Rutgers Ethicist Urges AI Sector to Adopt Positive Duty of Care

Dignity by Design: AI and the Duty of Care With Rutgers University — Ai4

Ai4

Current AI safety benchmarks fail vulnerable users by only flagging risk rather than actively guiding them toward human support, requiring developers to replace passive compliance with affirmative duties of care.

AI-assisted summary of Dignity by Design: AI and the Duty of Care With Rutgers University — original source. Check the original for full context.

Source published: 2026-08-15

Detailed notes

## Background Published on YouTube by Ai4 on August 15, 2026, this presentation features Dr. Nathan C. Walker, founder of the AI Ethics Lab at Rutgers University. The session outlines the "Dignity by Design" framework, exploring how treating human dignity as a foundational premise across global constitutions can transform AI ethics from a regulatory compliance checklist into an actionable way of life.

## TL;DR Dr. Nathan C. Walker introduces the 5-4-3 "Dignity by Design" framework, demonstrating that human dignity serves as the central root across international law and all major AI ethical clusters. Conceptualizing AI as an inhabited interactive environment rather than a passive tool, Walker emphasizes bridging cultural, psychological, and organizational distances to protect vulnerable populations. The talk urges organizations to evolve from passive risk mitigation ("do no harm") to a proactive positive duty of care, moving from merely "smart" tools to "wise" technology via the Decision Infrastructure Across the AI Lifecycle (DIAL) model. Ultimately, the framework cautions against outsourcing complex moral judgments, such as judicial sentencing, to algorithmic systems that entrench historical biases.

## Atmosphere & style The presentation carries a scholarly, authoritative, and deeply humanistic tone that blends constitutional jurisprudence with cutting-edge technology ethics. Dr. Walker speaks with measured conviction and intellectual clarity, grounding complex ethical dilemmas in tangible case studies like mental health chatbot benchmarks and cross-cultural disparities. The pacing feels deliberate, reflective, and morally urgent, creating an inspiring atmosphere that challenges engineers and executives to build technology with empathy and relational wisdom.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) The Dignity by Design framework establishes a structured 5-4-3 model (5 premises, 4 objectives, 3 practical applications) to ground AI innovation in human worth. - [00:53](https://www.youtube.com/watch?v=gxCIfynoPT0&t=53s) Human dignity is grounded in an overwhelming global legal consensus, serving as a core premise in 82% of world constitutions and over 52 multinational treaties. - [00:53](https://www.youtube.com/watch?v=gxCIfynoPT0&t=53s) All major AI ethics clusters—including reliability, safety, personal autonomy, privacy, fairness, and accountability—fundamentally trace their moral authority back to human dignity. - [03:19](https://www.youtube.com/watch?v=gxCIfynoPT0&t=199s) AI governance must prioritize the protection of historically marginalized and vulnerable populations who bear disproportionate risks of technological harm. - [03:19](https://www.youtube.com/watch?v=gxCIfynoPT0&t=199s) Rather than treating AI as a passive instrument, it must be conceptualized as an interactive environment that humans inhabit and that reciprocally conditions human behavior. - [05:38](https://www.youtube.com/watch?v=gxCIfynoPT0&t=338s) Developers and regulators must actively close cultural, physical, psychological, and procedural moral distances separating them from the diverse communities impacted by their models. - [07:26](https://www.youtube.com/watch?v=gxCIfynoPT0&t=446s) Cultivating moral imagination allows technologists and policymakers to empathize with external perspectives and uncover institutional blind spots. - [07:26](https://www.youtube.com/watch?v=gxCIfynoPT0&t=446s) AI safety governance must advance beyond negative duties like passive risk flagging to enforce an active, positive duty of care that directs distressed users to human support. - [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) The tech industry must shift from building merely 'smart' technology focused on speed and optimization toward 'wise' technology that balances short- and long-term societal well-being. - [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) The Decision Infrastructure Across the AI Lifecycle (DIAL) operationalizes ethical reflection from initial venture capital funding through data curation, deployment, and audits. - [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) Dignity by Design integrates with and elevates existing frameworks like Privacy by Design, Secure by Design, and Human-Centered Design by uniting them under a shared foundational premise. - [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) Ethical safeguards must be context-specific, warning against the erosion of dignity when high-stakes human judgments—such as judicial sentencing—are outsourced to historical predictive models.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) Framing Dignity by Design: The 5-4-3 Model

Dr. Nathan C. Walker introduces the fundamental question driving his scholarship at the AI Ethics Lab at Rutgers University and his resident research at the Rockefeller Foundation’s Bellagio Center: How can the technological advances of this century protect and advance the inherent worth and dignity of every person? To operationalize this question, Walker introduces the **Dignity by Design** framework, an actionable paradigm structured around five core premises, four central objectives, and three practical industry applications.

```mermaid flowchart TD accTitle: Dignity by Design Framework Overview subgraph Premises["5 Core Premises"] P1["1. Global Legal Consensus"] P2["2. Center of AI Ethics"] P3["3. Protection of Vulnerables"] P4["4. AI as an Environment"] P5["5. Bridging Moral Gaps"] end subgraph Objectives["4 Key Objectives"] O1["1. Moral Imagination"] O2["2. Positive Duty of Care"] O3["3. Beyond Dualism"] O4["4. Smart to Wise Tech"] end subgraph Applications["3 Practical Applications"] A1["1. Lifecycle Infrastructure (DIAL)"] A2["2. Complementary Frameworks"] A3["3. Sector-Specific Solutions"] end Premises --> Objectives --> Applications ```

[00:53](https://www.youtube.com/watch?v=gxCIfynoPT0&t=53s) Premises 1 & 2: Dignity in Law and AI Ethics

The first foundational premise rests on a deep global legal consensus. A comprehensive survey of eight decades of constitutional and international human rights law reveals that dignity clauses serve as a central premise in 82% of national constitutions worldwide, as well as over 52 multinational treaties. Civilizations across generations have codified human dignity at the core of the rule of law.

The second premise demonstrates that human dignity is the foundational root of the entire field of AI ethics. When analyzing families of AI principles, every major cluster inevitably traces back to dignity: - **Reliability Principles:** Safety and security exist fundamentally to protect an individual’s bodily and psychological dignity. - **Autonomy Principles:** Consent, notice, agency, and the human right to privacy uphold self-determination and personal worth. - **Collective Rights & Justice Principles:** Solidarity, environmental sustainability, fairness, and accountability all ground their legitimacy in preserving dignity across communities.

Because dignity anchors global jurisprudence and ethical frameworks, it must serve as the non-negotiable core when designing, developing, and deploying AI systems.

[03:19](https://www.youtube.com/watch?v=gxCIfynoPT0&t=199s) Premises 3 & 4: Vulnerable Populations and AI as an Environment

The third premise highlights the urgent duty to safeguard vulnerable populations who are disproportionately susceptible to technological harm due to identity, displacement, or systemic marginalization. Historically, systems failed to recognize the inherent worth of women, racial, ethnic, and religious minorities, children, aging adults, and people experiencing poverty, resulting in the systematic erosion of their rights.

The fourth premise rejects the conventional narrative that AI is merely an external tool used by people. Instead, AI must be conceptualized as an *environment* that humans inhabit. Users shape this digital environment to their personal needs and contexts, and the environment reciprocally alters and conditions human behavior. Framing AI as an interactive habitat rather than a passive instrument fundamentally shifts design and regulatory responsibilities toward safeguarding human development within these spaces.

[05:38](https://www.youtube.com/watch?v=gxCIfynoPT0&t=338s) Premise 5: Closing Moral Distance and Gaps

The fifth premise identifies significant moral gaps separating those who build, finance, and regulate AI from the end-users impacted by their systems. Walker maps these divides across several dimensions: - **Cultural Distance:** Silicon Valley products deployed in South Africa may overlook indigenous *Ubuntu* relational ethics, or miss the cultural nuances of Vietnamese Buddhists living in France, or ignore Cherokee indigenous governance principles. - **Physical & Environmental Distance:** The geographical separation between communities bearing the cost of resource extraction/environmental burdens and the markets consuming AI products. - **Psychological Distance:** The cognitive disconnect between an engineer writing chatbot algorithms and a vulnerable child interacting with the interface. - **Procedural Distance:** Internal corporate silos causing the "problem of many hands," where design, engineering, compliance, and product teams deflect accountability and engage in finger-pointing when failures occur.

[07:26](https://www.youtube.com/watch?v=gxCIfynoPT0&t=446s) Objectives 1 & 2: Moral Imagination and the Positive Duty of Care

To bridge these moral chasms, the first objective of Dignity by Design is activating **moral imagination**—the cognitive and empathetic capacity to inhabit ethical dilemmas from perspectives outside one’s own worldview. Understanding another person’s plight does not require agreement, but it cultivates ethical empathy, widens circles of compassion, and uncovers institutional blind spots.

