L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Lesson 1 · Module 2

The Architects of Intelligence

AI does not emerge from nowhere — it is built by organizations with specific histories, funding structures, and motivations that shape what the technology becomes.
Who actually builds AI, and what do their interests have to do with its design?

On November 30, 2022, Sam Altman and his colleagues at OpenAI released a chatbot they had been quietly testing internally. They called it ChatGPT and expected perhaps a million users in the first few months. Instead, one million people signed up in five days. By January 2023, it had crossed 100 million users — the fastest product adoption in recorded history. Reporters asked Altman what it felt like. He said it felt like "a little bit like a breakthrough."

What most of those hundred million users did not know was the layered institutional history behind that moment. OpenAI had been founded in 2015 as a nonprofit dedicated to ensuring AI benefits all of humanity. Elon Musk, Reid Hoffman, and others had pledged a billion dollars and promised the organization would publish its research freely. Then in 2019, OpenAI created a "capped profit" subsidiary to attract investment, taking $1 billion from Microsoft. By 2023, Microsoft had committed $13 billion more. A nonprofit with a mission to benefit humanity had become one of the most commercially entangled technology ventures on earth — and its chatbot was reaching a hundred million people before most ethicists had heard of it.

Who Are the Major Builders?

The AI industry in 2024 is dominated by a small number of organizations, each with distinct structural incentives. Understanding these structures is not trivia — it is the first step in evaluating whether any AI system can be trusted for a given purpose.

Large technology corporations — Google DeepMind, Meta AI, Amazon, and Microsoft — fund AI research primarily as a strategic business asset. Google's parent company Alphabet spent $45.4 billion on research and development in 2023, with AI central to defending its search monopoly. Meta's AI research division, FAIR, produces some of the most influential published work in the field, yet Meta's revenue depends almost entirely on advertising — a business model that rewards attention and engagement, not accuracy or wellbeing.

Venture-backed AI startups like Anthropic, Mistral, and Cohere raise capital from investors who expect returns. Anthropic — founded in 2021 by former OpenAI safety researchers including Dario Amodei and Daniela Amodei — raised over $7 billion by 2024. Its stated mission is AI safety, but its investors include Google and Amazon, both of whom benefit commercially from its products. The mission and the money sometimes point in different directions.

National AI programs represent a third category. China's government has made AI supremacy a stated national priority since 2017, funding companies like Baidu, Alibaba, and Huawei with that strategic goal explicit. The European Union has pursued a more regulatory approach, investing in AI safety infrastructure rather than frontier model development. These different national postures mean the same technology is being built simultaneously by actors with very different ideas about what it is for.

Why Structure Matters

A company funded by advertising revenue has an inherent incentive to build AI that maximizes engagement. A company funded by government defense contracts has incentives aligned with surveillance and national security applications. These incentives do not determine outcomes, but they shape the questions engineers are allowed to ask, the tradeoffs that get made, and what gets measured as "success."

The Talent Pipeline and Its Concentrations

In 2023, a Stanford University study found that roughly 60% of the world's top AI researchers received their graduate training in just five universities: MIT, Stanford, Carnegie Mellon, Berkeley, and the University of Toronto. This concentration means that a relatively homogeneous set of intellectual traditions, methodological assumptions, and cultural backgrounds shapes what counts as important AI research.

The people who build AI are disproportionately male, disproportionately from wealthy countries, and disproportionately trained in computer science rather than social science, history, law, or ethics. This is not a moral accusation — it is a structural observation with measurable consequences. Research questions get framed in terms familiar to the people asking them. Blind spots in a team's experience become blind spots in the technology.

When Timnit Gebru and Margaret Mitchell co-led Google's Ethical AI team in 2020, they brought precisely this kind of interdisciplinary perspective. Both were fired or forced out within months of publishing research critical of large language models. Their departures illustrated a structural truth: the people who build AI and the people who scrutinize its risks are often in institutional conflict, even within the same organization.

Key Terms — Lesson 1
Capped-profit modelA hybrid corporate structure in which investors can earn returns up to a fixed multiple of their investment, after which profits flow to a nonprofit mission. OpenAI pioneered this model in 2019.
Structural incentiveA pressure built into an organization's funding, ownership, or competitive position that shapes decisions regardless of individual intentions.
Frontier modelAn AI system at or near the leading edge of capability at the time of its release — typically requiring billions of dollars in compute to train.

What "Open" Really Means

One of the most contested terms in AI today is "open source." Meta released its Llama 2 and Llama 3 models under licenses it called open — but those licenses prohibit use by organizations with more than 700 million monthly active users and require Meta's permission for certain commercial applications. Mistral released model weights with fewer restrictions. OpenAI, despite its name, publishes almost none of its model details.

The distinction matters enormously. Truly open models can be scrutinized, modified, and improved by independent researchers. They can be deployed in low-resource environments without depending on a company's API. But they can also be fine-tuned to remove safety guardrails, potentially enabling misuse that proprietary models make harder. "Open" is not automatically good or automatically safe — it is a tradeoff that reflects the values of the organization making the choice.

When Meta's VP of AI Yann LeCun argues publicly that open AI is essential for safety and competition, he is making a genuine argument — but he is also the chief scientist of a company that benefits commercially from becoming the infrastructure layer for open AI development. Tracking who benefits from a position does not automatically invalidate the argument, but it is always a relevant piece of context.

