Written by: Techub News Compilation

Introduction

In the popular YouTube podcast "The Diary Of A CEO," host Steven Bartlett had an in-depth conversation lasting over two hours with senior tech journalist Karen Hao. Karen Hao, who served as an AI reporter for MIT Technology Review for a long time, later joined The Wall Street Journal for investigative reporting. The conversation centered around her new book "AI Empire: Dreams and Nightmares of Sam Altman at OpenAI," which is based on her over eight years of industry observation and interviews with more than 250 industry insiders, over 90 of whom are current or former employees and executives of OpenAI. This conversation is important because it systematically critiques the major corporations currently dominating AI development from the perspective of an insider and investigative journalist, revealing the unknown "imperial" logic, narrative manipulation, and systemic exploitation of labor, the environment, and public interests in their operations.

Summary

• Narrative Manipulation and the "Gaslighting" Effect: AI giants (such as OpenAI) manipulate the public, policymakers, and capital by flexibly defining concepts like "Artificial General Intelligence" (AGI) while simultaneously promoting their "utopian" prospects and "extinction-level" risks, thus securing resources for their unbridled development and resisting regulation.
• The Four Pillars of the "Empire": Karen Hao compares AI giants to an "empire," whose power is built on four pillars: the appropriation of data and intellectual property, the exploitation of global labor, the monopolization of knowledge production and control of research agendas, and the creation of a "good vs evil empire" narrative to justify their actions.
• Power and Division within OpenAI: Through numerous internal interviews, she reveals Sam Altman’s controversial leadership style, his fundamental disagreements with co-founder Ilya Sutskever, and the boardroom strife that led to his temporary dismissal and reinstatement, illustrating the chaos and human conflicts behind the company’s pursuit of AGI goals.
• Call to Action: Breaking the Empire and Finding Alternatives: Karen Hao argues that the current AI development path centered on "violent scaling" (which she terms "rocket") brings enormous social and environmental costs. She calls on the public to break the AI empire's monopoly through resisting data exploitation, opposing local data center construction, and supporting alternative AI development (such as "bicycle"-style efficient specialized models), thus promoting technology towards a fairer and more sustainable direction.

The "Empire" Metaphor and Narrative Manipulation in the AI Industry

At the beginning of the interview, Karen Hao pointed out that many operational methods in today's AI industry are "extremely inhumane." The core framework she uses to understand all of this is the "empire" metaphor. She believes that only this metaphor can fully encompass the operational dimensions, scale, and motivational behaviors of these AI giants.

She lists many similarities between the AI empire and historical empires: First, resource appropriation. They claim ownership of resources that do not actually belong to them to train models, including personal data, intellectual property from artists, writers, and creators, as well as "land grabs" to build the next generation of model training supercomputing facilities.

Second, labor exploitation. They employ hundreds of thousands of contract workers worldwide (including in the United States) for data labeling and other tasks to create these technologies. At the same time, the tools they design essentially replace labor, and when technology is deployed, it erodes labor rights. "This is a political choice," emphasizes Karen Hao.

Third, monopolizing knowledge production. They instill a notion among the public and policymakers that only they truly understand how this technology works. If the public is not satisfied, it is because they do not understand enough. They also "capture" most researchers who study AI limitations and capabilities through funding and employment relationships and will censor researchers who discover unfavorable results. She cites the dismissal of Dr. Timnit Gebru, co-lead of Google's ethical AI team, for publishing a paper criticizing the harmful impacts of large language models as an example.

Fourth, manufacturing "good vs evil empire" narratives. Empires always proclaim to the public: "We are the good empire; we must become an empire because there are still bad empires in the world." In the field of AI, early OpenAI depicted Google as the "evil empire" and later often pointed to China. This narrative of "if we do not act, the evil empire will, and the world will descend into hell" becomes an excuse for justifying their appropriation of resources and labor.

Karen Hao specifically points out that this narrative manipulation reaches the level of "gaslighting" the public. AI companies flexibly define core objectives (like AGI) to serve different audiences. She uses OpenAI as an example: "When Sam Altman talks to Congress, AGI is a system that can cure cancer, solve climate change, and eliminate poverty. When he speaks to consumers to sell products, AGI is the most magical digital assistant you will have. When entering an investment agreement with Microsoft, AGI is defined as a system that can generate hundreds of billions of dollars in revenue. Yet on OpenAI's own website, AGI is defined as 'a highly autonomous system that surpasses humans in most economic value work.'" She believes this is not a consistent vision of a technology but rather different definitions spoken to different audiences to evade regulation, garner consumer support, or attract more capital.