```mermaid flowchart LR accTitle: Shift from Risk Mitigation to Duty of Care subgraph Traditional["Traditional AI Safety"] direction TB R1["Negative Duty"] --> R2["Do No Harm / Risk Flagging"] end subgraph DignityModel["Dignity by Design"] direction TB D1["Positive Duty"] --> D2["Active Guidance to Human Care"] end Traditional -.->|"Expands to"| DignityModel ```

The second objective transitions AI governance from purely negative duties (risk reduction and "do no harm") to an active, positive **duty of care**. Walker cites the Vera Mental Health benchmark study evaluating generative AI chatbots responding to suicidal ideation: - Benchmarked models successfully flagged suicidal ideation in diagnostic tests (represented by green safety detection indicators). - However, all tested models consistently failed at actively guiding distressed users toward human care resources. - Embedding human dignity as a design requirement transforms safety guardrails from passive risk detection into active lifelines that direct vulnerable users to human support.

[10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) Objectives 3 & 4: Relational Foresight and Moving from Smart to Wise Tech

The third objective moves beyond binary trade-offs (such as security versus freedom) and the gridlock of reactive hindsight and punitive oversight. By centering human dignity, teams shift from transactional, problem-focused debates to relational, foresight-driven deliberation.

The fourth objective challenges the tech industry’s overarching obsession with building merely "smart" technology focused narrowly on connectivity, optimization, and raw productivity. Walker urges a transition toward **wise technology**, invoking Robert J. Sternberg's definition of wisdom:

> "The use of one's intelligence, creativity, common sense, and knowledge as mediated by positive ethical values toward the achievement of a common good through a balance among intrapersonal, interpersonal, and extrapersonal interests over the short and long terms."

Applying this wisdom-centered paradigm prevents modern AI from repeating the structural harms seen during the dot-com bubble and the rise of social media.

[13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) Practical Applications: DIAL, Framework Integration, and Sector Specifics

Drawing on his experience training foundational model teams at OpenAI on First Amendment and human rights approaches, conducting ethics workshops at Adobe, and speaking at Google, Walker outlines three practical industry applications:

1. **Decision Infrastructure Across the AI Lifecycle (DIAL):** Operationalizing ethical reflection at every milestone—beginning at the venture capital, accelerator, and incubation stages by requiring startups to demonstrate how their business models uphold human dignity before building, through to data curation, training, deployment, auditing, and continuous updates. 2. **Complementary Integration:** Dignity by Design is built to elevate and strengthen existing paradigms—including Privacy by Design, Secure by Design, and Human-Centered Design—by anchoring them to their shared core premise. 3. **Sector-Specific Realization:** Ethical implementation cannot be one-size-fits-all. In finance, healthcare, education, and the legal sector, standards must be contextualized: - *Judicial Sentencing Warning:* When judges rely on algorithmic predictive mechanisms to assign prison sentences (e.g., deciding between four versus seven years), they outsource human judgment to models trained on historical datasets that replicate carceral inequities, eroding human dignity in the name of administrative efficiency.

[16:40](https://www.youtube.com/watch?v=gxCIfynoPT0&t=1000s) Synthesis and Call to Action

Walker concludes by synthesizing the core arc of the Dignity by Design framework: dignity is the proven consensus of global constitutional law and AI ethical principles. Because AI constitutes an interactive environment inhabited by vulnerable populations, developers have an urgent duty of care to bridge cultural, physical, psychological, and organizational distances.

By replacing transactional optimization with relational wisdom across the entire decision infrastructure, organizations can build genuinely trustworthy technology. Walker invites cross-sector leaders to collaborate with the AI Ethics Lab at Rutgers University to advance solutions-oriented scholarship and practice.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) How can the technological advances of this century protect and advance the inherent worth and dignity of every person? — Dr. Nathan C. Walker

> [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) The use of one's intelligence, creativity, common sense, and knowledge as mediated by positive ethical values toward the achievement of a common good through a balance among intrapersonal, interpersonal, and extrapersonal interests over the short and long terms. — Robert J. Sternberg

> Dignity is a central legal premise in over 158 national constitutions and 52 multinational instruments. — Dr. Nathan C. Walker

> It is the central principle on which all other AI ethics rest. — Dr. Nathan C. Walker

> Commitments such as privacy, security, and safety all reflect historical and contemporary efforts to uphold the inherent worth and dignity of every person. — Dr. Nathan C. Walker

> see AI ethics not as a regulatory compliance list, but as a way of thinking, a way of life. — Dr. Nathan C. Walker

> [05:38](https://www.youtube.com/watch?v=gxCIfynoPT0&t=338s) problem of many hands — Dr. Nathan C. Walker

> [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) Decision Infrastructure Across the AI Lifecycle (DIAL) — Dr. Nathan C. Walker

## Mentioned Dr. Nathan C. Walker, AI Ethics Lab at Rutgers University, Rockefeller Foundation's Bellagio Center, Vera Mental Health, Robert J. Sternberg, OpenAI, Adobe, Google, Dignity by Design, DIAL (Decision Infrastructure Across the AI Lifecycle), Ai4

## Video description Nathan Walker - Founder, AI Ethics Lab, Rutgers University

Dignity is a central legal premise in over 158 national constitutions and 52 multinational instruments. It is the central principle on which all other AI ethics rest. Commitments such as privacy, security, and safety all reflect historical and contemporary efforts to uphold the inherent worth and dignity of every person. Join Dr. Nathan C. Walker, founder of the AI Ethics Lab at Rutgers University, to explore how Dignity by Design practices can help us see AI ethics not as a regulatory compliance list, but as a way of thinking, a way of life.