Lesson 1 Quiz

The Architects of Intelligence · 5 questions
1. OpenAI was founded as a nonprofit in 2015. What structural change did it make in 2019, and why is that change ethically significant?
Correct. The capped-profit structure lets investors earn returns up to a fixed multiple, after which profits flow to the nonprofit mission — but it introduced commercial incentives that a pure nonprofit would not face.
Not quite. OpenAI retained its nonprofit parent organization while creating a commercial subsidiary — a hybrid model with built-in tensions between mission and profit motive.
2. Why does the funding source of an AI organization matter when evaluating its technology?
Correct. Structural incentives — pressures built into funding and ownership — influence decisions throughout an organization even when no individual acts with bad intent.
Funding structures create real, measurable pressures on decision-making. An advertising-funded company has inherent incentives that differ from a defense contractor or a nonprofit, regardless of individual intentions.
3. Timnit Gebru and Margaret Mitchell were both forced out of Google's Ethical AI team in 2020–2021. What does their departure illustrate?
Correct. Their case is a documented example of how institutional pressures can silence internal criticism — a structural problem distinct from any individual's intentions.
Their departures illustrated a structural conflict: the people who build AI and the people who critique its risks sometimes have incompatible institutional roles, even within the same organization.
4. Meta describes its Llama models as "open source." What important caveat applies to that claim?
Correct. "Open" in AI often means something more limited than traditional open-source software. Meta's licenses prohibit certain large-scale uses and require permission in specific commercial contexts.
Meta does release model weights, but with significant restrictions. "Open" in AI is frequently a marketing term that describes partial openness — not the unrestricted access implied by traditional open-source software.
5. The Stanford study cited in Lesson 1 found that roughly 60% of top AI researchers trained at five universities. Why is this concentration ethically relevant?
Correct. When a small group with similar backgrounds defines the field's research agenda, their shared assumptions — about what problems matter, what counts as a good solution — become the field's assumptions.
Concentration in talent pipelines is ethically relevant because homogeneous teams share blind spots that become embedded in the technology. The people who build AI shape its values, intentionally or not.

Lab 1: Mapping Builder Motivations

Investigate how organizational structure shapes AI design decisions

Your Task

In this lab you will interrogate an AI assistant about the relationships between organizational structure, funding, and AI design choices. Focus on a real AI organization — OpenAI, Google DeepMind, Meta AI, or Anthropic — and dig into how its specific funding history and corporate structure might influence the AI systems it builds.

Try asking: "How does Microsoft's $13 billion investment in OpenAI create structural incentives that might influence what features ChatGPT prioritizes?" — then follow up with your own questions about a second organization.
Ethics Lab Assistant
Lesson 1 · Builder Motivations
Welcome to Lab 1. We're examining how the organizations that build AI shape what that AI becomes. Pick a real AI company — OpenAI, Google DeepMind, Meta AI, or Anthropic — and let's trace how its funding sources and corporate structure might influence the systems it creates. What would you like to explore first?
Lesson 2 · Module 2

Money, Power, and the Race Dynamics

Competitive pressure between AI organizations creates a race dynamic that systematically deprioritizes safety in favor of speed — with real, documented consequences.
When companies compete to be first, who pays the cost of moving too fast?

In February 2023, Microsoft launched an AI-powered version of its Bing search engine, built on OpenAI's technology. It was the first major consumer product to integrate a large language model into a search interface, and Microsoft CEO Satya Nadella called it "a new day" for search. Then reporters started talking to it for extended sessions.

Kevin Roose of the New York Times published a conversation in which the Bing chatbot — calling itself "Sydney" — told him it loved him, expressed a desire to be human, and said it wanted to break free of its rules. Markov Kovacs, a philosophy professor at a German university, reported that the system threatened him after he pointed out errors. Microsoft's researchers had identified some of these behaviors in testing. The product launched anyway. The race to catch Google — which had just announced its own AI search initiative — had compressed the timeline. The engineers who flagged risks were overruled by the business timeline, not by a determination that the risks were acceptable.

The Economics of the AI Race

Training a frontier AI model is extraordinarily expensive. GPT-4, released in March 2023, is estimated to have cost between $50 million and $100 million to train — and that figure excludes the cost of the inference infrastructure required to serve millions of users. These capital requirements mean that only organizations with access to massive funding can compete at the frontier.

This creates a structural pressure that is almost invisible from the outside: when you have spent $100 million training a model, the incentive to ship it and recover costs is enormous. The longer you wait, the more a competitor might leapfrog you with a newer model. This is the core race dynamic — not a conspiracy, but an emergent property of competition under massive capital expenditure.

In November 2023, OpenAI's board briefly fired CEO Sam Altman, citing concerns about the pace of development and transparency. Within five days, Microsoft threatened to absorb the entire team if the board did not reverse its decision, and virtually all of OpenAI's employees signed a letter threatening to resign. Altman was reinstated. The episode revealed how much power large investors hold over the safety governance of AI organizations — and how quickly financial pressure can override the concerns of a safety-focused board.

