深度融合。叙事从对脆弱群体的现实担忧平稳过渡到系统性的解决方案，语调沉稳而坚定。观众在聆听过程中不仅能感受到学术理论的扎实支撑，更能体会到呼吁技术向善、消除职场壁垒的紧迫感与行动号召力。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) AI 正在深刻重塑职场，若企业缺乏同步的赋能与支持，AI 将沦为残酷的“人员筛选机制”，导致无法跟上节奏的员工被加速边缘化。 - [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) 在缺乏针对性支持的培训环境中，多语种、神经多样性、残障以及数字素养起步阶段的劳动者是最容易受到技术排斥的四大脆弱群体。 - [01:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=115s) 企业在利用自适应学习等 AI 工具实现再培训时，核心挑战在于不能牺牲人类判断、导师指导与组织的伦理责任。 - [03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) AI 正在通过自适应进度调整、多模态访问体验、多语言与通俗语言支持以及低风险即时辅导四大机制，彻底告别传统“一刀切”的培训模式。 - [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) 若算法基于历史上偏向健全人或母语者的晋升数据进行训练，将把既有的结构性偏见固化为晋升标准，进一步压缩边缘群体的职场空间。 - [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) 机械采集员工的学习时长等遥测数据极易产生失真误判，过度监视更会侵蚀员工对组织的信任与参与积极性。 - [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) 过度依赖自动化会抹杀员工面对认知挑战时的人性化辅导，AI 无法替代真实的人际协作与同理心支持。 - [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) 以人为中心的 AI 包容模型强调明确分工：AI 负责提供底层数据支持，人类则负责提供同理心与价值诠释。 - [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) 组织落地框架应涵盖透明度、人类把关、无障碍设计、公平性审计、员工发声与学习型文化六大支柱，严禁 AI 作为人事决策的最终裁决者。 - [12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) 技术落地的第一步是全面排查组织内的隐性学习壁垒，审视并消除新工具在语言、认知与技术水平上预设的前置门槛。 - [14:16](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=856s) 推行“在边缘群体中进行试点”（Piloting within the margins）原则：若一款工具能让最容易受挫的员工顺利上手，它就能无缝适用于全体员工。 - [14:16](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=856s) 组织必须设立持续的公平性审计护栏，“绝不能把发展的阶梯锁在紧闭的门后”，防止技术红利演变为新的职场鸿沟。 - [15:56](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=956s) AI 既能拓展平权机会也能加剧不平等，组织在推进员工发展时绝不能听任技术野蛮生长，必须做出深思熟虑的“刻意设计选择”。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) AI 时代的员工再培训：核心使命与弱势群体面临的风险 来自索诺马州立大学（Sonoma State University）的教育学教授 Suzanne O'Keeffe 开启了题为《个性化、强大与公平》（Personalized, Powerful, and Equitable）的演讲。她强调，当前 AI 正在从根本上重塑工作方式，但核心问题在于：我们应如何利用 AI 为全体劳动者创造更公平的机会，确保 AI 是作为支持人类成长、学习与适应的辅助系统，而非直接取代人类岗位。如果组织仅仅加快 AI 的部署，而对员工的赋能与支持未能同步跟进，AI 就会沦为一种残酷的“人员筛选机制”（sorting mechanism），将无法跟上技术节奏的群体彻底边缘化。 在缺乏适当支持的背景下，最容易受到冲击的四类脆弱劳动群体包括： - **多语种学习者（Multilingual learners）**：非母语环境增加了理解技术指令与培训材料的门槛。 - **神经多样性学习者（Neurodivergent learners）**：需要差异化的认知路径与信息输入模式。 - **残障学习者（Disabled learners）**：依赖无障碍功能与替代性交互方式。 - **数字素养处于起步阶段的学习者（Emerging digital literacy learners）**：对新型数字工具的适应曲线较长。 --- ### [01:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=115s) 概念研究框架：AI 如何赋能个性化与包容性职业发展 Suzanne O'Keeffe 介绍了一篇跨学科概念性研究论文，该研究整合了负责任 AI（Responsible AI）、无障碍设计（Accessibility）、组织学习（Organizational Learning）、劳动力发展（Workforce Development）及包容性技术采纳（Inclusive Technology Adoption）等领域。