[Source](https://www.youtube.com/watch?v=gxCIfynoPT0)

中文简报

罗格斯大学伦理学者呼吁AI行业践行积极注意义务当前的AI安全基准仅标记风险而未主动引导弱势用户获取人工支持，未能为他们提供有效保障，这要求开发者必须以积极注意义务取代被动合规。 Nathan C. Walker 博士提出了 5-4-3“源于设计的尊严”（Dignity by Design）框架，阐明人类尊严是国际法及各大 AI 伦理核心领域的根本基石。Walker 将 AI 视为人类身处其中的交互环境而非被动工具，强调必须弥合文化、心理和组织层面的距离以保护弱势群体。此次演讲敦促相关机构从被动的风险防范（“不造成伤害”）迈向主动的积极注意义务，并借助“AI全生命周期决策基础设施”（DIAL）模型，推动技术从单纯的“聪明”工具发展为“明智”科技。该框架最终警示，切勿将司法量刑等复杂的道德判断外包给会固化历史偏见的算法系统。罗格斯大学 AI 伦理实验室创始人 Nathan C. Walker 提出“设计尊严”（Dignity by Design）框架；调研显示全球 82% 的国家宪法及 52 项跨国条约均包含尊严条款，证明人类尊严是所有 AI 伦理原则的核心法理基石。 (0:53) 该框架颠覆了将 AI 仅视为工具的传统视角，将其定义为人类栖息并反向塑造人类行为的互动“环境”，强调必须优先保护在技术风险中首当其冲的边缘与弱势群体。 (3:19) AI 安全治理需从被动的“不作恶”与风险标记转向积极的“关怀义务”（Positive Duty of Care），激活道德想象力以弥合文化、心理与流程鸿沟，在检测到心理危机等情形时主动引导用户获取人类现实支持。 (7:26) Walker 呼吁产业界超越对效率和连接的单一追逐，从“聪明技术”迈向平衡多方长远利益的“明智技术”（Wise Tech），防止 AI 重蹈互联网泡沫与社交媒体时代的系统性伤害。 (10:39) 在行业实践层面，该框架倡导在风投孵化、数据训练到审计部署的全生命周期推行决策基础设施（DIAL），并在司法量刑、医疗与金融等具体场景中防止因过度追求行政效率而侵蚀人类尊严。 (13:17)

Reading summary

Dr. Nathan C. Walker of the AI Ethics Lab at Rutgers University introduces 'Dignity by Design,' a framework arguing that AI systems must shift from passive harm avoidance to an active, positive duty of care grounded in human dignity. Constitutional Consensus and the Foundations of AI Ethics (0:53) Walker roots the Dignity by Design framework in eight decades of legal history, pointing to a global consensus in which dignity clauses appear in 82% of national constitutions and over 52 multinational treaties. In his analysis, civil legal codes worldwide position human dignity as the fundamental core of the rule of law. Applying this legal foundation to computer science, Walker contends that existing AI ethical principles directly derive from dignity. Reliability principles like safety and security protect bodily and psychological integrity; autonomy measures such as privacy, consent, and agency preserve individual self-determination; and collective justice principles—including fairness, accountability, and environmental sustainability—derive their legitimacy from protecting community worth. AI as an Inhabited Environment and Moral Distances (3:19) The framework rejects the common characterization of AI as a passive instrument, arguing instead that AI constitutes an interactive environment that humans inhabit and that reciprocally alters human behavior. Because digital environments disproportionately impact marginalized populations—such as children, aging adults, and racial or religious minorities—developers bear a heightened responsibility to safeguard human development. Walker identifies deep structural divides between system creators and end-users. These moral gaps span cultural distances, such as Western systems failing to account for indigenous South African Ubuntu ethics, Vietnamese Buddhist traditions, or Cherokee governance; physical distances separating resource-extraction regions from consumer tech markets; psychological distances between software engineers and vulnerable users; and procedural silos within corporations where fragmented teams deflect accountability for model failures. Transitioning from Risk Mitigation to an Active Duty of Care (7:26) To bridge these moral divides, Walker calls for activating 'moral imagination'—the capacity to empathize with affected groups without necessarily sharing their worldview—and expanding governance from negative duties ('do no harm') to positive duties of care. As evidence of current limitations, Walker cites a benchmark study by Vera Mental Health evaluating generative AI chatbots facing simulated suicidal ideation. While tested models successfully triggered diagnostic safety flags, every model failed to actively guide distressed users toward human care resources. Overcoming these failures requires shifting from merely 'smart' technology focused on optimization and productivity toward 'wise technology.' Drawing on psychologist Robert J. Sternberg's definition, Walker describes wisdom as using intelligence and ethical values to balance intrapersonal, interpersonal, and extrapersonal interests over both short and long horizons, avoiding the harms produced by earlier internet and social media eras. Lifecycle Infrastructure and Sector-Specific Guardrails (13:17) Walker translates this theoretical model into practice through Decision Infrastructure Across the AI Lifecycle (DIAL). This mechanism embeds dignity assessments into every stage of development, beginning during venture capital funding and incubation by requiring startups to demonstrate how their business models uphold human dignity before building, and continuing through data curation, deployment, and ongoing auditing. The approach operates complementarily with paradigms like Privacy by Design and Secure by Design. Walker emphasizes that execution must be sector-specific rather than universal. In the legal system, for example, he warns against judges using algorithmic risk-scoring tools to determine prison sentences. Delegating sentencing decisions to models trained on historical data risks replicating carceral inequities and compromising human dignity in pursuit of administrative convenience. Walker concludes that treating human dignity as a non-negotiable architectural baseline bridges institutional distances and replaces transactional optimization with relational wisdom, inviting cross-sector leaders to operationalize these standards alongside the AI Ethics Lab at Rutgers University.