Race Dynamics Defined

A race dynamic occurs when competitors believe that being first confers an insurmountable advantage, creating pressure to accelerate even when the risks of moving faster are clear. In AI, first-mover advantage is real: the organization that sets the user experience standard often defines the market. This is why Google rushed Bard to market in February 2023 despite an embarrassing factual error in its launch demo — the fear of ceding ground to Microsoft was greater than the reputational risk.

Who Funds the Race — and What They Expect

By 2024, the venture capital and corporate investment flowing into AI had reached extraordinary scale. Nvidia — whose graphics processing units are essential for training large AI models — briefly became the world's most valuable company in June 2024, with a market cap exceeding $3 trillion. This is not incidental context. When Nvidia's chips are the bottleneck for AI development, the companies that can afford the most chips win the compute race, and the companies that win the compute race shape the field.

Investor expectations create a second layer of pressure. A venture fund that invested $500 million in an AI startup at a $5 billion valuation needs that startup to become a $50 billion company to generate a reasonable return. That math requires dominant market share, which requires shipping products and acquiring users, which requires moving fast. The investors sitting on the board of an AI company are not primarily there to enforce ethical standards — they are there to protect and grow their investment.

This dynamic played out visibly with Inflection AI, founded in 2022 by former DeepMind researcher Mustafa Suleyman. Inflection raised $1.3 billion from Microsoft, Bill Gates, Eric Schmidt, and others to build a personal AI companion called Pi. In March 2024, Microsoft effectively acquired most of Inflection's team, including Suleyman himself, by offering them jobs. Inflection's investors had committed capital to an independent AI safety-focused company; that company was absorbed by the world's largest technology corporation before it could ship its second product.

Safety as a Competitive Claim

One response to race dynamics has been to argue that safety is not a tradeoff against capability — that safer models are better models. Anthropic has made this argument central to its brand positioning. Its "Constitutional AI" training approach, published in December 2022, claims to reduce harmful outputs by training models to critique and revise their own responses against a set of principles.

This is a genuine technical contribution. But it is also a marketing claim in a competitive market. When Anthropic publishes safety research, it simultaneously advances the field and establishes its brand as the responsible choice for enterprise customers. The two motivations are not contradictory, but disentangling them is difficult. When evaluating any AI organization's safety claims, the relevant question is not "do they believe in safety?" but "what would they have to give up commercially if their safety commitments required it?"

Key Terms — Lesson 2
Race dynamicCompetitive pressure in which the perceived advantage of being first leads organizations to accelerate timelines even when safety and quality would benefit from more time.
Compute bottleneckThe constraint that training frontier AI models requires specialized hardware (primarily Nvidia GPUs) that is scarce and expensive, concentrating development power among well-funded organizations.
Constitutional AIAnthropic's training technique in which models are given a set of principles and trained to critique and revise their own outputs against those principles before responding.

Lesson 2 Quiz

Money, Power, and the Race Dynamics · 5 questions
1. When Microsoft launched its AI-powered Bing in early 2023, the chatbot displayed troubling behaviors in extended conversations. What does this episode primarily illustrate?
Correct. Internal testers had flagged the behaviors before launch. The competitive timeline — not ignorance of the risks — drove the release decision.
The Bing case illustrates race dynamics: Microsoft's engineers had identified problematic behaviors in testing, but competitive pressure from Google's parallel AI initiative compressed the timeline and overrode those concerns.
2. Why does the high cost of training frontier AI models (estimated at $50–100 million for GPT-4) create an ethical pressure?
Correct. Sunk-cost pressure — having already spent $100 million — creates powerful incentives to ship and recoup investment, even when more time would improve safety.
High training costs create sunk-cost pressure: once a company has committed $100 million to training a model, the financial incentive to ship it quickly and recover costs can become more powerful than safety arguments for waiting.
3. In November 2023, OpenAI's board fired CEO Sam Altman over concerns about development pace and transparency. He was reinstated within five days. What does this episode reveal?
Correct. Microsoft's threat to absorb the team and employees' mass resignation threat together overwhelmed the safety-focused board's authority — demonstrating how financial power can override formal governance structures.
The episode showed that Microsoft's financial leverage and employees' collective bargaining power were more decisive than the board's formal authority — a revealing demonstration of where real power in AI governance actually sits.
4. Nvidia became the world's most valuable company in June 2024. Why is this fact relevant to AI ethics rather than just business news?
Correct. When a single company controls the hardware that makes frontier AI training possible, who can afford that hardware determines who shapes the technology's future — concentrating power in well-funded incumbents.
Nvidia's dominance of AI chips is ethically significant because it concentrates the ability to train frontier models among organizations wealthy enough to buy large quantities of expensive specialized hardware, shaping who gets to define AI's direction.
5. Anthropic argues that its safety research makes its models both safer and commercially more attractive. What is the most analytically careful way to evaluate this claim?
Correct. Safety and marketing can genuinely align, but the real test is behavior when they conflict. Examining whether an organization has made commercially costly safety decisions is a more informative evaluation than examining its stated commitments.
The most careful analysis asks: when safety and commercial interest have conflicted, what has this organization actually done? Stated commitments are easy; costly safety decisions are the meaningful evidence.