其核心研究问题在于：**组织如何在不牺牲人类判断、导师指导（Mentorship）与组织责任的前提下，合乎伦理地利用 AI 来实现员工再培训与专业学习的个性化？** 企业在专业培训场景中常见的 AI 工具包括：自适应学习平台（Adaptive Learning）、语音转文字工具（Speech-to-Text）、机器翻译系统、预测分析工具（Predictive Analytics）以及 AI 智能教练系统。 ```mermaid flowchart TD accTitle: AI 赋能的包容性员工培训框架 subgraph Core["核心目标"] A["个性化与包容性再培训"] end subgraph Modalities["四大支柱"] B["自适应进度 (Pacing)"] C["多模态访问 (Multimodal)"] D["语言支持 (Language)"] E["即时辅导 (Coaching)"] end A --> B A --> C A --> D A --> E ``` --- ### [03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) 告别“一刀切”：AI 推动个性化培训的四大核心机制 传统的“一刀切”（one-size-fits-all）培训模式正在被更加灵活、多语言、多模态且具有即时响应能力的个性化模型所取代： 1. **个性化进度调整（Personalized Pacing）**：AI 能够动态识别员工在特定概念或技能上的理解盲区，针对薄弱环节提供定向辅导，允许学习者按适合自己的节奏推进。 2. **多模态访问体验（Multimodal Access）**：通过文本、音频、实时字幕、智能摘要和示例演示等多种形式呈现培训内容，打破认知障碍。例如，处理信息较慢的员工可以同时边听边读，或通过模拟实操深化理解。 3. **多语言与通俗语言支持（Language Support）**：提供实时翻译与简明语言转换（plain language tools），让非母语员工能够用母语吸收核心业务逻辑，大幅拓宽参与度。 4. **低风险即时辅导（Low-Stakes AI Coaching）**：充当沙盒练习环境，为员工提供无心理负担的即时反馈，帮助其反复试错与练习，作为人类导师指导的重要补充。 --- ### [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) 伦理隐患：算法偏见、数据隐私与过度自动化的陷阱 若缺乏伦理约束，用于个性化赋能的技术可能会反向加剧职场不公、监视与排斥： - **历史数据固化偏见（Algorithmic Bias）**：若算法依据过往晋升数据进行训练，而历史上英语母语者或健全员工的晋升率远高于多语种、残障或神经多样性群体，模型便会将这种结构性偏差编码为“晋升标准”，进一步阻碍边缘群体的职业晋升。 - **数据隐私与误导性指标（Privacy & Telemetry Data）**：平台对员工完成模块的时长、按键频率等交互数据的机械采集极易产生误导。某些员工可能因残障、深思熟虑或工作中途被突发事项打断而耗时较长，若系统将其判定为“表现不佳”，则会导致评估严重失真。 - **过度监视削弱信任（Surveillance & Trust）**：当员工感到自己每一步操作都被严密监控而非被支持赋能时，对组织的信任感和参与积极性将直线下降。 - **过度依赖自动化（Automation Dependency）**：将 AI 视为人类辅导与同理心的替代品，会抹杀员工在真实协作与面对认知挑战（"tension between understanding a concept and having that breakthrough"）时的突破性成长机会。 --- ### [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) 以人为中心的 AI 劳动力包容模型（六大支柱） 针对上述挑战，Suzanne O'Keeffe 提出了一个“以人为中心”的系统性实施框架：**AI 的角色是提供底层数据支持，而人类负责提供同理心与价值诠释（"AI provides the data and the human provides the empathy and the interpretation"）**。 | 支柱维度 | 核心实践原则 | |:--- |:--- | | **透明度 (Transparency)** | 明确告知员工数据采集的目的、算法边界与监控范围，事先获得知情同意以建立组织互信。 | | **人类把关 (Human Oversight)** | 严禁将 AI 作为录用、解聘或晋升的最终决策者，必须由人类管理者综合考量算法无法感知的复杂情境。 | | **无障碍设计 (Accessibility)** | 构建多模态、灵活的无障碍学习环境，为不同认知与生理特质的员工提供平等的竞技场。 | | **公平性审计 (Equity Audits)** | 持续审查技术采纳的收益分配：明确新工具使谁受益、谁可能被阻隔在外，定期评估不同群体的适用性。 | | **员工发声 (Worker Voice)** | 突破自上而下的决策惯性，让一线员工参与工具的共同设计（co-design），确保工具真正契合工作实际并提升采纳意愿。 | | **学习型文化 (Learning Culture)** | 营造鼓励试错、及时反馈与安全的文化氛围，让员工确信 AI 是辅助个人技能精进的伙伴而非淘汰工具。 | --- ### [12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) 落地实践第一步：排查组织内部的学习壁垒 演讲最后转向领导者在“周一早晨重返工作岗位”时即可立即启动的三大实际行动。第一项具体举措为**梳理并绘制现有劳动力学习环境中的潜在壁垒（Map the barriers in your workforce learning environment）**，从物理无障碍、语言支持、数字技能门槛到认知负荷进行全面排查。 ### [13:09](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=789s) 识别学习壁垒与利用 AI 赋能多元化员工 在推行任何 AI 工具或再培训计划之前，组织必须首先深入审视员工队伍中现存的各种隐性壁垒。这包括语言障碍、资源获取途径受限、时间匮乏、缺乏自信，或是员工技术水平的参差不齐。管理者需要认真反思：在推出一款全新的 AI 工具时，我们在潜意识里对学习者预设了哪些门槛？学习者为了使用该工具或参与新工具的学习，究竟需要具备哪些前置条件？一旦明确了这些潜在阻碍，找准员工在成长、学习以及接触新技术时可能遭遇困难的节点，组织才能有的放矢地提供支持。 Suzanne O'Keeffe 举了一个非常具体的实际场景：如果培训对象是一位多语言学习者（Multilingual Learner），而组织分发给他们的却是一份充斥着大量复杂专业文本的培训手册或文档，这种培训往往收效甚微。此时组织应当思考的是，能够引入什么样的 AI 工具来协助这类员工，以一种对他们而言真正有意义、易于理解且符合其语言习惯的方式去吸收这些培训手册与核心资源，从而打破语言对技能提升的封锁。 ```mermaid flowchart LR accTitle: 员工培训壁垒识别与 AI 赋能流程 subgraph S1["壁垒评估"] A["审视员工现状<br/>(语言/时间/技术/信心)"] --> B["识别前置假设与学习痛点"] end subgraph S2["落地支持"] B --> C["部署针对性 AI 工具<br/>(如文本转化/语言辅助)"] C --> D["实现有意义的学习与吸收"] end ``` --- ### [14:16](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=856s) 从边缘群体切入试点与设立公平保障护栏 为了确保技术落地的普惠性，Suzanne O'Keeffe 提出了“在边缘群体中进行试点”（Piloting within the margins）的核心策略。这意味着在测试 AI 工具时，测试对象不应是那些自信满满、热衷于尝试新事物的前期尝鲜者（Early Adopters），而必须是那些最有可能被排斥、在技术应用中最容易遇到困难的边缘员工。组织需要让他们真正去深度体验和操作这些工具，并主动向他们提问：“工具的优势在哪里？阻碍又在哪里？你们目前是如何使用的？你们未来希望如何使用它？”这种设计逻辑在于：如果一款工具能够让最边缘化、最容易受挫的员工顺利上手并从中受益，那么它几乎能无缝适用于所有人。前期在这方面投入的时间，能够帮助组织在早期及时暴露并修复局限，避免在全面铺开后陷入部分员工无法使用的被动救火局面。 与此同时，组织必须设立严格的“护栏”（Guardrails），这意味着要推行持续的公平性审计（Equity Audits），防止技术带来收益分化，导致一部分人大幅受益而另一部分人被彻底边缘化。Suzanne O'Keeffe 用了一个生动的比喻来告诫大家：“千万不要把阶梯锁在紧闭的门后”（"Do not put the ladder behind a locked door."）。如果企业为员工提供了发展的阶梯、资源与 AI 工具，就必须确保每个人都能拥有平等的获取与使用权限。 - **边缘试点原则**：将最易被技术排斥的员工置于测试中心，通过前置反馈修复工具缺陷。 - **持续公平审计**：建立监测机制，防范因数字鸿沟导致的收益不均与职场不平等加剧。 - **平权核心理念**：工具和晋升阶梯必须保持开放，消除任何形式的隐性准入门槛。 --- ### [15:56](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=956s) 总结：以人为本的 AI 时代职业培训设计 在演讲的总结陈词中，Suzanne O'Keeffe 提炼了 AI 时代劳动力发展的三大核心法则：第一，AI 正在深刻改变工作形态，而员工随之亟需公平、包容的学习支持；第二，AI 是一把双刃剑，它既有能力极大拓展机会与准入范围，也可能进一步拉大并加深现有的不平等鸿沟；第三，组织在为员工创造学习与成长机会时，绝不能听任技术野蛮生长，而必须做出具有前瞻性与包容性的“刻意设计选择”（Intentional Design Choices）。 演讲最后在热烈的掌声中落下帷幕，Suzanne O'Keeffe 随后留出约三分钟的时间，开放现场提问并与与会者展开进一步的互动交流。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) Personalized, Powerful, and Equitable — Suzanne O'Keeffe > [00:04](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=4s) sorting mechanism — Suzanne O'Keeffe > [03:07](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=187s) one-size-fits-all — Suzanne O'Keeffe > [05:55](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=355s) tension between understanding a concept and having that breakthrough — Suzanne O'Keeffe > [09:22](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=562s) AI provides the data and the human provides the empathy and the interpretation — Suzanne O'Keeffe > [12:49](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=769s) Map