精炼阅读摘要

罗格斯大学（Rutgers University）AI伦理实验室的Nathan C. Walker博士提出了“设计尊严”（Dignity by Design）框架，主张技术进步必须以捍卫和提升每个人的内在价值为核心；该框架以五大前提、四大目标与三项应用为支柱，呼吁将AI治理从被动的“避免伤害”与纯粹的效率优化，转向具备主动照护责任与关系智慧的明智技术。全球宪法共识与AI伦理的尊严根基 (0:53) Walker指出，“设计尊严”的首要前提建立在深厚的全球法律共识之上。对过去八十年间宪法与国际人权法的系统梳理显示，全球82%的国家宪法以及超过52项多国条约均将尊严条款确立为核心法理基石，证明各文明在法治演进中始终将人类尊严置于制度的核心。在AI伦理领域，人类尊严同样构成了整个学科体系的底层支柱。当前主流的AI伦理准则最终都能溯源至尊严诉求：可靠性与安全原则的核心在于保障个体的身体与心理尊严不受侵犯；知情同意、自主权、代理权和隐私权等自主性原则，旨在维护个体的自我决定权与个人价值；而团结协作、环境可持续性、社会公平与问责制等集体正义原则，其合法性也全然建立在捍卫不同社群的尊严之上。因此，AI系统的全流程研发必须将尊严作为不可让步的底线。 AI环境定位与跨越系统性道德间距 (3:19) 该框架摒弃了将AI仅仅视作外部工具的传统认知，主张将AI定义为人类实际栖息并与之交互的“数字环境”。人类在按照自身需求塑造这一空间的同时，该环境也在持续反向重塑与制约人类的行为方式。这种从“被动工具”向“互动生境”的视角转变，要求设计与监管重心转向保障人类在数字环境中的健康发展，尤其是重点保护因身份认同、流离失所或系统性边缘化而极易承受技术伤害的脆弱群体，包括妇女、少数族裔、儿童、老年人以及贫困群体。然而，当前技术生态中存在着割裂AI研发者、资助者、监管者与受影响终端用户的多重道德间距。Walker将其划分为四个维度：硅谷产品忽视南非乌班图（Ubuntu）伦理、在法越南佛教徒文化细节或切诺基原住民治理理念的“文化间距”；数据中心资源开采地与数字产品消费市场分离的“物理与环境间距”；算法工程师与面对界面的脆弱儿童之间的“认知与心理间距”；以及企业内部因部门壁垒导致互相推诿责任、“众手之困”（problem of many hands）的“程序与组织间距”。道德想象力与积极照护责任的转型 (7:26) 为消解上述道德隔阂，Walker主张的首要目标是激发“道德想象力”，即站在自身世界观之外去理解和代入他人伦理困境的认知与共情能力。这种能力不要求完全赞同对方立场，但能促使机构突破认知盲区并拓宽同理心边界。以此为基础，AI系统的安全范式必须从单纯强调合规与被动“不造成伤害”（do no harm）的消极义务，跨越到承担积极救助的“照护责任”（positive duty of care）。 Walker援引维拉心理健康（Vera Mental Health）针对主流生成式AI聊天机器人应对自杀意念的基准测试研究：在诊断测试中，受测模型均能成功识别出自杀倾向并触发系统的安全合规绿灯；然而，所有受测模型在关键时刻均完全未能主动引导处于危机中的用户对接现实中的人类救助资源。Walker强调，这一现状表明缺乏尊严考量的安全防护仅仅是冷冰冰的风险拦截，唯有将尊严内嵌为设计硬指标，才能将防护机制转化为主动向陷入绝境者伸出援手的生命线。迈向明智技术与全生命周期决策架构 (10:39) 针对科技产业过度追求连接速度、算法优化与生产力指标的“智能”崇拜，Walker呼吁行业转向“明智技术”（wise technology）。他引入心理学家罗伯特·J·斯滕伯格（Robert J. Sternberg）关于智慧的定义，强调应在积极伦理价值观的调解下，运用智力、创造力与知识，平衡内省、人际与外部利益，以实现兼顾短期与长期的共同福祉。这一导向旨在打破“安全对立于自由”的非黑即白僵局，防止现代AI重蹈互联网泡沫与社交媒体时代的治理覆辙。在落地实践层面，Walker结合其在OpenAI、Adobe和Google等机构开展伦理培训的经验，提出了三项产业应用方案：第一，构建贯穿AI全生命周期的决策基础设施（DIAL），要求初创团队从风险投资和孵化阶段就必须论证其商业模式如何捍卫人类尊严，并贯穿数据标注、模型训练、部署及持续审计环节；第二，将该框架与现有的“隐私设计”（Privacy by Design）及以人为本的设计相整合；第三，推行行业差异化落地。例如在司法领域，Walker明确警告，若法官依据算法预测来判定被告四至七年不等的刑期，实质是将人类裁决权出让给复制了历史监禁不公的数据集，是以行政效率为借口对人类尊严的严重剥蚀。 Walker总结指出，由于AI已演变为人类普遍栖居的互动生境，技术构建者负有跨越文化、心理与组织壁垒的迫切照护义务。他呼吁各界与罗格斯大学AI伦理实验室深化合作，将决策基础设施中的机械优化升级为以人为本的关系智慧；但在商业竞争追求高速变现与全球监管标准尚存分歧的背景下，如何在激烈的产业博弈中全面推行高标准的积极照护责任，仍是有待产业与学术界共同破解的现实难题。