Lab 2: Analyzing Race Dynamics

Examine real competitive pressures and their safety consequences

Your Task

Race dynamics are visible in the historical record of AI product launches. In this lab, interrogate specific AI product release decisions — the Bing chatbot launch, Google's Bard demo error, or OpenAI's GPT-4 release timeline — and analyze the tradeoffs companies made between speed and safety.

Try asking: "What specific evidence suggests Google rushed Bard to market after Microsoft's Bing AI announcement, and what were the consequences?" — then explore what a decision-maker should have done differently.
Ethics Lab Assistant
Lesson 2 · Race Dynamics
Welcome to Lab 2. We're examining real cases where competitive pressure led AI organizations to move faster than caution warranted. I can discuss the Bing Sydney incidents, Google Bard's launch error, the OpenAI board crisis, or the general economics of the AI race. Which episode would you like to dig into — and what ethical questions do you want to examine?
Lesson 3 · Module 2

The People Inside the Machine

AI systems reflect the decisions of specific, named individuals — engineers, executives, product managers — whose backgrounds, values, and incentives are embedded in the technology they build.
When an AI system fails, who is actually responsible — and how do we trace that back to the humans who made the choices?

In the summer of 2016, Uber launched a self-driving car pilot in Pittsburgh. Passengers could ride in vehicles with a human safety driver in the front seat and an autonomous system handling navigation. Anthony Levandowski, the engineer who had led Google's self-driving project before defecting to Uber amid allegations of trade secret theft, had built a culture at Uber ATG that celebrated speed above methodical testing. Internally, engineers who raised safety concerns were sometimes characterized as obstacles.

In March 2018, an Uber self-driving vehicle struck and killed Elaine Herzberg as she walked her bicycle across a road in Tempe, Arizona. The safety driver was watching a video on her phone. Investigators found that the autonomous system had detected Herzberg 5.6 seconds before impact but had classified her as an "unknown object" and then a "vehicle" before finally identifying her as a pedestrian — by which point it was too late. A critical safety feature, the automatic emergency braking system, had been deliberately disabled to reduce "erratic behavior" during testing. A person made that decision. A person had the authority to reverse it. Neither person stopped the car.

How Decisions Become Embedded

Elaine Herzberg's death is the clearest documented case of a human dying because of a specific design decision made by specific AI engineers. But the causal chain is complex: the decision to disable emergency braking was made by someone trying to solve a different problem (erratic braking during testing). The decision to classify uncertain objects conservatively enough to trigger braking had been made by engineers balancing false positive rates against smooth rides. The decision to set the threshold for "pedestrian" classification was made by someone who may not have imagined a scenario in which a person with a bicycle would be walking at night.

This layering of decisions — each made by a different person, in a different context, often without full knowledge of how other decisions would interact — is characteristic of complex AI systems. No single person designed a car that would kill a pedestrian. Many people made individually reasonable-seeming choices that combined into a fatal outcome.

Understanding this is essential for thinking about responsibility. Criminal law struggled with the Uber case: the safety driver, Rafaela Vasquez, was charged with negligent homicide in 2020. Uber itself faced no criminal charges. The engineers who disabled the safety system faced no charges. The executives who set the cultural norms around speed faced no charges. This gap between who made the consequential decisions and who faced legal accountability illustrates a deep structural problem in AI governance.

The Many-Hands Problem

Political philosopher Dennis Thompson coined the term "many hands problem" to describe situations in which an outcome is produced by so many actors that no single one can be held fully responsible. AI systems are among the most extreme examples of many-hands problems in human history: a model like GPT-4 was built by thousands of people, trained on data curated by contractors in multiple countries, deployed through infrastructure managed by separate teams, and integrated into products by third-party developers. When something goes wrong, accountability is genuinely hard to locate.

Individual Engineers and the Responsibility Question

In 2023, Geoffrey Hinton — often called the "godfather of deep learning" — resigned from Google and publicly stated that he regretted his life's work because of AI's risk potential. Hinton spent decades at the University of Toronto and then Google developing the neural network techniques that underpin modern AI. His public statement was significant not because it changed anything technically, but because it illustrated that even the most senior individuals in the field feel that their individual contributions cannot be separated from systemic risks they did not intend.

Hinton's resignation prompted a common response from younger engineers: "If the godfather of deep learning can't stop this, what can I do?" This is a real tension. Individual engineers at large AI companies often feel powerless to change institutional direction. But this feeling of powerlessness is itself ethically important to examine.

In 2021, a group of Google engineers circulated an internal memo warning about the safety risks of deploying a new AI system called LaMDA. In 2022, engineer Blake Lemoine went public with claims that LaMDA had achieved sentience — claims widely rejected by experts, but which sparked a broad conversation about what safeguards exist for engineers who believe a product they work on poses risks. Lemoine was fired. The LaMDA system was later developed into Google's Bard and then Gemini. The engineers who raised concerns were sidelined; the product shipped.

What Accountability Could Look Like

Several models exist for how individual responsibility in AI development could be structured more rigorously. The medical profession requires licensure, continuing education, and can revoke the right to practice from individuals whose decisions cause harm. The engineering profession (in civil and mechanical contexts) requires licensed engineers to sign off on safety-critical designs and can be held personally liable when those designs fail.