the barriers in your workforce learning environment — Suzanne O'Keeffe > [14:16](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=856s) Piloting within the margins — Suzanne O'Keeffe > [14:16](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=856s) Do not put the ladder behind a locked door. — Suzanne O'Keeffe > [15:56](https://www.youtube.com/watch?v=PuGpoOQb9n4&t=956s) Intentional Design Choices — Suzanne O'Keeffe ## Mentioned Suzanne O'Keeffe, Sonoma State University, Ai4 ## Video description Suzanne O'Keeffe, Professor, Sonoma State University Artificial intelligence is transforming the workplace, but will it create new opportunities for everyone, or leave some workers behind? As organizations increasingly rely on AI, I believe we have a responsibility to ensure that workforce retraining is accessible, equitable, and responsive to the diverse needs of today's employees. In this session, I will explore how AI can personalize learning and support workers with disabilities, multilingual backgrounds, and varying levels of digital literacy while also addressing the ethical challenges of algorithmic bias, data privacy, and overreliance on automation. I will share a human-centered framework for implementing AI that balances innovation with inclusion and keeps people, not technology, at the center of workforce development. Whether you work in education, industry, or organizational leadership, you will leave with practical ideas for using AI to build a workforce where everyone has the opportunity to learn, adapt, and thrive. [Source](https://www.youtube.com/watch?v=PuGpoOQb9n4)

Source quotes

[0:04] Suzanne O'Keeffe: Personalized, Powerful, and Equitable [0:04] Suzanne O'Keeffe: sorting mechanism [3:07] Suzanne O'Keeffe: one-size-fits-all [5:55] Suzanne O'Keeffe: tension between understanding a concept and having that breakthrough [9:22] Suzanne O'Keeffe: AI provides the data and the human provides the empathy and the interpretation [12:49] Suzanne O'Keeffe: Map the barriers in your workforce learning environment [14:16] Suzanne O'Keeffe: Piloting within the margins [14:16] Suzanne O'Keeffe: Do not put the ladder behind a locked door. [15:56] Suzanne O'Keeffe: Intentional Design Choices

Source quotes

[0:04] Suzanne O'Keeffe: Personalized, Powerful, and Equitable: The Role of AI in Workforce Retraining [0:52] Suzanne O'Keeffe: sorting mechanism [3:07] Suzanne O'Keeffe: one-size-fits-all [9:22] Suzanne O'Keeffe: AI belongs inside a human-centered workforce learning system and not to replace one. [12:49] Suzanne O'Keeffe: Do not put the ladder behind a locked door.

