Empire of AI

“Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI” by Karen Hao is an exhaustively researched look into one of the most important AI companies in Silicon Valley. Here's what I learned by reading it.

1. Sam Altman, one of the most important founders of what was originally a not for profit organization, is an insanely ambitious business leader who has a gift for fundraising, but is wholly untrustworthy as a manager, colleague, or any other kind of associate. He'll tell anyone whatever benefits him at a particular moment in furtherance of his aims, which are: to win the race to Artificial General Intelligence (AGI), and enrich himself to levels that are incomprehensible to most humans. He was famously fired for his duplicitousness, but eventually re-admitted because OpenAI's staff rebelled, feeling he was the spiritual leader they needed to be successful. No staff, no company.

2. AI is an incredibly resource-extractive technology. The data centers it relies on require huge amounts of electricity to run the CPUs and billions of gallons of water for cooling. Because of this, many AI data centers are being located in the Global South where local authorities are easily bought off and citizens have little power to fight back. Data centers require millions of gallons of not just water, but *potable* water for cooling, water that human beings need to survive. AI will make climate change worse by dramatically increasing global power consumption and absorbing resources needed for life of all kinds. All so you can get a quick answer to a question, or to have a program in the cloud write your term paper for you.

3. OpenAI was staffed by sections dedicated to AI safety, that is, ensuring that the technology could be developed in such a way that it is "aligned with human values" and will not lead to an existential risk for humanity. Beyond the debate over whether AGI is achievable, lies the debate over whether its creation would lead to the extinction of human life. This is not a science fiction scenario. Many people closest to the cutting edge of this tech believe this is possible. What OpenAI did over time was to reduce its attention to the Safety group and in some cases marginalize it...lest OpenAI lose the race to AGI. Basically, safety simply gets in the way. An old story in the development of any prized technology.

4. There's another human cost to AI that's largely hidden from the public: the poor wage laborers who are used to "train" the AI models, stripping out hideous sexual content and violent fantasy material. Guard rails have to be in place to ensure that the public can't ask ChatGPT to generate Child Sexual Abuse Material (CSAM). The only way to train the model is to force humans to strain such material out. The labor is menial, pays nearly nothing, and leaves human beings with broken psyches behind. People can't look at this kind of material for months on end without becoming deeply disturbed. Again,. all of this takes place largely in the Global South where wealthy northerners don't have to be "dirtied" by seeing it. We always shift the ugliest parts of our economy to places where we don't have to see it so out consciences won't be bothered.

One star off only because of the time dedicated to the "palace intrigue" of various managers at OpenAI, the descriptions of which became repetitious and tedious to read.