Karen Hao - Prologue

Who is the author of Empire of AI, and what is the prologue called?

Karen Hao is the author of Empire of AI. The prologue is called "A Run for the Throne." Hao is a journalist who has reported on AI and OpenAI for years. She uses OpenAI's 2023 leadership crisis as an entry point into a much larger argument about who has the power to shape AI and who bears the costs of its development.

What is Karen Hao's main argument in the prologue of Empire of AI?

Hao argues that the future of AI is not inevitable or determined only by technical progress. It is being shaped by choices made by a relatively small number of powerful people and companies. She uses OpenAI's transformation and its 2023 leadership crisis to show how control over an enormously consequential technology has become concentrated among Silicon Valley elites, corporations, and investors, while many of AI's social, labor, and environmental costs are pushed onto less powerful people.

Why does the prologue begin with Sam Altman's firing?

Hao uses Altman's sudden firing and rapid return as more than Silicon Valley drama. The crisis provides a concrete example of her central question: WHO GOVERNS AI?

Decisions about the leadership and direction of one of the world's most influential AI companies were made by a tiny group behind closed doors, even though the technologies being developed could affect people around the world.

What happened to Sam Altman on November 17, 2023?

OpenAI's board abruptly fired CEO Sam Altman. Board member and chief scientist Ilya Sutskever informed him through a Google Meet while Altman was in Las Vegas. The board publicly stated that Altman had not been "consistently candid" in his communications with them and that this interfered with the board's ability to oversee the company.

Why was Altman's firing so shocking?

OpenAI appeared to be extraordinarily successful. ChatGPT had become a massive consumer product, the company's valuation was rapidly increasing, and Altman had become the public face of the generative AI revolution. Most employees and even some senior executives had little warning that he was about to be removed, creating confusion about why the board would fire the leader of such a successful company.

Who was Ilya Sutskever, and what role did he play in the crisis?

Ilya Sutskever was OpenAI's chief scientist and a member of its nonprofit board. He participated in the decision to remove Altman and initially defended the board's actions as consistent with OpenAI's mission. However, as employee opposition intensified, Sutskever reversed course, expressed regret for participating in the board's actions, and signed the employee letter demanding Altman's return.

Who was Mira Murati during the OpenAI crisis?

Mira Murati was OpenAI's chief technology officer and was initially chosen by the board to serve as interim CEO after Altman's firing. However, she and other company leaders soon pushed for Altman to return as frustration with the board grew. The board then sought another interim CEO.

Why were OpenAI employees angry with the board after Altman's firing?

Employees were given very little explanation for why Altman had been removed. During an all-hands meeting, Ilya Sutskever repeatedly declined to provide specifics and directed employees back to the vague public statement. Employees felt that a decision dramatically affecting their work, financial futures, and company had been made without transparency.

Why did OpenAI employees have financial reasons to fear the company's collapse?

OpenAI was completing a tender offer that would allow eligible employees to sell shares potentially worth millions of dollars. The leadership crisis threatened that deal and possibly the company's survival. Employees therefore had both professional and financial reasons to resist the board's decision.

Why was Microsoft so important during the OpenAI crisis?

Microsoft was OpenAI's biggest financial and technological partner and provided the computing infrastructure OpenAI relied on to train models and operate its products. This gave Microsoft enormous leverage. After Altman's firing, Microsoft CEO Satya Nadella eventually offered Altman, Greg Brockman, and other OpenAI employees positions in a new Microsoft AI group.

What happened after Microsoft offered jobs to Altman and OpenAI employees?

The offer dramatically increased employees' leverage against the board because they now had somewhere to go if OpenAI collapsed. More than 700 of roughly 770 OpenAI employees eventually signed a letter threatening to leave and join Microsoft unless Altman was reinstated and the board resigned. Their message was essentially that OpenAI's value depended on retaining its people.

What was significant about the statement "OpenAI is nothing without its people"?

