Paul Graham, founder of the tech accelerator Y Combinator and current CEO of OpenAI, Sam Altman, sent a reply to a tweet from the former YC President in April. Altman had just said goodbye to GPT-4 in public on X. Graham had a question.
If you had [GPT-4’s model weights] inscribed on a metal piece in its most compressed form, Graham wrotereferring to values that determine the behavior of the model. “How big would the metal piece have to be?” This is a serious question. These models are history and digital data will disappear by default.”
It is clear that OpenAI achieved something historic in 2022 with the release of ChatGPT version 3.5. It has set off an AI arms race which has already altered the world in many ways. The long-term effects of this AI arms race will be even greater than the short-term disruptions in areas like education or employment we are already seeing. We are still pondering the future of humanity and will likely be doing so for some time. Two recent books attempt to understand it by describing what two leading technology reporters saw at the OpenAI Revolution. Karen Hao, in Empire of AI : Dreams and Nightmares In Sam Altman’s OpenAI (19459017), tells the story of OpenAI’s rise to prominence and its impact on the world. Keach Hagey of the Wall Street Journal (19459017) wrote Optimist: Sam Altman and OpenAI in the Race to Invent the Future, which focuses on Altman’s life from his childhood to the present. Both portray Altman as a brilliantly successful yet deeply flawed creature from Silicon Valley, someone who can always get what he wants but often by manipulating other people. Hao, a former reporter for MIT Technology Review,began reporting on OpenAI when she was at this publication. She is still an occasional contributor. This reporting was the basis for one chapter of her book. In fact, Hao states in the acknowledgements of Empire of AIthat some of her reporting on MIT Technology Review– a series on AI imperialism – “laid the foundation for the thesis and ultimately, the title of the book.” You can take this to mean we are predisposed towards Hao’s works.
Having said that, Empire of AIa powerful book, bristling with not only great reporting but big ideas. This is done in service of two main themes.
First, it is a story about ambition trumps ethics. Hao’s (and Hagey’s) version of OpenAI’s history is a story of a company founded with the idealistic goal of creating a safety-focused AI general intelligence, but which instead became more focused on winning. This is a familiar story in Big Tech. Google is the closest analogy, as it went from “Don’t be evil” (at least to the courts) to an illegal monopolist. Consider how Google went from being cautious about releasing their language model to the public to rushing to release a chatbot to beat OpenAI. No matter what the original intention, winning is always the goal in Silicon Valley.
Hao’s book has a second theme that is more complex. It is called AI colonialism. The idea is that large AI companies behave like traditional empires by syphoning wealth away from the bottom rungs in the form of labor, creative work, raw materials, etc. to fuel their ambitions and enrich those on top of the ladder. She writes, “I have found only one metaphor to describe what these AI power-players are: empires.”
During the long period of European colonialism empires seized resources that were not owned by them and exploited labor from the people they subjugated in order to mine, grow, and refine these resources for the empires’ enrichment. She then chronicles her own growing disillusionment. She writes: “With increasing clarity, I realized that the revolution promising a better future for people on society’s margins was instead, reviving darkest remnants from the past.” To document this, Hao leaves her desk and travels to the world to see how this empire is spreading across the planet. She travels to Colombia in order to meet data labelers who are tasked with teaching AI how to interpret images. One of them she describes running back to her apartment to earn a few dollars. She documents how workers who moderated content for OpenAI in Kenya, who were responsible for data-labeling, became traumatized after seeing so much disturbing material. She documents in Chile how the industry uses precious resources – water, power, copper and lithium – to build data centers.
Her focus is on how people are fighting back against AI empires around the world. Hao draws on New Zealand where Maoris are trying to save their language by creating a small language-model of their own. It’s based on voice recordings of volunteers and runs on only two graphics processing units (GPUs), not the thousands used by OpenAI. It’s designed to benefit the community rather than exploit it. Hao writes in
that she is not anti-AI. Hao writes: “What i reject is the dangerous idea that AI benefits can only be gained by a vision that demands the complete capitulation our privacy, agency, and worth, including our labor and art to an imperial centralization project. will everarise from this…. [The New Zealand model] demonstrates another way. It shows how AI can be the exact opposite. The models can be small, task-specific and their training data contained, allowing them to eliminate the incentive for widespread, exploitative, and psychologically damaging labor practices, as well as the all-consuming extractivism that comes with producing and running massive computers. She also focuses on the OpenAI CEO’s attempt to build an empire. Indeed, “Altman’s departure from YC had not slowed his civilization-building ambitions,” Hagey writes. She then goes on to describe how Altman, a man who had considered running for governor of California in the past, set up income distribution experiments via Tools for Humanity – the parent company of Worldcoin. She quotes Altman as saying, “I thought that it would be interesting to find out… how far technology could achieve some of the goals which used to be accomplished by nation-states.” Hagey’s book is jam-packed with insider scoops, insights and intrigue from behind the scenes. The book is a great read, especially the second half when OpenAI takes over. Hagey seems to have had more access to Altman, his inner circle, both personally and professionally, than Hao, which allows for a more complete telling of the CEO’s story. Both writers, for example, cover the tragic tale of Altman’s older sister Annie, who was estranged from the family and made accusations about being sexually abused by Sam (something that he and the Altman family deny vehemently). Hagey’s account provides a more nuanced view of the situation and gives more insight into the family dynamics. Hagey ends by describing Altman’s reckoning with the long arc in human history, and what it means to create a “superintelligence.” His place in this sweep is clearly something that has consumed the CEO’s thoughts. Altman was ready to respond when Paul Graham asked him about preserving GPT-4. Altman replied that they had already thought about thisand that the sheet metal would have to be 100 meters in size.