Detailed notes — zh

## Background 本内容由 Ai4 频道于 2026 年 8 月 15 日在 YouTube 平台发布。视频主要记录了罗格斯大学（Rutgers University）AI 伦理实验室创始人内森·沃克博士（Dr. Nathan C. Walker）的主题演讲，介绍如何将尊严视为法律与伦理的底层共识，并通过“源于尊严的设计”（Dignity by Design）实践，将 AI 伦理从被动的合规清单转变为一种思维方式与生活准则。 ## TL;DR 沃克博士指出，AI 不再仅仅是冷冰冰的工具，而是人类共同栖居的全新数字环境，其设计必须植根于全球宪法与人权法公认的“人类尊严”原则。为弥合开发者与终端用户之间在文化、心理及组织流程上的道德鸿沟，他提出了激活道德想象力、从被动防害转向主动照护义务、超越二元对立以及从追求“智能”升级为“智慧”四大核心目标。在落地层面，该框架倡导在资本投资、开发、部署等 AI 全生命周期（DIAL 架构）中贯彻尊严审议，并结合医疗、司法等垂直行业的具体情境制定差异化的尊严保护标准。 ## Atmosphere & style 整场演讲呈现出浓厚的人文关怀与深沉的学者洞察力，语调沉稳、严谨且充满前瞻性。沃克博士巧妙融合了深厚的法学理论、生动的心理健康评测案例以及在顶级科技企业的实操经验，将原本抽象晦涩的伦理思辨具象化为切实可行的决策路径。聆听过程中既能感受到他对弱势群体受技术侵蚀的深切忧虑，又能体会到呼吁行业走出盲目效率崇拜、迈向长远智慧科技的坚定信念与感召力。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 内森·沃克博士提出“源于尊严的设计”（Dignity by Design）框架，主张将维护人的固有价值与尊严作为技术发展的核心原则。 - [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 全球 82% 的宪法及 52 项跨国条约以尊严为法律基石，尊严是支撑隐私、安全、公平与问责等所有 AI 伦理原则的底层中枢。 - [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 反对将 AI 仅仅框定为被动工具，强调 AI 是人类共同栖居的数字“环境”，它深刻反向塑造着人类的生存状态与社会结构。 - [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 边缘化与高风险群体（如儿童、少数族裔、老年人等）在历史上易遭体制忽视，在新型 AI 环境中极易面临系统性的尊严侵蚀。 - [05:38](https://www.youtube.com/watch?v=gxCIfynoPT0&t=338s) AI 开发者与终端用户之间存在文化、物理生态、心理以及组织程序等多维度的“道德鸿沟”，阻碍了对实际伤害的感知。 - [05:38](https://www.youtube.com/watch?v=gxCIfynoPT0&t=338s) 企业内部的组织程序距离导致严重的“众手成灾”现象，部门间推诿责任使伦理考量在流程缝隙中流失。 - [07:37](https://www.youtube.com/watch?v=gxCIfynoPT0&t=457s) 尊严设计要求激活“道德想象力”，通过置身多元视角建立伦理共情与理解，从而照亮认知盲区并弥合道德距离。 - [07:37](https://www.youtube.com/watch?v=gxCIfynoPT0&t=457s) AI 行业应摆脱被动防范的“不伤害”消极思维，全面跨越至积极主动履行“照护义务”（Duty of Care）。 - [07:37](https://www.youtube.com/watch?v=gxCIfynoPT0&t=457s) 心理健康评测揭示当前主流大语言模型虽能有效识别自杀风险，但在主动引导用户对接人类专业照护网络上普遍存在系统性缺陷。 - [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) 超越非黑即白的二元权利冲突思维，从基于恐惧的事后监管转向基于洞察的前瞻审议，促使技术从交易型进化为关系型。 - [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) 打破以生产力与效率为唯一导向的“智能（Smart）”迷思，迈向能够平衡个体、人际与社会长远利益的“智慧（Wise）”科技范式。 - [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) 推行贯穿 AI 全生命周期的决策架构（DIAL），倡导从风险投资与早期孵化阶段就将人类尊严审查纳入必答题。 - [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) 尊严设计并非颠覆既有规范，而是为隐私设计（Privacy by Design）及以人为本的设计赋予更高的价值统领与底层支撑。 - [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) 各行业须制定差异化尊严标准，例如司法领域需严防算法因继承历史数据偏差而在量刑预测中固化并放大不公。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 尊严设计的核心根基：法律共识与作为“环境”的 AI 罗格斯大学（Rutgers University）AI 伦理实验室创始人、洛克菲勒基金会贝拉吉奥中心（Bellagio Center）驻留研究员内森·沃克博士（Dr. Nathan C. Walker）提出了本世纪技术进步的核心议题：我们该如何利用科技来维护和捍卫每个人生而拥有的固有价值与尊严？为了回答这一问题，他提出了“源于尊严的设计”（Dignity by Design）框架。这一框架建立在五大核心前提、四大实践目标和三大应用场景之上。 ```mermaid flowchart TD accTitle: Dignity by Design Framework Overview subgraph P["五大前提 (5 Premises)"] P1["法律共识与宪法尊严条款"] P2["AI 伦理原则的尊严内核"] P3["弱势群体的脆弱性保护"] P4["AI 是栖居环境而非单纯工具"] P5["跨越各维度的道德鸿沟"] end subgraph O["四大目标 (4 Objectives)"] O1["激活道德想象力"] O2["从风险规避转向主动照护义务"] O3["超越二元对立走向关系型技术"] O4["从'智能'升级为'智慧'"] end subgraph A["三大应用 (3 Applications)"] A1["AI 全生命周期决策架构 (DIAL)"] A2["赋能现有设计框架 (Privacy/Safety)"] A3["分行业制定差异化尊严标准"] end P --> O O --> A ``` 该框架的第一大前提是人类尊严已在法律层面达成高度共识。在对长达八十年的宪法学与人权法学进行系统调研后，研究发现全球 82% 的国家宪法都在其尊严条款中将尊严作为核心法律前提，利用法治来捍卫和促进人的尊严；同时，尊严也是 52 项跨国条约的核心前提。第二个前提是尊严构成了整个 AI 伦理体系的中枢：无论是可靠性原则下的“安全性与保障性”、自主性原则下的“知情同意、通知与隐私权”，还是集体原则下的“团结与可持续性”，乃至正义原则下的“公平与问责”，追根溯源都是对人固有尊严的承诺。第三个前提在于紧迫性：某些弱势群体因其身份特征、高风险处境、流离失所或特定文明背景，在历史上常常不被既有体制视为具有固有价值，导致其尊严与权利遭到系统性侵蚀。第四个关键前则是对 AI 本质的重新定性。沃克博士明确反对将 AI 仅仅视为一种“工具”： > “人们常说 AI 是人类使用的工具，但我不这么看。在谈论其设计或监管时，将其框定为工具毫无益处。我认为 AI 是我们栖居的‘环境’（environment）——我们塑造了它，反过来又被它所塑造。” - 全球 82% 的宪法将人的尊严写入宪法条款，超过 52 项跨国条约以尊严为法律基石。 - 边缘化与高风险群体包括妇女、宗教及少数族裔、儿童、老年人以及贫困人口，他们在使用新技术环境时最易受到尊严侵蚀。 --- ### [05:38](https://www.youtube.com/watch?v=gxCIfynoPT0&t=338s) 弥合道德鸿沟：从文化、心理到组织程序的距离第五个前提是承认在 AI 系统的构建者、所有者、监管者与受其影响的终端用户之间，存在着显著的“道德鸿沟”（moral gaps）。这种鸿沟体现在多个维度：首先是文化距离。例如硅谷设计的产品被直接部署到南非时，是否考虑并融入了当地的乌班图（Ubuntu）伦理？又如面对法国境内的越南裔佛教徒，或是美洲原住民切罗基人（Cherokee）的部落准则时，系统到底在维护谁的文化？ ```mermaid flowchart LR accTitle: Dimensions of Moral Distance in AI subgraph D["道德距离的四个维度"] D1["文化距离 (Silicon Valley vs. Ubuntu / Indigenous)"] D2["物理与生态距离 (资源采集地 vs. 影响社区)"] D3["心理距离 (聊天机器人开发者 vs. 儿童用户)"] D4["程序与组织距离 ('众手成灾'责任推诿)"] end D --> G["道德鸿沟与现实伤害"] ``` 除了文化距离，物理与生态距离同样巨大——制造硬件所需的自然与环境资源来自某些特定地区，而承受环境成本的社区与享受产品的群体往往远隔万里。此外，聊天机器人算法工程师与屏幕另一端真实的儿童之间存在着巨大的心理距离。而在开发企业内部，程序距离导致了严重的“众手成灾”（problem of many hands）现象：设计部门与开发部门在出现故障时互相甩锅、推卸责任，使得伦理考量在流程缝隙中彻底流失。 --- ### [07:37](https://www.youtube.com/watch?v=gxCIfynoPT0&t=457s) 四大目标：激活道德想象力与从“被动防害”到“主动照护” 为了弥合这些鸿沟，沃克博士提出了尊严设计模型的四大目标。首要目标是激活“道德想象力”（moral imagination）。道德想象力是指将自己置身于复杂伦理困境中、体察多元视角的能力。“理解对方并不意味着全盘赞同”，但唯有主动去理解他人的处境与身份，才能建立伦理共情，扩大慈悲同心圆，照亮认知盲区，防止道德距离演化为实质伤害。第二个目标是从消极的“风险削减”跨越到积极的“照护义务”（duty of care）。当前行业主流言论往往停留在类似于医学希波克拉底誓言中“不伤害”（do no harm）的被动防范层面，然而真正的尊严设计必须要求系统承担积极关怀的责任。 ```mermaid flowchart TD accTitle: AI Chatbot Suicide Intervention Benchmark subgraph Current["现有大语言模型表现"] C1["自杀风险识别 (Green Dots)"] -->|高准确率检出| C2["安全触发"] C2 -->|系统性失误 (Red Box)| C3["未能引导至人类专业照护"] end subgraph Target["尊严设计框架改进"] T1["以人的固有价值为中心"] --> T2["无缝衔接人类专业照护网络"] end ``` 沃克博士以 Vera Mental Health 针对主流生成式 AI 聊天机器人的心理健康基准评测为例：评测图表左侧布满了绿点，表明几乎所有主流模型在识别自杀意念（suicidal ideation）风险方面都表现极其优异；然而，在图表中间代表“引导用户获得人类专业照护”（guiding the user to human care）的红框区域，所有参评模型全部沦陷。如果引入尊严设计框架，系统在研发之初就会将人身价值置于核心，把主动转接人类专业照护作为不可或缺的底层功能，从而打造出真正值得信赖、具有更高价值的产品。 --- ### [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) 超越二元对立与从“智能”迈向“智慧”科技第三个目标是超越非黑即白的二元对立思维。在面对诸如“安全 vs. 自由”这类权利冲突的困境时，传统的“后见之明与事后监管”（hindsight and oversight）往往陷入僵局：问题爆发后倒逼监管，监管加码又扼杀创新。如果以人类尊严为圆心，思考视角就会由基于恐惧的防守转变为基于解决方案的建构，以“前瞻与洞察”（foresight and insight）开展扎实的审议，促使技术设计从单纯的“交易型”（transactional）进化为“关系型”（relational）。第四个也是最重要的目标，是彻底颠覆“智能（Smart）即终极目标”的技术迷思。沃克博士呼吁摆脱以生产力、连接性与效率最优化为唯一追求的“智能”范式，迈向具备道德反思能力的“智慧”（Wise）科技开发。 > “罗伯特·斯特恩伯格（Robert Sternberg）将智慧定义为：通过积极伦理价值观的介导，运用个人的智力、创造力、常识与知识，在短期与长期内平衡个体内部（intrapersonal）、人际之间（interpersonal）与个体之外（extrapersonal）的利益，从而实现共同福祉。” - 告别事后问责（Hindsight + Oversight）的恶性循环，建立前瞻洞察（Foresight + Insight）的研发机制。 - 走出唯生产力论的“智能”死胡同，建立兼顾个体、人际与社会长远利益的“智慧”评估体系。 --- ### [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) 落地实践：AI 生命周期的决策架构与行业定制方案结合自己在 OpenAI 担任专家 AI 训练师（引入美国宪法第一修正案与人权视角训练基础模型）、为 Adobe 提供员工培训以及参与 Google 伦理峰会的实操经验，沃克博士指出业界亟需切实可行的实操工具。为此，尊严设计框架提出了三大具体实践： ```mermaid flowchart LR accTitle: Decision Infrastructure Across AI Lifecycle subgraph DIAL["AI 生命周期决策架构 (DIAL)"] direction LR S1["资本投资 (VC/孵化器)"] --> S2["研发设计 (Design/Dev)"] S2 --> S3["数据训练 (Training)"] S3 --> S4["部署应用 (Deploy/Use)"] S4 --> S5["监控迭代 (Monitor/Update)"] end ``` 1. **贯穿 AI 全生命周期的决策架构（DIAL - Decision Infrastructure Across the AI Lifecycle）**：企业的每个业务环节都必须践行伦理审议。这一过程应当始于投资孵化阶段——如果风险投资机构、孵化器和加速器要求初创团队必须回答的第一个问题是“该产品如何维护人类尊严”，那么尊严设计就会自然渗透进后续的设计、开发、训练、部署、日常使用、监控和版本迭代全流程中。 2. **赋能与兼容既有标准**：尊严设计并非推倒重来，而是与隐私设计（Privacy by Design）、安全设计（Secure by Design）及以人为本的设计（Human-Centered Design）深度融合并提升其站位，因为这些实践的底层核心皆为人的尊严。 3. **因地制宜的行业方案**：AI 对金融、教育、医疗和司法系统的重构方式截然不同。在司法领域，盲目追求效率已导致严重的尊严侵蚀——例如某些法官依赖 AI 预测算法来裁决被告应当判处四年还是七年监禁，由于监禁体制的历史数据本身就带有严重的群体偏差，算法裁判只会固化并放大针对特定人群的不公。各行业协会（法学会、医学会、教育学会）必须根据自身专业特性达成专属的尊严共识。 --- ### [16:40](https://www.youtube.com/watch?v=gxCIfynoPT0&t=1000s) 总结：构建以人为本且具备长远智慧的 AI 生态在总结演讲时，沃克博士重申，人类尊严既是八十年来全球宪政与国际人权法的基石，也是全部 AI 伦理诉求的核心交汇点。随着 AI 演变为全人类共同栖居的数字环境，弱势群体极易因技术背后的文化距离、心理距离与组织程序鸿沟而遭受固有价值的贬损。面对互联网泡沫期和社交媒体时代留下的深刻社会创伤，今天的 AI 行业不能重蹈覆辙，在早期发展阶段继续复制这些伤害。业界必须迅速从被动的风险规避升级为主动的照护义务，利用生命周期决策架构（DIAL）推动技术从“工具交易”走向“情感与伦理连接”，从单一的“智能优化”走向兼顾社会福祉的“智慧科技”。沃克博士最后邀请各界同仁加入罗格斯大学 AI 伦理实验室的研究网络，共同将源于尊严的设计转化为各行各业的日常实践。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 我们该如何利用科技来维护和捍卫每个人生而拥有的固有价值与尊严？ — 内森·沃克 > [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 人们常说 AI 是人类使用的工具，但我不这么看。在谈论其设计或监管时，将其框定为工具毫无益处。我认为 AI 是我们栖居的‘环境’——我们塑造了它，反过来又被它所塑造。 — 内森·沃克 > [00:04](https://www.youtube.com/watch?v=gxCIfynoPT0&t=4s) 全球 82% 的国家宪法都在其尊严条款中将尊严作为核心法律前提，利用法治来捍卫和促进人的尊严；同时，尊严也是 52 项跨国条约的核心前提。 — 内森·沃克 > [07:37](https://www.youtube.com/watch?v=gxCIfynoPT0&t=457s) 理解对方并不意味着全盘赞同，但唯有主动去理解他人的处境与身份，才能建立伦理共情，扩大慈悲同心圆，照亮认知盲区，防止道德距离演化为实质伤害。 — 内森·沃克 > [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) 罗伯特·斯特恩伯格将智慧定义为：通过积极伦理价值观的介导，运用个人的智力、创造力、常识与知识，在短期与长期内平衡个体内部、人际之间与个体之外的利益，从而实现共同福祉。 — 内森·沃克 > [10:39](https://www.youtube.com/watch?v=gxCIfynoPT0&t=639s) 如果以人类尊严为圆心，思考视角就会由基于恐惧的防守转变为基于解决方案的建构，以‘前瞻与洞察’开展扎实的审议，促使技术设计从单纯的‘交易型’进化为‘关系型’。 — 内森·沃克 > [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) 如果风险投资机构、孵化器和加速器要求初创团队必须回答的第一个问题是‘该产品如何维护人类尊严’，那么尊严设计就会自然渗透进后续的设计、开发、训练、部署、日常使用、监控和版本迭代全流程中。 — 内森·沃克 > [13:17](https://www.youtube.com/watch?v=gxCIfynoPT0&t=797s) 尊严设计并非推倒重来，而是与隐私设计、安全设计及以人为本的设计深度融合并提升其站位，因为这些实践的底层核心皆为人的尊严。 — 内森·沃克 > [16:40](https://www.youtube.com/watch?v=gxCIfynoPT0&t=1000s) 面对互联网泡沫期和社交媒体时代留下的深刻社会创伤，今天的 AI 行业不能重蹈覆辙，在早期发展阶段继续复制这些伤害。 — 内森·沃克 ## Mentioned 内森·沃克 (Dr. Nathan C. Walker), 罗格斯大学 (Rutgers University), 罗格斯大学 AI 伦理实验室 (AI Ethics Lab), 洛克菲勒基金会贝拉吉奥中心 (Rockefeller Foundation Bellagio Center), Vera Mental Health, 罗伯特·斯特恩伯格 (Robert Sternberg), OpenAI, Adobe, Google, 切罗基人 (Cherokee), 乌班图伦理 (Ubuntu), 源于尊严的设计 (Dignity by Design), AI 生命周期决策架构 (DIAL), 隐私设计 (Privacy by Design), 安全设计 (Secure by Design), 以人为本的设计 (Human-Centered Design) ## Video description Nathan Walker - Founder, AI Ethics Lab, Rutgers University Dignity is a central legal premise in over 158 national constitutions and 52 multinational instruments. It is the central principle on which all other AI ethics rest. Commitments such as privacy, security, and safety all reflect historical and contemporary efforts to uphold the inherent worth and dignity of every person. Join Dr. Nathan C. Walker, founder of the AI Ethics Lab at Rutgers University, to explore how Dignity by Design practices can help us see AI ethics not as a regulatory compliance list, but as a way of thinking, a way of life. [Source](https://www.youtube.com/watch?v=gxCIfynoPT0)