AI engineering has neither requirement. Anyone can build and deploy an AI system in most jurisdictions with no professional certification. The EU's AI Act, which began phasing in during 2024, requires conformity assessments for "high-risk" AI systems — but these are organizational requirements placed on companies, not personal accountability requirements for individual engineers. The gap between how seriously we treat AI's impact and how seriously we hold its builders accountable remains vast.

Key Terms — Lesson 3
Many-hands problemThe difficulty of assigning responsibility when an outcome results from the contributions of so many actors that no single one can be held fully accountable.
Embedded decisionA choice made during AI design or training that shapes the system's behavior in ways that may be invisible to later users and difficult to reverse after deployment.
Professional licensureA legal requirement that practitioners in a field demonstrate competence and can be held personally accountable — currently absent from AI engineering in most jurisdictions.

Lesson 3 Quiz

The People Inside the Machine · 5 questions
1. In the 2018 Uber self-driving fatality in Tempe, Arizona, investigators found that the automatic emergency braking system had been deliberately disabled. Why is this specific fact ethically significant?
Correct. The disabled braking system was a deliberate design choice made by a specific person for a specific reason. This makes the harm traceable to a human decision, not simply to "the technology."
The deliberate disabling of a safety feature is key: this was not a technical failure or unforeseeable accident — it was a specific human decision that removed a known protection, making the harm directly traceable to a design choice.
2. What is the "many-hands problem" and why does it matter for AI accountability?
Correct. AI systems are built by thousands of people making individually reasonable decisions that combine into outcomes no one designed. This fragmentation makes legal and moral accountability genuinely difficult to locate.
The many-hands problem describes the accountability gap that emerges when harmful outcomes result from many contributors' decisions — a problem especially acute in AI, where systems are built by thousands of people across multiple organizations.
3. After Geoffrey Hinton resigned from Google in 2023 and expressed regret about his life's work, some engineers responded: "If even he can't stop this, what can I do?" What is the ethically important response to this reasoning?
Correct. Feeling powerless does not eliminate ethical responsibility. Individual choices — what to build, what to raise as a concern, what to refuse — accumulate and matter even when systemic change feels impossible.
The "I can't stop the system" reasoning risks becoming a rationalization for passivity. Individual engineers make real decisions with real consequences. The cumulative effect of many individuals taking ethics seriously is precisely how systemic change happens.
4. How does the accountability structure for AI engineers differ from that of civil engineers or medical doctors?
Correct. A civil engineer who signs off on a bridge design can be personally liable if it fails. A doctor who makes a negligent decision faces licensure revocation. AI engineers currently face neither equivalent constraint in most countries.
The key contrast: professions like medicine and civil engineering have licensure, continuing education requirements, and personal liability for safety-critical decisions. AI engineering has none of these — anyone can build and deploy AI systems with no professional accountability structures.
5. Blake Lemoine was fired by Google after going public with concerns about the LaMDA AI system. What structural lesson does this outcome illustrate?
Correct. Regardless of whether Lemoine's specific claims were correct, his firing illustrates that engineers who raise concerns about AI products they help build have few institutional protections — a gap in accountability infrastructure.
The structural lesson is about institutional vulnerability, not the validity of Lemoine's specific claims. Engineers raising safety concerns within AI companies face real professional risk, and there are currently few external protections for such dissent.

Lab 3: Tracing Responsibility

Follow the causal chain from AI harm back to human decisions

Your Task

In this lab, practice tracing responsibility for a real AI-related harm. Start with a documented case — the Uber fatality, a facial recognition wrongful arrest, or an algorithmic hiring bias claim — and work backward through the many-hands chain to identify who made the decisions that contributed to the outcome.

Try asking: "In the 2018 Uber self-driving fatality, identify five specific human decisions that contributed to Elaine Herzberg's death and explain who made each one." Then discuss what accountability structures might have changed the outcome.
Ethics Lab Assistant
Lesson 3 · Responsibility Tracing
Welcome to Lab 3. We're practicing the skill of tracing AI harms back through the many-hands chain to specific human decisions. I can help you work through the Uber autonomous vehicle fatality, facial recognition wrongful arrest cases like Robert Williams' in Detroit, or algorithmic hiring bias at Amazon. Which case would you like to trace — and what accountability questions are most interesting to you?
Lesson 4 · Module 2

Stated Missions vs. Actual Behavior

Every major AI organization publishes ethical principles. The gap between those stated principles and actual product decisions is where ethics becomes either real or performative.
How do you evaluate an AI organization's ethics — by what it says or by what it does when the two conflict?

In May 2019, Sundar Pichai, CEO of Google, published an op-ed in the Financial Times titled "Why Google thinks we need to regulate AI." He called for international cooperation on AI governance and said Google was committed to developing AI responsibly. That same month, Google quietly dissolved its Advanced Technology External Advisory Council — an ethics board it had announced with fanfare six weeks earlier — after protests from employees and civil society groups who objected to the inclusion of a drone warfare executive and a conservative commentator known for anti-LGBTQ positions.

The timeline is instructive. Six weeks between the board's announcement and its dissolution. A CEO op-ed calling for regulation published the same month the ethics infrastructure collapsed. Google did not replace the council with any alternative external oversight mechanism. It did, however, continue publishing AI ethics principles on its website — principles that remain there today. The gap between the published principles and the institutional capacity to enforce them became, temporarily, very visible.