The phrase became a rallying cry among employees opposing the board. It demonstrated that employees themselves possessed power because OpenAI's valuable research and products depended on their expertise. The threat of a mass resignation helped make an OpenAI without Altman increasingly difficult to sustain.

How did the November 2023 OpenAI crisis end?

After several days of negotiations, the board agreed to bring Altman back as CEO. Altman and Greg Brockman agreed not to immediately return to the board. Helen Toner and Tasha McCauley stepped down, Adam D'Angelo remained, and Bret Taylor and Larry Summers joined a reconstituted board. OpenAI then publicly emphasized unity and stability.

Why was OpenAI's board different from a normal corporate board?

OpenAI had an unusual structure in which a nonprofit ultimately controlled the company. The board was supposed to prioritize OpenAI's mission of ensuring that artificial general intelligence benefited humanity rather than simply maximizing financial returns to shareholders. This meant the board theoretically had authority to make decisions—even firing the CEO—that could hurt the company's financial value if it believed doing so protected the mission.

Why is Helen Toner's statement about potentially destroying OpenAI important?

During the crisis, Toner emphasized that if a board action destroyed the company, that outcome could theoretically still be consistent with the nonprofit mission. Her point reflected OpenAI's unusual governance structure: preserving the company itself was not necessarily supposed to be the board's highest priority. The mission of ensuring beneficial AGI was theoretically more important than OpenAI's financial survival.

What did OpenAI originally claim its mission would be?

OpenAI was founded as a nonprofit with the stated goal of developing artificial general intelligence for the benefit of humanity rather than primarily for shareholder profit. Its founders emphasized openness, collaboration, research rather than commercial products, and broad sharing of scientific findings. The name "OpenAI" reflected this original commitment to openness.

What is AGI in the context of Empire of AI?

AGI stands for artificial general intelligence. OpenAI described it as a highly capable form of AI with broad abilities rather than a system specialized for one narrow task. The possibility of AGI became central to OpenAI's mission and justification for developing increasingly powerful AI systems.

What unusual promise did OpenAI make about competition in its early years?

OpenAI warned that a competitive race toward AGI could encourage companies to sacrifice safety. It therefore promised that if another organization came close to developing beneficial AGI before OpenAI, OpenAI would stop competing with that organization and instead assist it. Hao presents this promise as an example of how strongly OpenAI originally claimed to prioritize humanity and safety over winning.

Why did OpenAI's original nonprofit model become difficult to maintain?

Developing the kind of AI OpenAI wanted required enormous amounts of computing power and therefore enormous amounts of money. After Elon Musk left the organization and withdrew his financial support, OpenAI faced financial pressure. Altman responded by creating a for-profit arm that could raise investment capital and commercialize technology.

How did OpenAI's organizational structure change in 2019?

OpenAI created a for-profit entity, OpenAI LP, nested within and ultimately controlled by the nonprofit. This allowed it to raise capital, commercialize products, and provide financial returns to investors while formally retaining its nonprofit mission. Soon afterward, Microsoft announced a $1 billion investment.

According to Hao, how did OpenAI change from its original ideals?

Hao argues that OpenAI gradually became the opposite of what it originally promised to be. It moved from openness toward secrecy, from nonprofit research toward aggressive commercialization, from collaboration toward competition, and from caution about an AGI race toward helping accelerate that very race. Products such as ChatGPT became central to its commercial growth.

Why does Hao view the 2023 OpenAI crisis as a failure of its governance experiment?

OpenAI had created its nonprofit governance structure specifically to prevent commercial interests from overriding its mission. Yet when the board tried to exercise its authority by removing Altman, pressure from employees, investors, Microsoft, and other powerful interests made the company's survival increasingly difficult. The board ultimately reversed course. For Hao, the episode demonstrated how difficult it was for OpenAI's mission-focused governance structure to withstand enormous financial and corporate pressures.

What larger question does Hao believe the OpenAI leadership crisis raises?

