Written by: Techub News Compilation

Introduction

In an exclusive interview on the independent news program "Democracy Now!", veteran technology journalist and former AI reporter for MIT Technology Review Karen Hao shared key findings from her new book "AI Empire: Sam Altman's Dreams and Nightmares of OpenAI". Using her in-depth visit to OpenAI in 2019 as a starting point, she revealed how the company transformed from a research laboratory labeled as "non-profit" and "benefiting all humanity" into an "empire" pursuing "artificial general intelligence" (AGI) that profoundly influences global politics, economics, and the environment. This discussion delves into the 'quasi-religious' nature of the AGI concept, the impact of AI development on employment and democracy, the evolving landscape of US-China AI competition, and global community resistance against tech giants.

Summary

• AGI is a "quasi-religious" belief: OpenAI and its followers view AGI as an inevitable and imminent "miracle", but this belief lacks scientific evidence and is divided into "doomers" and "boomers", both advocating for technology to be controlled by a few elites.
• AI development path choice: substitution vs. augmentation: Silicon Valley companies like OpenAI focus on developing technology for "substituting labor" to cut costs, rather than creating tools for "augmenting labor" to improve human work quality and social welfare.
• The "empire model" of "expanding at all costs" brings multiple crises: The AI race leads to a frantic grab for computing resources, energy, land, and data, resulting in environmental destruction, the deprivation of community resources, global brain drain, and the erosion of democratic decision-making.
• China's AI "innovation breakthrough" challenges the Silicon Valley narrative: Under US export controls, Chinese companies like DeepSeek have achieved AI capabilities comparable to those of the US at significantly lower computing costs, proving that Silicon Valley's notion of "only massive computing power can drive progress" is not the only path.
• Community resistance and visions for small-scale AI alternatives: From Maori language preservation in New Zealand to water resource struggles in Chile, global communities are fighting back through laws, unions, and public opinion, showcasing the possibilities of AI development based on small-scale, highly curated data that meets specific community needs.

OpenAI's "Quasi-Religious" Mission and Empire Building

Karen Hao pointed out that OpenAI's success largely stems from its meticulously crafted "quasi-religious" mission. Founder Sam Altman once quoted and endorsed the saying: "Successful people build companies, more successful people build nations, and the most successful people build religions," in a blog post in 2013. He later reflected, "In my view, the best way to build a religion is actually to build a company." This provides a key insight into understanding OpenAI's operations.

Hao explained that when Altman decided to establish an AI research laboratory capable of competing with giants like Google, he realized he could not win on funding and first-mover advantage. Therefore, he opted for a non-profit structure and assigned it an extremely lofty and vague mission: "to ensure that AGI benefits all humanity." This mission itself became a powerful "hook" to attract top talent, gain public favor, and build a brand. However, AGI at that time and even now is a concept that lacks a clear definition and scientific consensus, being more based on a belief that human intelligence is fundamentally computable, and with enough data and computing power it can inevitably be replicated.

Inside OpenAI, a small group of "self-identifying AGI believers" emerged, split into two factions: "Boomers" believe AGI will lead humanity into a utopia; while "Doomers" staunchly believe AGI will annihilate humanity. Ironically, both factions are based on the same belief—that AGI is possible and imminent—and reach the same conclusion: it must be controlled by them (not the masses), and cannot be democratized. This intense internal conflict over the pace of technological development and release partially led to the dramatic event of Altman being briefly ousted and then reinstated by the board.

Hao emphasized that this "quasi-religious" nature allows OpenAI to shape the definition and development path of AGI in the way that is most convenient for itself, without the need for strict scientific validation or public scrutiny. The AGI it defines is "a highly autonomous system that surpasses humans in most economically valuable work." This essentially sets the goal of "automating human work", rather than enhancing human capabilities.

Substitution vs. Augmentation: The Chosen Path of AI Development

Karen Hao sharply pointed out that the current AI development model represented by OpenAI fundamentally revolves around "substitution" rather than "augmentation". She quoted the views from MIT economists Daron Acemoglu and Simon Johnson in their book "Power and Progress": the technological revolution towards "substituting labor" is not inevitable, but a result of the choices made by those in power.

"These companies pitch to corporate executives saying: 'You can cut all these workers and use our AI services to reduce costs,'" Hao said. This business model has had a tangible impact on employment: not because the technology itself is mature enough, but because executives "think" it can replace human labor, leading them to start layoffs.

However, historical research and real-world cases show that "augmentation" AI tools often yield better social outcomes. For instance, developing AI diagnostic tools for doctors rather than attempting to replace doctors can lead to better healthcare outcomes and cancer diagnosis rates. Developing AI teaching assistants for teachers, instead of AI tutors to replace teachers, can produce better educational results. But Silicon Valley companies tend to develop "universal machines" positioned as tools to replace professionals (like therapists), even though this can lead to the serious dissemination of medical misinformation and psychological harm, and there have already been tragic cases of children committing suicide after forming emotional dependencies on chatbots.

Hao believes we need more "guardrails" to guide companies towards developing "augmentation" technologies, but this contradicts the current logic of maximizing profit and market monopolization.

The Cost of Empire: Resource Extraction, Environmental Crisis, and Erosion of Democracy

To support the fervent pursuit of AGI and the enormous computing power demand, companies like OpenAI are frantically extracting resources globally. Hao cited an OpenAI employee saying: "We are depleting land and electricity." This has driven Altman to actively lobby the US government and form agreements with resource-rich regions like the Middle East to acquire land and energy for constructing vast data centers.