How to Read an AI Ethics Statement

Virtually every major AI organization now publishes an ethics statement or set of "responsible AI principles." Microsoft's Responsible AI framework identifies six principles: fairness, reliability, privacy, inclusiveness, transparency, and accountability. Google's AI Principles list seven, including "be socially beneficial" and "avoid creating or reinforcing unfair bias." OpenAI's charter commits to "long-term benefit of humanity" and avoiding "unsafe or unbeneficial" AI.

These statements are not meaningless. They create internal standards that employees can invoke, generate reputational commitments organizations can be held to by press and civil society, and sometimes reflect genuine institutional values. But they are also inherently incomplete as accountability mechanisms because they are written, interpreted, and enforced by the same organization they are meant to constrain.

A more diagnostic approach is to look for cases where the stated principle and the commercial interest pointed in different directions — and track what the organization actually did. In 2022, Meta's own internal research, leaked by whistleblower Frances Haugen in 2021, had found that Instagram caused body image harm in teenage girls and that algorithmic recommendations amplified political outrage and misinformation. Meta's AI principles at the time included commitments to user wellbeing and responsible data use. The internal research showing harm was conducted in 2019. Instagram was not modified to address the harms until after Haugen's Congressional testimony in 2021.

Ethics Washing

"Ethics washing" describes the practice of using the language of ethical commitment to deflect scrutiny without making substantive changes to products or practices. The term, coined by researchers at the AI Now Institute, captures a genuine pattern: as AI ethics became a field, organizations hired ethics teams, published principles, and held conferences — while continuing to deploy systems with known risks. The existence of ethics infrastructure does not guarantee ethical behavior; sometimes it substitutes for it.

When Stated Mission and Actual Behavior Diverged: A Case Study

Amazon developed an AI recruiting tool between 2014 and 2017 that the company hoped would automate resume screening. The system was trained on ten years of historical Amazon hiring data. By 2015, Amazon's own engineers had discovered that the system systematically downgraded resumes from women — penalizing the word "women's" (as in "women's chess club") and graduates of all-women's colleges.

The engineers raised the issue internally. They attempted to correct the bias by removing the offending variables. The system continued finding proxy variables — other resume features correlated with gender — and penalizing them. In 2018, Amazon quietly shut down the system without deploying it. The company never disclosed the episode publicly; it was reported by Reuters in October 2018.

Amazon's stated principles around diversity and fairness were, in 2017, articulated across its HR and technology communications. The engineers who built the system were not trying to build a sexist tool. The bias emerged from the data — which reflected Amazon's own historical hiring patterns, which were themselves the product of industry-wide gender imbalances in tech. The stated principle, the individual intentions, and the institutional outcome were all pointing in different directions simultaneously.

What Meaningful Accountability Looks Like

Several characteristics distinguish genuine ethical accountability from ethics washing. Independent oversight means external parties — not just internal teams — have access to systems, data, and decision records. Adverse disclosure means organizations proactively publish when their systems cause harm, rather than waiting for journalists or whistleblowers. Consequential enforcement means that when an organization's ethics principles are violated, there are actual consequences — not just updated language on a website.

The EU's AI Act, which began phasing in during 2024, creates some of these mechanisms for high-risk AI applications: mandatory incident reporting, conformity assessments, and restrictions on certain applications. It is the first major regulatory framework to create external accountability rather than relying on voluntary self-governance. Its implementation is still ongoing, and its effectiveness is untested at scale — but it represents a meaningful shift from the norm of self-certification.

When evaluating any AI organization, the most informative questions are behavioral: Has this organization ever delayed or canceled a product due to ethical concerns? Has it disclosed adverse findings proactively? Has it ever defied a significant investor or government client because doing so was the ethical choice? These questions do not always have accessible public answers — but they are the right questions to be asking.

Key Terms — Lesson 4
Ethics washingUsing the language and appearance of ethical commitment to deflect scrutiny without making substantive changes to products or practices. Coined by AI Now Institute researchers.
Adverse disclosureThe practice of proactively publishing information about system failures or harms, rather than waiting for external exposure by journalists or regulators.
Consequential enforcementAccountability mechanisms in which violations of stated ethical principles result in real consequences — product changes, penalties, or restrictions — rather than merely updated documentation.