The larger question is: "How do we govern artificial intelligence?" Because AI increasingly influences areas such as health care, education, law, finance, journalism, and government, decisions about how AI develops can affect society broadly. Hao argues that the question of who governs AI is therefore ultimately a question about who gets to shape our collective future.

Why does Hao argue that AI's future is NOT inevitable?

Today's dominant forms of AI did not simply emerge because they were technically destined to win. Hao argues that they resulted from thousands of subjective choices made by researchers, executives, investors, and other powerful decision-makers. Future AI systems can therefore also take different forms depending on who has decision-making power and which goals society chooses to prioritize.

How does ideology influence AI development according to Hao?

Hao argues that AI evolves not only according to technical merit but also according to the beliefs and goals of the people developing it, as well as commercial incentives and hype. Different developers make choices about what AI should accomplish, how it should be built, and what resources should be devoted to it. Technologies therefore embody particular visions of how the world is and how it should be.

Why does Hao say generative AI represents only one possible form of AI?

Artificial intelligence includes many different technologies and approaches. Large language models and generative AI currently dominate public attention, but Hao argues that their dominance was not inevitable. They represent one particular approach shaped by decisions to prioritize enormous models, huge data sets, computing power, and commercial products.

What is Hao's criticism of the idea that AI development is purely technical?

The direction of AI is influenced by human choices, ideology, corporate incentives, money, hype, and competition—not simply by which technical approach is scientifically "best." This means decisions about AI development are also social and political decisions because they determine what technologies receive resources and whose interests those technologies serve.

What hidden resources does generative AI require according to Hao?

Generative AI systems require enormous amounts of DATA, HUMAN LABOR, COMPUTING POWER, ENERGY, WATER, and physical infrastructure. Their polished interfaces can make them appear almost immaterial, but Hao emphasizes that they depend on extensive physical and human resources distributed around the world.

Why does Hao emphasize data-labeling and content-moderation workers?

AI systems depend on human workers who label, categorize, clean, and moderate data used to train models. Hao argues that many of these workers, particularly in lower-income regions, experience low pay and poor working conditions while the companies using their labor accumulate enormous wealth. Their often-invisible work challenges the idea that AI systems are purely automated.

What examples does Hao give of AI's environmental costs?

Large AI systems require massive data centers and computing infrastructure that consume substantial electricity and water. Hao describes reporting in places such as Arizona and Chile, where residents and activists were concerned about data centers consuming scarce water resources. These examples show that seemingly digital AI technologies have material environmental footprints.

Who receives the benefits and who bears the costs of AI according to Hao?

Hao argues that the economic benefits of the current generative AI boom disproportionately flow upward toward powerful technology companies and wealthy actors, while many of the human and environmental costs fall on workers and communities with much less power. This unequal distribution is central to her critique.

Why does Karen Hao use the metaphor of an "empire" to describe major AI companies?

Hao compares today's powerful AI companies to empires because they accumulate wealth and power by extracting resources from broad populations and territories. AI companies consume people's data and creative work, rely on global labor, and require land, energy, water, and computing infrastructure. Meanwhile, the economic benefits become concentrated among a relatively small number of powerful companies.

Is Hao saying AI companies are literally identical to historical colonial empires?

No. Hao explicitly acknowledges that today's AI companies are not engaged in the same overt violence and brutality that characterized historical colonialism. Her comparison focuses on a structural similarity: powerful institutions extract resources and labor from others, justify expansion through narratives of progress and competition, and concentrate the resulting wealth and power.

How does resource extraction connect to Hao's "empire" metaphor?

Historical empires extracted land, natural resources, and labor from the territories they controlled. Hao argues that AI companies similarly depend on resources they did not independently create: writings and artwork used as training data, personal information produced online, global data-labeling labor, and natural resources such as water and energy needed for data centers.

How does competition between AI companies strengthen the "empire" metaphor?