Original transcript

[0:04] Hi everyone, welcome. Thanks for uh being here  after lunch. So I'm going to present um on a [0:10] a a presentation on research called personalized,  powerful, and equitable. Um, so this is really [0:16] about AI in the workforce and how we can create  more equitable opportunities for our workers [0:23] and making sure that AI is not taking over these  positions but being a support system for humans to [0:30] grow, learn and adapt. So essentially um I came up  with research on AI supported workforce retaining [0:38] and inclusion. Um, as I stated earlier, my name  is Suzanne O'Keefe. I'm a professor of education [0:44] at Sonoma State. Um my research is really embedded  in equitable learning opportunities and inclusion. [0:52] Uh so go ahead and get started here. Um so why  does this matter? So AI is changing the way that [1:00] work is done and so organizations need to  rethink how we support workforce learning. [1:07] Um so essentially risk of job displacement  and unequal adaptation of AI is um [1:14] essentially making some workers a little  bit more vulnerable than others. And so [1:19] the workers who are at the greatest risk include  multilingual learners, neurodeivergent learners, [1:25] disabled learners, and then also emerging digital  literacy learners. So if AI training is moving [1:33] faster than the support we're giving, then AI  becomes a sorting mechanism for the workers we [1:39] have. And if we think about it, if we're allowing  AI to take over these jobs, then we're leaving a [1:45] lot of people in um positions where they're unable  to keep up with the growing demands of AI. So, [1:55] so okay. So, what I examined, um, I wrote a  conceptual paper. What I'm going to present [2:01] today is not going to be research heavy or,  um, or theoryheavy. It's really going to be [2:07] about what is, uh, next steps and ways that we  can provide opportunities for all different uh, [2:13] learners and all different individuals in the  workforce. And so what I examined is how can [2:19] organizations use AI ethically to personalize  workforce retraining and professional learning [2:25] without replacing human judgment,  mentorship or organizational responsibility. [2:32] So in this paper, I looked at research from lots  of different angles. I looked at responsible AI, [2:38] accessibility, organizational learning, workforce  development, and inclusive technology adoption. [2:45] Um the tools that I was specifically looking at  when I was uh researching this were the tools that [2:51] we commonly see in professional learning spaces  such as adaptive learning, speechtoext tools, [2:59] translation tools, predictive  analytics and AI and coaching. [3:07] So how can AI uh opportunities for growth and  learning? How can we provide this opportunity [3:15] for workers to grow in the age of AI? And so  really what it comes down to is ensuring that [3:22] the trainings we are providing are flexible,  multilingual, multimodal, and responsive. [3:29] So instead of using AI to replace workers, we're  using AI in tandem with workers to ensure that [3:36] that uh that our organizations are growing uh on  a human level and using AI to produce stronger um [3:45] stronger and more equitable learning experiences.  So here's some way that we can do that. [3:51] Instead of looking at workforce training as this  one-sizefits-all model, looking at it in a more [3:57] individualized um responsive to the learner model.  So for instance, if we are training our workers or [4:04] providing uh professional development, AI gives  us an opportunity to create personalized pacing, [4:11] which means that if there's workers that are  struggling in certain areas or they're struggling [4:15] to understand certain concepts, AI can help us to  identify what those concepts are and provide more [4:21] strategic training for those specific workers. AI  also gives us multimodal access which means that [4:29] when we're training workers whether again we're  training on AI or we're training on on other um [4:35] other soft skills or hard skills we can use  text audio captions um summaries and examples [4:42] to reduce barriers to learning. So if you have  someone who might take a little bit longer to [4:47] process information, maybe they are both reading  it and listening to it at the same time or maybe [4:52] there are hands-on learning opportunities that can  provide can be provided through AI experiences. [4:59] The other piece of it is language support. So  AI offers uh transition translation and plain [5:06] language tools that can widen participation.  