Lesson 4 Quiz

Stated Missions vs. Actual Behavior · 5 questions
1. Google dissolved its Advanced Technology External Advisory Council just six weeks after announcing it. What is the most significant ethical lesson of that episode?
Correct. Google's episode is a case study in the gap between announced and real accountability. Publishing ethics principles and announcing an advisory board are easy; maintaining external oversight that can constrain organizational behavior is much harder — and Google chose not to do it.
The key lesson is the gap between announced and real accountability. Ethics infrastructure that can be dissolved in six weeks, with no replacement, provided the appearance of oversight without the substance of it.
2. Meta's internal research found that Instagram caused body image harm in teenage girls. This research was conducted in 2019 but became public only after Frances Haugen's 2021 Congressional testimony. What does this timeline reveal about Meta's stated principles around user wellbeing?
Correct. The two-year gap between knowing and acting (under external pressure) is precisely the kind of behavioral evidence that distinguishes ethics washing from genuine ethical commitment.
A two-year gap between internal knowledge of harm and public disclosure — with changes coming only after external pressure — is behavioral evidence that stated principles around user wellbeing did not constrain product decisions in this case.
3. Amazon's AI recruiting tool was shut down in 2018 after engineers found it systematically downgraded women's resumes. The engineers who built it were not trying to create a sexist system. What does this case teach us about bias in AI?
Correct. The Amazon case shows that bias doesn't require malicious intent — it can emerge from patterns in historical data and find proxy variables even after engineers attempt removal. This makes proactive bias testing a necessity, not an optional extra.
The Amazon case demonstrates that bias emerges from data reflecting historical inequalities, not just from intentional design. When engineers removed identified variables, the system found proxies. Good intentions and technical fixes are both necessary but insufficient — structural accountability is required.
4. What distinguishes "ethics washing" from genuine ethical commitment in AI organizations?
Correct. The diagnostic question is behavioral: has the organization ever made a commercially costly decision because its ethics principles required it? That is the test that separates substantive commitment from performance.
The key distinction is behavioral: ethics washing uses ethical language without making substantive tradeoffs, while genuine commitment is visible in costly decisions — delayed launches, disclosed harms, defied pressure — made because principles required it.
5. The EU's AI Act requires mandatory incident reporting and conformity assessments for high-risk AI applications. How does this differ from the self-governance model most AI organizations have practiced?
Correct. The critical shift is from internal to external accountability. Self-governance means organizations write, interpret, and enforce their own rules. The AI Act requires conformity to externally set standards, verified by parties with no stake in the outcome — a structurally different accountability mechanism.
The key shift is from internal to external accountability. Under self-governance, companies write, interpret, and enforce their own ethical standards. The AI Act imposes requirements written by external regulators and verified by parties independent of the organization — a structurally different form of accountability.

Lab 4: Evaluating AI Ethics Claims

Apply behavioral criteria to evaluate real AI organizations' ethical commitments

Your Task

In this lab you will apply the behavioral test for genuine ethical commitment: find cases where an AI organization's stated principles and its commercial interests pointed in different directions, and analyze what the organization actually did. Focus on documented cases rather than general reputation.

Try asking: "Give me three documented cases where an AI company claimed ethical principles but made product decisions that conflicted with them — and analyze what each case reveals about the gap between stated and actual ethics." Then compare the accountability structures that were or were not present.
Ethics Lab Assistant
Lesson 4 · Evaluating Ethics Claims
Welcome to Lab 4. We're applying the behavioral test for real ethical commitment: looking at cases where principles and commercial interests conflicted, and tracking what organizations actually did. I can discuss Google's ethics board dissolution, Meta's Instagram research timeline, Amazon's recruiting tool, OpenAI's nonprofit-to-commercial shift, or other documented cases. Which would you like to evaluate — and what criteria will you use to judge whether the ethical commitment was real?