Hao argues that empires historically justified expansion partly by claiming they had to compete with rival empires. Similarly, AI companies argue that they must develop larger and more powerful systems because competitors are doing the same. This "arms race" logic can make continued expansion seem necessary even when it creates social or environmental costs.

What does Hao mean by the AI "race to scale"?

The dominant AI paradigm assumes that increasingly powerful systems require larger models, more training data, more computing infrastructure, and greater financial investment. Companies therefore compete to "out-scale" one another. Hao argues that OpenAI helped establish this approach as the industry's dominant strategy.

Why is the race to scale politically important?

Building extremely large AI systems is so expensive that only a small number of wealthy corporations can compete effectively. This concentrates technological power in organizations with access to enormous amounts of capital, computing infrastructure, and data. Scale therefore does not just affect model performance—it influences who is capable of participating in AI development.

How can the current generative AI paradigm reduce diversity in AI research?

When money, computing resources, and researchers are redirected toward building large generative AI models, alternative approaches receive fewer resources. Independent researchers without industry funding struggle to compete, and researchers may align their work with industry priorities to remain employable. Hao argues that this narrows the range of ideas shaping AI's future.

What is the relationship between AI development and concentrated corporate power in Hao's argument?

The enormous cost of developing frontier generative AI favors companies with massive financial and computing resources. As those companies become dominant, governments, businesses, researchers, and consumers may increasingly depend on a small group of firms for access to advanced AI. Hao warns that this can create oligopolistic control over technologies presented as essential to the future.

How does Hao challenge claims about AI-driven productivity?

Hao contrasts industry promises of dramatic productivity growth with evidence suggesting more mixed results. She cites reports questioning whether enormous AI investments are producing comparable economic value and a worker survey in which many employees said generative AI increased their workloads. Her point is that claims of inevitable economic transformation should be critically examined rather than accepted at face value.

What role does "hype" play in Hao's argument?

Hype helps portray increasingly powerful AI systems as inevitable, revolutionary, and necessary for progress. This can encourage society to accept enormous investments and social costs today because of promised future benefits. Hao argues that these narratives can obscure who currently benefits from AI development and who currently pays its costs.

How does Hao characterize OpenAI's use of AGI rhetoric later in its development?

Hao argues that AGI increasingly functions as a rhetorical justification for continued expansion. Promises of extraordinary future benefits can justify acquiring more money, computing power, data, and influence today. She wants readers to distinguish speculative future promises from the present-day social and environmental consequences of AI development.

What is the connection between AI, progress, and power in Empire of AI?

AI companies often frame increasingly powerful systems as synonymous with technological progress. Hao challenges that assumption by asking who defines "progress," who receives its benefits, and who bears its costs. A technology can become more powerful while simultaneously increasing inequality, exploitation, environmental pressure, or concentrated corporate control.

Does Hao argue that AI itself should be abandoned?

No. Hao argues that AI does not have to take its current form. She rejects the assumption that progress requires ever-larger models and ever-greater resource consumption. Smaller AI models and alternative technological approaches could potentially help address problems in health care, education, environmental protection, and other areas without following the same race-to-scale paradigm.

What alternative future for AI does Hao propose?

Hao calls for greater democratic control and a wider diversity of AI approaches rather than allowing a handful of companies to determine the technology's direction. She argues for stronger privacy and transparency rules, updated intellectual property protections, improved labor standards for workers such as data labelers, and greater funding for alternative forms of AI research.

Why are privacy and intellectual property protections important to Hao's alternative vision?

AI companies rely heavily on data and creative work produced by other people. Hao argues that stronger privacy protections and updated intellectual property rules could give individuals greater control over whether and how their data, writing, artwork, and other creations are used in AI development.

Why does Hao call for stronger protections for AI workers?

The current AI industry depends on human labor that is often hidden behind supposedly automated systems, including workers who label data and filter harmful content. Hao argues that international labor standards, minimum wages, humane working conditions, and broader labor protections are necessary so technological progress does not depend on exploitation.

Why does Hao think funding alternative AI research matters?