This manipulation is even reflected in the early statements of the company’s founders. She mentions that before OpenAI was officially announced in 2015, Altman wrote in a blog post, "The development of superhuman machine intelligence may be the greatest threat to the continued existence of humanity... AI may be the most likely way to destroy everything." Karen Hao's investigation shows that at that time, Altman was trying to persuade Elon Musk to co-found OpenAI, while Musk was spending considerable time warning about the enormous existential threats posed by AI. "If you line up the language Altman used at that time with what Musk said concurrently, they mirror everything Musk was saying," she argues, suggesting that this was a deliberate linguistic match to win Musk's trust and involvement. Musk later left OpenAI feeling manipulated, partly due to this.

Power, Division, and "Myth" Immersion within OpenAI

Based on interviews with many current and former OpenAI insiders, Karen Hao paints a picture filled with power struggles, personal conflicts, and ideological divisions. She points out that opinions on Sam Altman are highly polarized: "Some see him as the greatest tech leader of this generation, akin to a modern-day Steve Jobs; others believe he is very manipulative, abuses power, and lies." She argues that this divide ultimately depends on whether an individual's vision for the future aligns with Altman's.

She cites Dario Amodei, the current CEO of Anthropic, as an example. Amodei, a former OpenAI executive, initially believed Altman shared his vision but gradually felt that Altman was actually completely opposite to him and thought that Altman was using his intelligence, capabilities, and skills to realize a future vision he fundamentally disagreed with. This is precisely why many people eventually became discontented with Altman.

This division climaxed in late 2023 when the OpenAI board briefly dismissed Altman. In her book, Karen Hao recounts the event in detail through interviews with six or seven individuals directly involved or knowledgeable about the decision-making process. The core conflict arose from the serious concerns about Altman's leadership style expressed by Chief Scientist Ilya Sutskever and then Chief Technology Officer Mira Murati.

They reported to independent board members (such as Helen Toner) that Altman had created a highly chaotic environment within the company, causing teams to oppose each other and convey different messages to different people, resulting in a lack of trust among personnel and competition rather than collaboration. Following the release of ChatGPT, the company's rapid unprepared growth exacerbated this chaos, with frequent server crashes and hasty hiring and firing.

The key point is that board members realized they were not running an ordinary company (like Instacart). They believed they were building AGI technology that could "make or destroy the world." In this context, the "instability" and "chaos" caused by Altman's behavior were viewed as unacceptable risks. Karen Hao quotes Ilya's concerns at the core: "I don't think Sam (Altman) is the person who should press the AGI button."

Furthermore, the independent board members discovered other inconsistencies, such as the fact that the OpenAI startup fund was legally structured as "Altman’s fund" rather than "OpenAI's fund." These accumulating doubts ultimately led the board to decide to take swift action, dismissing Altman without widely informing major stakeholders (such as Microsoft), which in turn triggered subsequent employee protests and Altman's quick reinstatement.

Karen Hao observed that many of OpenAI’s early core members ultimately left due to conflicts with Altman's vision and founded competitors: Musk founded xAI after leaving, Dario Amodei founded Anthropic, Ilya Sutskever founded Safe Superintelligence Inc., and Mira Murati founded Thinking Machines Lab. "Every tech billionaire wants to create AI in their own image, which is why they can never get along. In fact, they not only cannot get along; they will ultimately hate each other after collaborating."

So, do these leaders really believe they are "summoning demons" (as Musk said ten years ago)? Karen Hao offers a complex psychological explanation. She believes that, on the one hand, they actively engage in "myth-making," as internal documents show they are very aware of how to guide the public through dazzling technology demonstrations and carefully crafted lofty-sounding missions to gain more tolerance and support for their companies. On the other hand, many are also immersed in this "myth." "Because they must live, breathe, and embody this myth day in and day out." She draws an analogy with the story of Paul Atreides in "Dune": the protagonist initially knows it is a myth but uses it to control the people, ultimately blurring the lines between myth and reality himself. She believes that when Dario Amodei says, "There is a 10% to 25% chance that things will go catastrophically wrong," he is both actively engaging in myth-making and potentially losing himself in the myth to some extent.