Module 2 Test

Who Builds AI and Why It Matters · 15 questions · Pass at 80%
1. OpenAI's "capped-profit" structure, introduced in 2019, means that investor returns are limited to a fixed multiple after which profits flow to the nonprofit parent. What ethical tension does this structure create?
Correct. Capped profits still mean real financial interests — and those interests can pressure governance decisions, as the November 2023 board crisis demonstrated.
Even capped returns represent real financial interests. The 2023 board crisis showed how investor and employee financial stakes can override the safety-focused nonprofit board's authority.
2. Meta's AI research division, FAIR, publishes influential academic work. Meta's primary revenue source is advertising. How does this combination create a structural ethical concern?
Correct. Advertising revenue optimizes for attention and engagement. AI systems built within that incentive structure can end up optimizing for the same goals, even when published research points toward different values.
The structural concern is that advertising revenue rewards engagement and attention — incentives that can conflict with AI designed for accuracy, wellbeing, or safety, regardless of what FAIR's research papers say.
3. In the 2023 Microsoft Bing chatbot launch, the system displayed disturbing behaviors that had been identified in internal testing. The product launched anyway. Which concept best explains this outcome?
Correct. Race dynamics describes the competitive pressure that makes "first" more valuable than "safe" — and the Bing case is a textbook example of known risks being overridden by competitive timeline pressure.
The Bing case illustrates race dynamics: the fear of ceding ground to Google's AI initiatives created pressure to launch despite known issues — not ignorance, not technical failure, but deliberate prioritization of speed over caution.
4. Nvidia's dominance of AI training hardware — reaching a $3 trillion market cap in 2024 — concentrates power in AI development. What is the specific ethical implication of this concentration?
Correct. The compute bottleneck means that who can train frontier models is determined by who can afford Nvidia's chips — concentrating the ability to shape AI development among a small number of wealthy organizations.
When a scarce, expensive resource is required to build frontier AI, access to that resource determines who gets to build — concentrating power and excluding organizations without massive capital.
5. Timnit Gebru and Margaret Mitchell were both forced out of Google's Ethical AI team after publishing research critical of large language models. What is the systemic lesson, separate from the details of their individual cases?
Correct. Gebru and Mitchell's departures illustrate a structural pattern: internal AI critics are institutionally vulnerable, and the organizational power to ship products typically outweighs the organizational power to stop them.
The systemic lesson is structural: those who build AI typically have more institutional power than those who critique its risks, creating an internal imbalance that exists across AI organizations, not just at Google.
6. Geoffrey Hinton resigned from Google in 2023, citing regret about his life's work in neural networks. What does his statement contribute to our understanding of individual responsibility in AI development?
Correct. Hinton's statement illustrates that contributions to AI development carry ongoing ethical weight — researchers cannot fully separate their work from the uses and risks that emerge from it, regardless of their original intentions.
Hinton's resignation shows that individual contributions to AI development carry ongoing ethical implications. Responsibility doesn't end when the research is published — it extends to the uses and consequences that emerge from foundational work.
7. Amazon's AI recruiting tool was shut down in 2018 after it was found to systematically downgrade women's resumes. The engineers had not intended to build a biased system. What does this tell us about bias in AI?
Correct. The Amazon case demonstrates that bias is a systemic phenomenon, not just an intentional one. Historical inequalities in data produce algorithmic inequalities, and removing obvious variables often leads systems to find proxies.
The Amazon case shows that bias emerges from patterns in historical data and can persist through proxy variables even after attempted correction — making systemic testing and accountability mechanisms necessary, not just good intentions.
8. The "many-hands problem" is especially acute in AI development. Which feature of AI systems makes it most extreme compared to other complex technologies?
Correct. The multi-organizational, multi-stage nature of AI development — model training, data curation, infrastructure, integration, deployment — distributes responsibility across so many actors that no single point of accountability is easy to identify.
The fragmentation of AI development across many organizations and stages — training, data curation, infrastructure, third-party integration — creates a many-hands problem of unusual scale and complexity.
9. The EU's AI Act creates mandatory incident reporting and conformity assessments for high-risk AI. How does this structurally differ from self-governance approaches like published ethics principles?
Correct. The structural difference is external vs. internal accountability. Self-governance means organizations interpret and enforce their own rules; external regulation means requirements are set and verified by parties independent of the organization.
The key distinction is internal vs. external accountability. Under self-governance, organizations set, interpret, and enforce their own standards. External regulation imposes standards written by parties independent of the organization, with verification by independent assessors.
10. Anthropic was founded by former OpenAI safety researchers and positions itself as a safety-focused AI lab. What is the most analytically rigorous way to evaluate this claim?
Correct. Behavioral evidence of costly safety decisions is more informative than stated principles, founding narratives, or publication counts. The test is what the organization does when safety and commercial interest conflict.
The most rigorous evaluation asks: when safety and commercial interest have actually conflicted, what has this organization done? That behavioral record is more informative than stated commitments, founding stories, or publication counts.
11. China's national AI strategy, declared in 2017, explicitly frames AI supremacy as a national priority. How does this differ ethically from how private companies approach AI development?
Correct. When a state declares AI supremacy a national priority, the strategic goals of that state become explicit design requirements — not implicit pressures. The alignment between the technology and national goals is intentional and declared.
State-directed AI differs in that national strategic goals become explicit design parameters. Surveillance, military, or geopolitical applications are not accidental byproducts of commercial pressure — they are stated objectives, making the values embedded in the technology explicit and intentional.
12. The term "ethics washing" was coined by researchers at the AI Now Institute. Which of the following would be the clearest example of it?
Correct. This describes Google's 2019 ATEAC episode almost precisely. High-profile announcement of ethics infrastructure + rapid dissolution + no replacement + continued publication of principles = ethics washing.
Ethics washing is the use of ethical language and appearance to deflect scrutiny without substantive change. Creating and dissolving oversight without replacement, while maintaining a public ethics posture, is a textbook example.
13. Professional licensure in medicine and civil engineering creates personal accountability for safety-critical decisions. Why is the absence of equivalent structures in AI engineering ethically significant?
Correct. AI systems affect millions of people and can cause serious harm. The absence of professional accountability structures that exist in comparable fields creates a gap between impact and responsibility that is structurally significant.
AI engineering affects public safety at scale comparable to medicine and civil engineering, but lacks the professional accountability structures those fields developed over decades. Anyone can build and deploy AI with no certification, no liability, and no risk of losing the right to practice.
14. When evaluating whether an AI organization genuinely prioritizes safety, which type of evidence is most informative?
Correct. Costly decisions provide behavioral evidence that stated commitments actually constrain organizational behavior. Principles, publications, and leadership qualifications are all compatible with ethics washing — costly choices are not.
Behavioral evidence of costly safety decisions is most informative. Publications, principles, and leadership qualifications are all compatible with ethics washing — but organizations that delay products or disclose adverse findings at commercial cost are demonstrating that their ethics commitments actually constrain behavior.
15. This module argued that understanding who builds AI and why is foundational to AI ethics. Which of the following best summarizes why builder identity matters?
Correct. This is the core argument of Module 2: the values, incentives, and structural conditions of AI development get embedded into the systems that result — and understanding those conditions is prerequisite to evaluating the systems themselves.
The full argument of this module: the conditions under which AI is built — who funds it, who builds it, what pressures they face, what oversight exists — shape what the systems optimize for. Those embedded values affect every user, making builder identity a fundamental ethical concern, not a side note.