If nearly all funding and computing resources support the same large-scale generative AI paradigm, researchers have less ability to explore fundamentally different approaches. Supporting independent and diverse research could expand society's choices about what AI becomes rather than allowing commercial incentives to determine its future.

What does Hao mean by asking "Who will get to shape AI?"

The question reflects her belief that technology is shaped by human decisions rather than predetermined technical evolution. If a small group of executives, investors, and corporations make those decisions, AI will reflect their priorities. Broader participation could produce technologies built around different social goals and values.

What is the relationship between OpenAI's governance crisis and Hao's "empire" argument?

The governance crisis shows the concentration of decision-making power at the top of the AI industry. The "empire" argument expands the analysis outward: those powerful institutions then gather data, labor, capital, computing power, and natural resources from around the world. Together, they illustrate both the concentration of control and the unequal distribution of AI's costs and benefits.

Why is the title "A Run for the Throne" significant?

The "throne" symbolizes control over one of the world's most influential AI organizations and, more broadly, power over the direction of AI itself. The struggle over Altman's position reveals that the people competing for control of OpenAI possess extraordinary influence over technological decisions that could affect far more people than those included in the company's internal power struggle.

What is the biggest contradiction Hao identifies in OpenAI?

OpenAI was created partly to prevent powerful AI from being controlled by narrow commercial interests and to ensure that AGI benefited humanity. Yet Hao argues that OpenAI itself became increasingly commercial, secretive, competitive, and powerful. An organization created to solve the problem of concentrated AI power ultimately became an example of that very problem.

How does Hao connect AI governance to democracy?

Hao argues that decisions about AI affect society broadly but are often made by a tiny number of corporate leaders, investors, and technical experts behind closed doors. The democratic problem is therefore not only whether AI systems are safe; it is whether the people affected by AI have meaningful influence over what technologies are developed, how they are used, and what goals they serve.

What does Hao want readers to understand about the people building AI?

The people building AI are not neutral conduits for inevitable technological progress. Their beliefs, ambitions, incentives, rivalries, and decisions leave "fingerprints" on the technologies they create. Understanding AI therefore requires examining the people and institutions behind it, not simply the technical characteristics of the models.

What does Hao want readers to understand about the apparent "magic" of generative AI?

The polished output of a chatbot can hide the enormous system required to produce it. Behind generative AI are human workers, training data, copyrighted and personal material, computer chips, data centers, energy, water, corporate investment, and political choices. Hao wants readers to look beneath the interface and examine this infrastructure and its costs.

What is the relationship between AI and inequality in the prologue?

Hao argues that the current AI system can reinforce inequality because the rewards are concentrated among already-powerful companies and investors while many costs are externalized onto workers, artists, local communities, and other less powerful groups. The central issue is therefore not merely whether AI creates wealth, but how that wealth and its associated costs are distributed.

Why is "Who benefits?" an important question when evaluating AI according to Hao?

Claims that AI "benefits humanity" can hide unequal outcomes. Hao argues that we should identify who receives profits, productivity gains, or technological capabilities and compare them with who supplies the data, labor, creative work, water, energy, and other resources required to create those benefits.

What does Hao mean by saying society can "wrest back control" of AI's future?

Because the current direction of AI resulted from human choices, society can make different choices. Hao argues that governments, researchers, workers, human-rights organizations, and the public can establish rules and institutions that redistribute control over AI development. Her point is that society should not treat the industry's current trajectory as inevitable.

What is the overall takeaway from the Empire of AI prologue?

AI'S FUTURE IS A QUESTION OF POWER AND GOVERNANCE. Hao uses OpenAI's leadership crisis to show that a tiny group of powerful people can make decisions about technologies with enormous social consequences. She then argues that today's generative AI industry resembles an "empire" because it extracts data, creative work, labor, energy, water, and other resources while concentrating wealth and decision-making power at the top. But this path is not inevitable: different governance, protections, research priorities, and technological approaches could produce a different future for AI.