So again, this allows for multiple um [5:13] opportunities for folks to engage in the  learning process, maybe in their native language, [5:19] which might increase their ability to understand  the concepts a little bit uh more clearly. [5:25] And then lastly, there's uh timing coaching. So  AI can be used to provide low stakes feedback, [5:31] not to be replaced by a mentor or um a human  support, but low stakes feedback that can [5:39] create opportunities for the worker  to um consider their learning needs [5:45] and um have opportunities to practice and  in areas that they need to um get better in. [5:55] So, not only can AI provide these opportunities,  but there there can be um opportunities for a [6:02] deepened inequity if AI is not used correctly  or not used ethically. And so, some of the same [6:09] tools that we use to personalize the supports can  also intensify bias, surveillance, and exclusion. [6:16] So, for example, bias and AI when we're talking  about workforce learning, historical data can [6:23] be used to reproduce inequities uh for  folks that are flagged unsupported or um [6:31] not advancing. So for instance, if a a company  is looking at the metrics in terms of promotions [6:37] and they're looking at specific workers and  the metrics that were used to promote them, [6:42] you might find that native English speakers  or um maybe workers without disabilities [6:49] are promoted at a much higher rate um than  folks with disabilities, neurodeivergent folks [6:55] or um multilingual learners. And so if that's the  case then uh the the algorithms that are being [7:01] created around advancement in that organization  may provide bias information um when we're looking [7:10] at other types of learners. The other issue is  privacy. Um the training systems that we use [7:16] are collecting data all the time. And so if we're  looking at the data in terms of how quickly um a [7:22] person gets through a module or keystrokes or how  they're able to interact with AI, we might be um [7:29] the the person may have a disability or they may  have a um need a little bit more time processing [7:35] or maybe they got interrupted when they were in  the middle of their training. And so if we're [7:38] using those those metrics to see if they are a  strong worker or not, we're we're not providing [7:45] we're not getting the the most up-to-date  realistic information that we um [7:50] that we would need to to inform our decisions.  The other piece is surveillance. Um if workers are [7:56] feeling like they're constantly monitored instead  of supported, then we're reducing that trust and [8:02] participation. So really being clear about how  the AI is being used and um and how workers are [8:09] potentially being surveiled so that they don't  feel like they're being watched all the time. [8:14] And then lastly, there's this automation  dependency. So if we are um using AI and [8:21] and training our our workers to use AI as just  a a means of replacing this coaching practice [8:28] and judgment and we're relying solely on AI to  inform our decisions, then we might be missing the [8:35] the human empathy component that could be really  important in helping support a um a worker in in [8:43] an organization. And so really thinking about how  learning and collaboration happen and that tension [8:50] between understanding a concept and um kind of  having that breakthrough and how important that is [8:56] for the learning process. So when we just  constantly fall back on that automation or ask our [9:02] our workers to fall back on that automation,  we are losing that ability for them to grow. [9:08] So really this is a shift away from the the  scoring and the sorting and the monitoring [9:13] of workers and moving towards supporting  workforce learning in a more holistic way. [9:22] So uh what I'm proposing in the the kind of crux  of the research here is a human- centered model [9:29] for AI workforce inclusion. So AI belongs inside  a human- centered workforce learning system and [9:37] not to replace one. So there are six different  facets here to to think of when we're building [9:43] this workforce. Um when we're we're building this  inclusive workforce. So the first is transparency. [9:51] Um if you're using AI in your organization, being  really transparent about how data is being used, [9:57] um limits and also the consent. So if if there is  going to be surveillance and and how it's going [10:04] going to be used, it's important that workers  understand that upfront so that they have um [10:10] they've built trust with their organization.  The other piece is human oversight. So [10:16] never allowing AI to be the end- all beall. So  never allowing AI to make decisions on hiring or [10:22] firing or advancement without having some type  of uh a human oversight to ensure that the um [10:30] the the variables that AI cannot pick up on  are being considered. Another piece of this [10:37] is accessibility. And so again, when we provide  accessible and um inclusive multimodal, flexible [10:46] uh systems of learning, then we're providing  an equal playing field for everyone and we're [10:51] not just um providing folks that may fit into a  certain box that uh that pathway to advancement. [11:00] Um the next piece is equity audits.  