The Real Costs: Labor, Environment, and the "Abandoned Ones"

Karen Hao's critique extends beyond the internal power narratives of companies to the broad societal and environmental costs arising from AI technology development. She points out that AI companies claim automation will create "new jobs beyond our imagination," but many of the jobs created are far worse than the original ones.

She specifically describes the current state of data labeling work: this work has not decreased; instead, it is increasing. These workers feel treated like machines, merely to extract value from their labor to sustain this "automated labor machine." Their working conditions, pay, and the value created are vastly mismatched.

The environmental costs are equally alarming. Karen Hao details the impact of the giant data centers (which she calls "supercomputing facilities") being built by AI giants worldwide on local communities, especially vulnerable ones. For example, OpenAI's "Gateway" plan in Abilene, Texas, involves facilities as large as Central Park, operating millions of chips, with power demands exceeding 20% of New York City's total. Meanwhile, Meta is building a supercomputing facility in Louisiana that will be four times the size of the former, consuming power comparable to half of New York City's average energy demand.

When these facilities enter communities, they cause a surge in electricity demand and a decrease in grid reliability. They also require vast amounts of freshwater for power generation and cooling, often competing with already drought-stricken communities for scarce water resources. Worse still, some facilities are built alongside power plants. She cites Elon Musk's construction of the "Colossus" supercomputer in Memphis, Tennessee, to train Grok: this facility uses 35 methane gas turbines and is located in a working-class, Black, and brown community, where residents were not even informed of the facility's existence until they noticed a gas leak-like smell in their living rooms. These turbines emit thousands of tons of toxins into the air, exacerbating children's asthma symptoms and other respiratory diseases, while the community already faces a legacy of environmental racism and a higher incidence of lung cancer.

"This is what I mean by 'the havers' and 'the have-nots' being further torn apart," Karen Hao concludes, "If you unfortunately belong to the 'have-nots,' you might end up with a job that is worse than the original (like data labeling), your air is polluted, bills increase, and the affordability crisis worsens. How can this make people more 'human'?"

Breaking the Empire: What Can the Public Do?

Faced with the seemingly powerful AI empire, Karen Hao does not believe that the public is powerless or "the horse has bolted." On the contrary, she points out that a thriving grassroots movement is currently exerting immense pressure on the empire's agenda. "80% of Americans in the latest poll believe the AI industry needs regulation. When was the last time 80% of Americans agreed on an issue?" She mentions that dozens of protests against data centers have erupted across the United States and even globally, some successfully delaying or stopping project construction.

Her core agenda for action is: Break the empire and develop alternatives.

First, she distinguishes between different types of AI. She compares the current mainstream models, which rely on massive data and computing power for "violent scaling," to a "rocket in AI." They consume enormous resources, providing massive profits to a few while imposing enormous development costs on the broader population. She calls for the development of more "bicycle-style AI" — those AI systems that use refined small datasets, require less computing resources, and can deliver significant benefits at a low cost. She cites DeepMind's AlphaFold (used for predicting protein folding) as an example; this system received the 2024 Nobel Prize in Chemistry for its contribution to accelerating drug discovery, exemplifying "bicycle" style AI.

"My point is not to say that these technologies are without value," she clarifies, "but rather that the political and economic structures currently supporting the production of these technologies are causing significant harm to people. However, research indicates that the same capabilities can be developed through much more efficient and resource-saving methods."

For the general public, she offers specific action suggestions:

• Examine your intersection with AI empire resources: You are a "data donor" to these companies and can consider resisting data being appropriated without compensation by supporting lawsuits of artists and writers.
• Focus on local data center construction: If your community is planning to build a data center, participate in or support local protests and discussions, exercising your democratic rights.
• Engage in AI policy discussions in your environment: Whether in schools or companies, discussions about AI adoption policies are happening. Actively participate, express your concerns, and do not let the adoption process go "smoothly."
• Support alternative AI development: Pay attention to and support those committed to developing more efficient, focused, and socially and environmentally friendly AI technologies and projects.

Finally, Karen Hao emphasizes that being in awe of technology and worrying about its unintended consequences are not contradictory. "This tension doesn't necessarily have to be a tension because we can actually retain the utility and benefits of these technologies but develop and design them in different ways to avoid all these unintended consequences." The key lies in liberating the dialogue about the social and environmental impacts of technology from the industry-led narrative into a broad social discussion and democratic decision-making process. The research, writing, and speaking she engages in are aimed at promoting this crucial dialogue.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。