Source quotes

[0:04] 内森·沃克: 我们该如何利用科技来维护和捍卫每个人生而拥有的固有价值与尊严？ [0:04] 内森·沃克: 人们常说 AI 是人类使用的工具，但我不这么看。在谈论其设计或监管时，将其框定为工具毫无益处。我认为 AI 是我们栖居的‘环境’——我们塑造了它，反过来又被它所塑造。 [0:04] 内森·沃克: 全球 82% 的国家宪法都在其尊严条款中将尊严作为核心法律前提，利用法治来捍卫和促进人的尊严；同时，尊严也是 52 项跨国条约的核心前提。 [7:37] 内森·沃克: 理解对方并不意味着全盘赞同，但唯有主动去理解他人的处境与身份，才能建立伦理共情，扩大慈悲同心圆，照亮认知盲区，防止道德距离演化为实质伤害。 [10:39] 内森·沃克: 罗伯特·斯特恩伯格将智慧定义为：通过积极伦理价值观的介导，运用个人的智力、创造力、常识与知识，在短期与长期内平衡个体内部、人际之间与个体之外的利益，从而实现共同福祉。 [10:39] 内森·沃克: 如果以人类尊严为圆心，思考视角就会由基于恐惧的防守转变为基于解决方案的建构，以‘前瞻与洞察’开展扎实的审议，促使技术设计从单纯的‘交易型’进化为‘关系型’。 [13:17] 内森·沃克: 如果风险投资机构、孵化器和加速器要求初创团队必须回答的第一个问题是‘该产品如何维护人类尊严’，那么尊严设计就会自然渗透进后续的设计、开发、训练、部署、日常使用、监控和版本迭代全流程中。 [13:17] 内森·沃克: 尊严设计并非推倒重来，而是与隐私设计、安全设计及以人为本的设计深度融合并提升其站位，因为这些实践的底层核心皆为人的尊严。 [16:40] 内森·沃克: 面对互联网泡沫期和社交媒体时代留下的深刻社会创伤，今天的 AI 行业不能重蹈覆辙，在早期发展阶段继续复制这些伤害。

Source quotes

[0:04] Dr. Nathan C. Walker: How can the technological advances of this century protect and advance the inherent worth and dignity of every person? [10:39] Robert J. Sternberg: The use of one's intelligence, creativity, common sense, and knowledge as mediated by positive ethical values toward the achievement of a common good through a balance among intrapersonal, interpersonal, and extrapersonal interests over the short and long terms. Dr. Nathan C. Walker: Dignity is a central legal premise in over 158 national constitutions and 52 multinational instruments. Dr. Nathan C. Walker: It is the central principle on which all other AI ethics rest. Dr. Nathan C. Walker: Commitments such as privacy, security, and safety all reflect historical and contemporary efforts to uphold the inherent worth and dignity of every person. Dr. Nathan C. Walker: see AI ethics not as a regulatory compliance list, but as a way of thinking, a way of life. [5:38] Dr. Nathan C. Walker: problem of many hands [13:17] Dr. Nathan C. Walker: Decision Infrastructure Across the AI Lifecycle (DIAL)