So, who benefits from this particular u [11:07] AI that you're bringing into your workforce?  So, um when you're thinking about creating a [11:14] workflow or you're thinking about adopting AI,  who is benefiting from this? Is there anybody [11:20] who may not be able to um access it? Is there is  there any population that may not be able to um [11:28] be successful with that specific uh AI tool? So  really thinking about who are your users and um [11:36] and and constantly auditing that. The next is um  worker voice. So this one's a little controversial [11:43] because we're typically in organizations we  kind of feel like hey we're going to make those [11:46] decisions on our own based on data really. But  um having that that that voice from the worker [11:53] allows you to co-design what tools you're going to  use ahead of time so that um again you're getting [11:59] that buy in. Is this something that could help  our organization? Is it something that you think [12:04] would be useful? So really getting that feedback  um is a powerful way of of providing that buy in. [12:12] And then lastly is the the learning culture. So  when AI is being implemented or when uh we're [12:20] learning new skills in the workplace, is there  opportunities for feedback? Is there opportunities [12:26] for practice? um building trust with the worker  and the organization so that um the worker [12:32] doesn't feel like AI is replacing them but it's  a tool for them to become better at what they do. [12:39] So AI provides the data and human the human  provides the empathy and the interpretation. [12:49] So practical implications of something you can  do uh Monday morning when you return to work. [12:56] So three practical moves leaders can act  on immediately. The first thing you can do [13:01] is you can map the barriers in your workforce  learning environment. So consider the barriers [13:09] in your workforce right now in terms of language,  access, time, confidence or technology skills [13:16] the employee may or may not have. So really  looking at if you're rolling out an AI tool, [13:22] what is the assumptions that  the the learner needs to have [13:28] in that that setting to be able to access this  tool or to be able to access learning a new tool. [13:35] So once you can map those barriers and identify  where it might be difficult for a learner to [13:41] to grow and learn and access a new tool, then  you can find ways to support that learner. [13:49] So for instance, if you have a learner who's a  multil- language learner and you provide them [13:55] with a a training document that is very textheavy.  So, what what type of AI tool can you put in place [14:02] that's going to allow this multilingual learner  to understand um this document or this training [14:09] manual or uh this resource that you've given  them in a way that that is meaningful to them. [14:16] The next is piloting within the margins. So, this  means you're actually testing your AI tools with [14:22] workers that are most likely to be excluded.  So, we're not taking these confident early [14:29] adapt um adopters and saying, "Okay, you know, we  want you to play with these tools." We are taking [14:35] the folks that we think are probably going to  have the most difficult time with these tools [14:39] and we're having them really play with it and say,  "Where are the strengths? Where are the barriers? [14:44] Um how are you using it? How would you like to  use it?" and really asking those questions because [14:49] we know that if we can get these tools adopted  with the uh folks that we consider kind of [14:55] more marginalized or excluded, then we know that  we can get these these tools to be working for [15:00] almost anyone. And so um the thought here is  that upfront investment of time helps identify [15:08] limitations early so that you're not scrambling  after the the AI tool roll out um happens and [15:15] then you see that there are some some workers that  are not able to access it. And lastly, you want [15:21] to put in guard rails. So this just means that  there's ongoing equity audits to make sure that [15:28] there aren't folks that are benefiting and then  folks that are not benefiting from these new tools [15:33] or these new um technologies that are put in  place. So essentially do not put the ladder behind [15:40] a locked door. So if you are providing access and  tools um and AI for your workers then essentially [15:49] allowing everybody to be able to use it and  access it equally is going to be your best bet. [15:56] So final thoughts here. AI changes work.  Workers need equitable learning support. [16:03] AI can expand access or deepen inequity. And then  organizations must make intentional design choices [16:11] when providing opportunities for workers  to grow and learn in the workforce. [16:19] And my final thoughts here and then I have about  three minutes for questions if anyone has any. [16:34] Yeah. [applause]