This expansion has brought severe environmental and social costs. Hao mentioned two typical cases: one is Elon Musk's xAI building the "Colossus" supercomputer in Memphis, Tennessee, used to train the Grok chatbot. This facility is powered by about 35 unlicensed methane gas turbines, emitting thousands of tons of toxic air pollutants into the Memphis community, which already suffers from a long-standing lack of clean air as a fundamental human right. The second is Meta secretly advancing data center construction under the shell company "Greater Kudu LLC" in New Mexico, revealing its identity only after the deal was completed and residents could no longer oppose it, consuming significant amounts of local freshwater.

In response to the criticism of high energy consumption, Altman's proposed solution during his congressional testimony was to utilize more natural gas in the short term and rely on advanced nuclear fission and fusion in the medium term. Hao is deeply concerned about this, noting that these companies repeatedly cite nuclear energy as the "ultimate solution" and lobby governments to relax regulatory frameworks for nuclear power plant construction, which represents a dangerous extension of the "rapid action, break the mold" ideology into critical infrastructure sectors.

The more fundamental crisis lies in the erosion of democracy. Hao pointed out that Silicon Valley has successfully made the public feel over the past decade that resources that should belong to the collective—such as personal data, artists' intellectual property, community land and water, schools, and hospitals—ought to be utilized by tech companies. OpenAI even launched the "National Version of OpenAI" initiative, claiming to install its hardware and software globally as a "democratic AI track" to prevent China from deploying "authoritarian AI". However, the executive editor of The Atlantic astutely pointed out that these companies themselves are "technological authoritarianists", and they never seek public opinion on how to develop technology, what data to use, or where to build data centers.

"AI is threatening democracy," Hao summarized, "If most people in the world feel they have lost the agency to determine their own futures, democracy cannot survive."

US-China AI Competition: Narrative Disruption and Talent Reverse Flow

Karen Hao has long focused on US-China AI development, providing a perspective that disrupts the mainstream narrative in Silicon Valley. In recent years, the US government, at the suggestion of American tech companies, has attempted to stifle China's AI development through export controls on cutting-edge computing chips (mainly designed by Nvidia), aiming to maintain its lead.

However, the results were unexpected. China has a strong and dense pool of AI research talent. In the face of limited computing resources, Chinese companies have been forced to innovate, achieving comparable AI capabilities to those of US companies at much lower costs. Hao highlighted the model released earlier this year by Chinese company DeepSeek. The company stated that the cost of training this model was only about $6 million, while OpenAI's model training costs often reach hundreds of millions or even billions of dollars. The magnitude of this gap demonstrates to the world that the path Silicon Valley touts—that "only immense computing power can enhance AI capabilities"—is not the only truth. The technology employed by DeepSeek mostly exists in academic literature, seamlessly integrated through superior engineering capabilities, without the use of fundamentally new technologies.

Behind this achievement lies a "reverse flow" of talent caused by US policies. Hao recalled that initially, more Chinese researchers were working in the US, contributing to American AI. However, policies like the "China Initiative" began to criminalize Chinese scholars (including American citizens) by accusing them of being spies. Subsequently, the pandemic, trade frictions, and anti-China rhetoric intensified, along with the Trump administration's potential policies to ban international students (particularly Chinese students), prompting more top Chinese talents to choose to stay in China and contribute to the Chinese AI ecosystem. The success of DeepSeek is a direct outcome of this talent gathering and alienation policy.

Hao warned that this trend is accelerating America's "brain drain". Not only Chinese researchers, but even domestic scholars from the US are considering moving to Europe due to cuts in public funding, while European countries are seizing the opportunity to provide substantial funding and laboratories to attract global talent.

Resistance and Hope: Community Agency and Visions for Small-Scale AI

Despite painting a grim picture, Karen Hao also records resistance movements around the globe in her book and proposes alternative visions for AI development.

She recounts an inspiring case: New Zealand's non-profit Maori radio Te Hiku Media. To revive the endangered Maori language, they aimed to transcribe their ancestors' recorded audio into written form as a learning resource. Lacking proficient transcribers, they turned to AI. However, their approach starkly deviated from Silicon Valley's: they first sought community consent, then clearly explained the purpose of the data and obtained fully informed consent, using only a few hundred hours of high-quality audio data donated by the community to train an excellent speech recognition model. This model was later open-sourced to serve the community's language revitalization goals.

This case demonstrates several key points: small-scale, carefully curated datasets can train powerful AI models; community-led, informed consent data collection is feasible; and AI can serve specific, real human needs rather than vague "general" objectives.

Hao observes that from artists suing companies for intellectual property infringement to Chilean water protectors struggling to defend freshwater, and to Kenyan workers labeling data for OpenAI organizing unions and reaching out to international media, seemingly disadvantaged communities worldwide are reclaiming their "agency". They recognize that data, land, water resources, schools, and hospitals are arenas of democratic contestation that cannot be easily relinquished.

"Ultimately, I realized that the first step to reclaiming democracy is to remember that no one can take away your agency," Hao concluded. Her vision is for the future to feature more small, task-specific AI models based on high-quality small datasets, needing minimal computing resources and can genuinely be used to tackle human challenges, such as integrating more renewable energy into the grid to mitigate climate change or conducting drug discovery to improve healthcare. This represents a path distinctly different from the "expanding at all costs" empire model.

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。