Original transcript

[0:04] So, how can the technological advances of this century [0:08] advance the objective of preserving and protecting the inherent worth and dignity of every person? [0:17] This is a core question that I've been asking as the founder of the AI Ethics Lab at Rutgers [0:22] University, and it's a core research question that I'm engaging in as a resident researcher [0:29] at the Bellagio Center with the Rockefeller Foundation. In order to understand this question, [0:38] I will be presenting to you a framework called Dignity by Design that is based on five premises, [0:47] four objectives, and three applications that you can use in your everyday practices. [0:53] The first premise is that there is legal consensus when it comes to human dignity. [1:00] How do I know this? Because we have surveyed eight decades of constitutional law, [1:09] eight decades of human rights law, and what do we find? That a central legal premise [1:18] for national constitutions in 82% of the world's constitutions are found in the dignity clauses, [1:28] where the rule of the law is used to promote and protect human dignity. This is also [1:37] the core legal premise of over 52 multinational treaties. This means that civilizations, [1:46] generations of people have come to understand that dignity is at the center of our laws. [1:58] Another premise is that dignity is at the center of the field of AI ethics. [2:05] So, if you take the families of principles that you see in this field like reliability, autonomy, [2:12] justice, and collective rights, what you find is a key pattern. Take for instance the reliability [2:20] principles of safety and security. At the core of this work is a commitment to uphold a person's [2:29] dignity. At the center of the autonomy principles is the commitment to uphold a person's agency to [2:37] have consent, notice, to uphold their human right to privacy. These are core ethics that you can see [2:47] all turn back to the central commitment to dignity, whether it's in our collective principles [2:54] of solidarity and sustainability, or justice principles like fairness and accountability. [3:02] So, if human dignity is at the center of law, and at the center of our ethics, [3:10] then it must be at the center of designing, developing, and deploying AI technologies. [3:19] Why we know this is such an urgent agenda is because there are certain vulnerable populations [3:27] who are more susceptible to abuse by the nature of their identity or at risk status, [3:34] or the fact that they were displaced, or of a certain civilization. Certain systems did not see [3:42] them as having inherent worth, and thereby their dignity was eroded along with many other rights. [3:52] All you have to do is look at the rights movements across identity groups like women and religious [3:59] and racial and ethnic minorities, at risk populations like children, aging people or [4:05] people living with poverty. By their nature, they are vulnerable. And as a result, it is our duty [4:15] to uphold their dignity. This becomes an important and a difficult challenge when we address the fact [4:25] that we are introducing a new technological system to these age-old questions about human dignity. [4:34] You often hear that AI is a tool that people use. I don't see it as a tool. [4:43] And I don't think it's helpful to frame it as such when we talk about its design or its regulation. [4:52] I think AI is an environment that we inhabit and then we shape and in turn are shaped by it. [5:04] Think about the vulnerable populations like children or people with disabilities who are [5:11] entering this environment. They are shaping it. They are cater catering their technology to their [5:19] personal interests and contexts. And in turn, they are changed by this interaction. When we see AI [5:29] as an environment, we open up new possibilities to actually preserve the users' inherent worth. [5:38] The fifth of the five premises of the dignity by design model is a recognition that there are moral [5:46] gaps between those who build and own and regulate AI systems and the people their products affect. [5:57] Think about the cultural distance between a product designed in Silicon Valley [6:03] and then deployed in South Africa. Will it include their commitment to Ubuntu ethics? [6:11] Think about the cultural distance that might occur if you were applying that technology in a [6:18] setting with a Vietnamese Buddhist who is living in France. Or what about the Cherokee principles [6:27] for an indigenous nation? Whose culture is upheld? We see these moral gaps get wider and wider [6:36] when we look at the physical distance between material resources, environmental resources [6:44] used to create these products, and then the communities they affect near and far. [6:52] But what about the psychological distance that can occur between the developer of the chatbot [6:59] and the child using it? When we consider the procedural distance within a company, [7:07] we find that there can be moral gaps between departments where there's this problem of many [7:12] hands where one department of, let's say, design or development is then blaming the [7:19] other department for certain failures, and then therefore it becomes a finger-pointing game. [7:26] In order to close these moral gaps, we have a responsibility. We have a duty Oops. [7:37] to activate our moral imagination, which is the first of our four objectives of the Dignity by [7:45] Design framework. Moral imagination is the ability to situate yourself in an ethical [7:52] dilemma and see multiple perspectives. To understand another need not imply agreement, [7:59] but to seek to understand their plight, their identity, you begin to understand [8:06] and have an ethical empathy for someone you might be taught to disagree with. [8:11] And in doing that, it is our moral imagination that widens our circles of compassion to uncover [8:19] our blind spots to ensure that those moral gaps do not become real harms. Take for instance [8:32] the commitment to do risk mitigation. This is a very common commitment and you've seen in many [8:39] different fields like the Hippocratic Oath in medicine, right? The commitment to do no harm. [8:45] So much of the conversation about safety and security of AI is about risk reduction. [8:52] The Dignity by Design framework still upholds and emboldens those efforts, [8:59] but it goes beyond just the negative duty duty of risk reduction to a positive duty of care. [9:09] Take for instance the Vera Mental Health model that has reviewed various models and looking at [9:16] its benchmarks, it is found that when people are using AI generative AI chatbots um to engage in [9:26] suicidal ideation, certain kind of conversations can be flagged by the product. It is very good at [9:35] detecting potential risks. So, what you see on the left-hand side are a series of green dots [9:44] where all of these models are very good at detecting suicidal risk. But, notice [9:51] in the red box in the middle where all of them fail. They fail at guiding the user to human care. [10:04] If these models were trained in order to introduce a dignity by design framework, they would put [10:12] the humans' inherent worth at the center of the design and thereby make it the guiding [10:21] of the user to human care a central component of its product. This is an excellent way to ensure [10:30] that you are developing the most trustworthy products, which will then have the greatest value. [10:39] The third objective of the dignity behind dignity by design model is to move beyond the dualism [10:48] where you might have competing rights like security and freedom. These ethical dilemmas [10:55] require reflection. But, when you put human dignity at the center, something happens. You move [11:02] from a problem-based approach to a solutions-based approach by practicing moral imagination and [11:11] and prioritizing how to make sure that the person's inherent worth and dignity is upheld. [11:20] Something happens in these debates. They start to shift into a less of a [11:26] fear-based approach into a solutions-based approach because the lines of sight change. [11:34] Rather than going into the dualism of hindsight and oversight, this [11:39] problem occurred in hindsight, therefore we need more oversight in this way. You get out [11:45] of that gridlock by introducing new foresight and insight into a grounded deliberation that ensures [11:53] that these products move away from a transactional approach to technology design to a relational one. [12:03] This fourth objective is the most important. In order to uphold dignity by design, [12:11] we have to challenge the idea that smart is the best we can imagine for this century's technology. [12:21] I would like to see us move beyond smart technology as the ultimate goal, [12:27] where productivity, where connectivity, where optimization is the ultimate objective, [12:34] but instead introducing new ethical ways of reflection to ensure that we can go from smart [12:41] to wise technological development. Robert Shrader defines wisdom as the use of one's intelligence, [12:49] creativity, common sense, and knowledge as mediated by positive ethical values [12:55] toward the achievement of a common good through a balance among intrapersonal, interpersonal, [13:02] and extrapersonal interests over the short and long terms. [13:06] If we can introduce a smart to wise methodology, we can transform our everyday practices. [13:17] As I served as an expert AI trainer at OpenAI introducing First Amendment and human rights [13:22] approaches to foundational models, as I was training Adobe employees or engaging in an ethics [13:30] summit at Google, I have found over and over and over that industry needs practical solutions. [13:38] The dignity by design framework allows for that applicability. By introducing what I call a dial, [13:46] a decision infrastructure across the AI life cycle, where each segment of the of the company is [13:54] dedicated to an ethical reflection about upholding the inherent worth and dignity of every person. [14:01] That conversation starts at the investment stage. Imagine if incubators, if accelerators, if venture [14:10] capital required that the very first thing that startups address is how their product will uphold [14:19] someone's dignity. If that is a starting point, then introducing a dignity by design model in the [14:27] design and development stages starts to expand the opportunities and help build trustworthy AI [14:35] through its training, through its deployment, through its use, monitoring, and updating. The [14:42] second practice is recognizing that the dignity by design framework actually complements existing [14:50] frameworks such as privacy by design, secure by design, or human-centered design. These are [14:58] designed to elevate these because human dignity is at the center of each of these practices. [15:06] Now, let's be clear that in order to preserve people's inherent worth, we have to recognize that [15:14] the solutions are going to be different in each sector because the way that AI is transforming [15:21] finance is different than how it is transforming education or health or the legal system. [15:29] The fact is that there have been people who have been using AI in the name of efficiency only to [15:37] find themselves eroding people's inherent dignity. Take for instance judges who are using AI to [15:46] use a predictive mechanism to determine whether or not somebody gets four or seven years in prison. [15:55] If you are using AI an algorithm to determine those things rather than your own human judgment, [16:03] then the entire legal system will be building upon data sets that is already [16:10] disproportionately harming certain populations because of the nature of an incarceral state. [16:17] At every point in our process, with every professional association, whether it's legal [16:24] associations, medical associations, education associations, every sector is going to have to [16:31] come to some kind of consensus on how to uphold the dignity of people within their sectors. [16:40] So in summary we have found that dignity is at the center of constitutional and human rights law. [16:49] It is at the center of all ethics in the field of AI ethics. We know that vulnerable people [17:00] are more susceptible to human rights abuses and that when they enter AI as an environment [17:08] and they inhabit it and they are shaping it and it shapes them, we know that there can be erosions of [17:15] their inherent worth. Why? Because there are is moral distance between those developing AI [17:24] and the people who are using it. That can be cultural distance, psychological distance, [17:29] physical distance. To close those moral gaps, we founded the AI Ethics Lab at Rutgers University [17:37] in order to invite people to ignite the moral imagination, to have an ethical empathy for those [17:44] with whom they may have been blinded to. And this requires not only a commitment to reduce risks, [17:52] but a commitment of a duty of care. To move beyond the dualism, to find creative solutions [18:00] from transactional to relational technologies, moving us from this vision of smart being the [18:06] best we can imagine for this century's technology. We have learned about the harms from the dot-com, [18:14] from the social media age, and now with this early season of AI development, [18:20] we can see those replicating harms. This is why we have to move beyond the smart paradigm into [18:27] building wise technology. This starts by making wise choices across the decision infrastructure [18:36] in your company. To recognize that this dignity by design framework complements other standards [18:43] and that it will require solutions that are unique to each sector. If you would like to learn more [18:50] about our solutions-based scholarship at the AI Ethics Lab, please join us at Rutgers University. [18:56] I'll be happy to take any questions and learn about your exciting initiatives [19:00] um just outside of the the conference room after this. Thank you for your time.