Brief Summary
Karen Hao's "Empire of AI" is examined, revealing the inner workings of OpenAI and the broader AI industry. The discussion covers the environmental and social costs of AI development, the major players in the field, and the exploitative practices in the Global South. It also addresses the quasi-religious fervor driving AI development, the role of key figures like Sam Altman, and the potential threats to democracy.
- AI is poorly defined and encompasses various technologies simulating human tasks.
- Deep learning, a subset of machine learning, is the most common type of AI used by major companies.
- The AI industry's resource consumption is accelerating climate and public health crises.
- OpenAI, co-founded by Elon Musk and Sam Altman, started as a nonprofit but shifted to a for-profit model.
- Sam Altman, a master manipulator, has cultivated powerful networks in Silicon Valley.
- Exploitative practices in the Global South, such as content moderation in Kenya and data annotation in Venezuela, are prevalent.
- The US government's hands-off approach to AI regulation is concerning.
- Individuals can play an active role in shaping the AI development trajectory by contesting the use of resources and spaces.
Intro
The discussion introduces Karen Hao, an MIT graduate and journalist, who has written "Empire of AI," an inside account of OpenAI and the perils of artificial intelligence. Hao's work is based on 300 interviews, including 150 with OpenAI employees, and access to company emails and Slack channels. The book explores the burgeoning AI industry in the United States and its potential impact on society.
AI, Deep Learning & Machine Learning Explained
Artificial intelligence is poorly defined, referring to technologies simulating human behaviors or tasks. The term was coined in 1956 to attract attention and funding. AI encompasses a range of technologies, from Siri to Chat GBT, which operate differently and have varying resource consumption levels. Deep learning systems, which train on large datasets and use neural networks to calculate patterns, are commonly used by companies like Meta and OpenAI. Deep learning is a subcategory of machine learning, which is a branch of AI focused on building software that calculates patterns in data.
Data Centres & Resource Consumption
The resource consumption required to develop and use generative AI models is significant. Data centers globally account for about 3-3.5% of CO2 emissions, with AI data centers being a growing fraction of that. A McKinsey report projects that the current pace of data center expansion for AI development will require adding half to 1.2 times the amount of energy consumed in the UK annually to the global grid in the next 5 years, largely serviced by fossil fuels. Elon Musk's XAI supercomputer in Memphis, Tennessee, is powered by unlicensed methane gas turbines, pumping toxic air pollutants into the community. Data centers use fresh water for cooling, often drawing from public drinking water supplies, exacerbating water scarcity in communities.
Who Are The Major Players
The major players in the AI space in the United States include OpenAI, Anthropic, Google, Meta, Microsoft, Super Safe Super Intelligence, Thinking Machines Lab, and Amazon. These companies are racing to deploy AI technologies, particularly artificial general intelligence (AGI). In China, companies like ByteDance, Alibaba, Baidu, Huawei, and Tencent are building chatbots similar to Chat GBT. Europe has smaller players like Mistral in France. The business case for AI is unclear, with Microsoft pulling back investments in data centers. The fervor is driven by the belief in recreating human intelligence, rather than purely financial incentives.
What’s Behind Open AI’s Rapid Rise
OpenAI started as a nonprofit in 2015, co-founded by Elon Musk and Sam Altman, to create an AI research lab without commercial pressures. They aimed to be the anti-Silicon Valley, focusing on openness, transparency, and collaboration. The initial bottleneck was talent, and the nonprofit status was a recruitment tool, competing on a sense of mission rather than salaries. The shift to a for-profit model occurred when they realized the need for massive capital investment to compete with Google. This led to a falling out between Musk and Altman, with Altman becoming CEO. The term "open" in OpenAI originally stood for open source, but this changed as they pursued scale.
Who Is Sam Altman?
Sam Altman has spent his entire career in Silicon Valley. He was a startup founder and part of the first batch of companies that joined Y Combinator. He later became president of YC, expanding its portfolio to include hard tech engineering challenges. Altman is described as a master manipulator and a once-in-a-generation fundraising talent. He cultivated relationships with politicians early on, positioning himself as a gateway into Silicon Valley. Despite being described as a people pleaser and conflict-averse, Altman has led OpenAI to a significant valuation. He understands human psychology, acquires talent, and persuades others to join his vision.
Who Is Karen Hao?
Karen Hao studied mechanical engineering at MIT and worked in Silicon Valley before transitioning to journalism. Her interest in sustainability and mitigating climate change led her to realize that Silicon Valley's incentive structures were not aligned with the public interest. She worked at MIT Technology Review and the Wall Street Journal, covering AI. Her engineering background helps her have honest conversations with people in Silicon Valley and understand the technology. She believes that technology is a product of human choices and that a small group of people should not have too much power to develop technologies affecting billions of lives.
Colonial Similarities To AI Companies
OpenAI's actions are compared to those of the East India Company, a corporate empire abetted by the British Empire. The US government, under the Trump administration, sees corporate empires like OpenAI as empire-building assets. There's a tenuous alliance between Silicon Valley and Washington, with each trying to use the other. Silicon Valley has a growing popularity of the idea of a politics of exit, suggesting that democracy doesn't work anymore. Tech leaders see territories as resources to be acquired, similar to older empires. Data center expansion often occurs in economically vulnerable communities that are not fully informed about the costs.
DeepSeek
DeepSeek, a Chinese AI model, matched and even exceeded some performance metrics of American models with significantly fewer computational resources. This demonstrates that these capabilities can be achieved with less compute. However, American companies have shown an unwillingness to adopt these techniques. The scaling approach is maintained to monopolize the technology and lock out others. Path dependence also plays a role, as companies are not nimble enough to switch to different approaches. Stable diffusion, developed by a European academic, also showed that image generation could be achieved with fewer resources, but OpenAI continued to pursue massive scaling approaches.
Exploitative Practices In Global South
OpenAI contracted workers in Kenya to build a content moderation filter, exposing them to reams of the worst text on the internet and AI-generated text. These workers suffered trauma and PTSD, with their families and communities also affected. The workers were paid a few dollars an hour. Generative AI is not the only thing that leads to data annotation. In Colombia, Venezuelan refugees were exploited for data annotation labeling for self-driving cars, working under extreme conditions for low pay. These stories highlight the logic of empire, where workers in the Global South are paid pennies while those in the Global North receive million-dollar compensation packages.
How To Stop Big Tech
Individuals worldwide can play an active role in shaping the AI development trajectory. Resources and spaces needed by AI companies are places of democratic contestation. Artists and writers are suing companies for using their intellectual property. People are exercising their data privacy rights. Communities are pushing back against data center development. Teachers and students are debating the use of AI in schools. Employees should participate in drafting AI policies in their workplaces. By democratically contesting every stage of the AI development and deployment pipeline, it is possible to reverse the imperial conquest of these companies and move towards a more broadly beneficial trajectory for AI development.
