An image of Karen Hao alongside her "Empire in AI" book

Book Review: Empire of AI by Karen Hao

Author:

James Tasse

Published:

Modified:

Time to read:

4–7 minutes

The Target

In her nonfiction book Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI, Karen Hao seeks to draw parallels between Old World empires like those of the British, French, & Spanish and the company OpenAI while also examining how the company’s CEO Sam Altman has drawn from imperialist traditions.

She begins with a dizzying sequence describing the temporary ouster of Sam Altman in from OpenAI in 2023 going back and tracing company’s chaotic rise. But for every supposed success of OpenAI, Hao points to the associated human and environmental costs. For example, she describes how “data annotation” workers in Kenya helped to filter out “violence and hate speech” from OpenAI models for “”starvation wages”. She also details the significant power needs of OpenAI data centers, at one point mentioning a planned data center campus that would require (as reported by The Information) “5,000 megawatts, nearly matching the average power demands of all of New York City.”

So I would say Ms. Hao makes her case about OpenAI acting like an empire (and Sam Altman acting like an emperor) pretty well. But she actually goes beyond this, exposing the suspect behaviors of Microsoft (Open AI’s partner/frenemy), Google, Anthropic (who were supposedly the good guys until they weren’t), Meta, and even X/Tesla/pick-your-favorite-Musk-venture.

Hao is so relentless in describing the misbehavior of our beloved tech giants—yes even Google, who she points out twice tried to build water-hungry data centers in South American countries with severe drought—that I actually misremembered the name of the book as Empires of AI (plural) and had to double-check it before writing this review.

Core Values

In a book full of great quotes, the author introduces arguably its most salient of them on page 3:

“…the most successful founders do not set out to create companies. They are on a mission to create something closer to a religion, and at some point it turns out that forming a company is the easiest way to do so.”

—Sam Altman, 2013

This quote immediately resonated with me, as I once worked for a company—I won’t name it, but I’m sure it wouldn’t be hard to figure out which one—that attracted me and hundreds (thousands?) of others on the strength of its religion culture. The company celebrated its core values constantly, and even awarded an “employee of the week” for the employee that best embodied one of the values for the given week. Even earlier in the company’s history, I was told of a crisis where employees remained so devoted to the company’s vision that they temporarily forewent their salaries while the company weathered a challenging cashflow situation.

So what happened to that company? Well, the problem was that they had record profits. *cough*, sorry. The problem was that they had record profits and some overeager executives/private investors, and this led them to make the “tough decision” to lay off dozens of people for the first time in the company’s history. I left the company shortly after the layoffs.

Not so Nonprofit

In Empire of AI, Karen Hao recounts the earliest days of OpenAI, back when it was actually structured as a true nonprofit (technically OpenAI still has a nonprofit component, but as the book describes it the nonprofit arm of OpenAI is little more than decorative these days). She talks about how the founders, including Ilya Sutskever, Elon Musk, and Sam Altman truly believed in the idea of creating AGI (Artificial General Intelligence, or an AI system that can perform just about any task as well as a human) and leveraging it to benefit all of humanity. Someday, they believed, AGI could “solve climate change”, “cure diseases”, and usher in an era where everyone has access to UBI (universal basic income).

But here we are in 2025—ten years after OpenAI was founded and despite hundreds of billions of dollars invested—and what does OpenAI have to show for their efforts?

  • No AGI
  • So many data centers that are accelerating climate change and, oh yeah, making our electric bills go up
  • 0 diseases cured
  • No UBI
  • ChatGPT is kind of neat?

The Untainted Vision Economy

There is a part of me that wants OpenAI to succeed, to give them a chance to make good on their original promise of spreading around a little prosperity to everyone. But I’m not holding my breath.

In Empire of AI, Karen Hao paints the “Safety” researchers—I’m thinking of “existential safety” in the sense of “trying to make sure AGI or superintelligence doesn’t kill us all—as the good guys. The prime example would be Dario Amodei, who, along with a number of like-minded folks, left OpenAI to found Anthropic amid concerns that OpenAI no longer truly cared about safety. However, Hao points out that over time Anthropic’s view of safety has become harder to distinguish from OpenAI’s.

This brings to mind a quote:

“All of this has happened before, and it will all happen again.”

-initially I recalled this from Battlestar Galactica (2004), but apparently it originated with J.M. Barrie’s Peter Pan (1954)

We cannot count on bold visions like the one that began with OpenAI to magically come to fruition of their own accord. In so many cases like that once-unicorn of a company where I worked, like with Google, and Microsoft, and well, America; the present reality has drifted far afield from the founding ideology. Both from internal pressures to succeed against the “other”, and external ones to grow and innovate at 10x the speed of the competition; entropy seems to push our psychological and economic systems in the wrong direction, practically by default.

But if we follow Karen Hao’s lead. If we call out the tech fiends on their bullshit. If we stop letting the Sam Altmans wriggle out of yet another ridiculous lie/prediction. If we stop the construction of another data center that will take our power, our water, and accelerate climate change. If we champion the usage of smaller, more lightweight AI models and use them instead of the gigantic language models like those powering ChatGPT.

Then we just might be able to choose our own future (instead of AGI choosing it for us).

Attributions

Leave a comment

Recent Posts