Book Review: Hao, K., (2025), Empire of AI: Inside the Reckless Race for Total Domination, Allen Lane/Penguin

This impressive book is written by Karen Hao, a mechanical engineering graduate from MIT and a specialist on the impacts of AI on society. She covers many of the same issues as Olsen covers in Supremacy. But while Olsen’s somewhat shorter book discusses the race between the UK-based Deep Mind firm, backed by Google, and San Francisco based Open AI, backed by Microsoft, in more general terms, Hao’s book focuses much more in depth on Open AI and on its CEO, Sam Altman. Many of the other key executives in Open AI are also discussed, including co-founder Ilya Sutskever and the CTO Greg Brockman. The culture of the firm is discussed. Open AI’s board’s role is also examined, and so is the firm’s close link to Microsoft.

While in one way, this book is more focused on Open AI and its strive to be the first (or to perish) when it comes to the development of AI, it is simultaneously also much broader than Olsen’s, in that it also covers other closely related topics, such as the link with Elon Musk, Altman’s stressful relationship with Annie Altman (Sam’s sister), the use of relatively inexpensive ways to acquire large data volumes (Kenya) and the development of new data centers, Critical resources are also discussed, such as extensive uses of energy, as well as of water for cooling. The extra use of rare metals is also a concern, to build the many super-computers needed.

In this book review, we shall not focus so much on these issues, but rather concentrate on Sam Altman and Open AI.

The book falls into four parts. The first includes various issues that characterize the AI industry, the second covers some of the dysfunctional aspects of AI’s development, the third looks at how key industry players, Open AI in particular, seem to be on top, In the final part of the book, there is a detailed discussion of how Open AI’s board fired Altman, for them to rehire him soon thereafter but only after the board had been reconstituted. It should also be noted that Altman’s ousting had been initially discussed in a prologue.

Altman is described as rather manipulative. While he is clearly a good listener, as well as excellent in fund-raising, he comes across as quite calculating, keeping his own concerns at the center. Half-truths and divergent messages to various people seem to be central to his way of managing.

Altman sees the strive to come up with “the best” way to incorporate AI as a battle, with Deep Mind seemingly being the major competitor. So he seems to prefer speedy releases of new approaches, rather than to take the extra time that would be typically called for to “ensure” higher quality. It seems that incremental improvements are acceptable: “learning” from customers.

Although concerned with aspects of the “AI bubble”, Altman creates the impression that “all is under control”. His worry appears to be potential competition from abroad, perhaps especially from China. He calls for legislation to protect his US-based AI industry against this but yet is clear that to regulate the domestic industry would be wrong!

Altman is of course not alone in focusing heavily on speed in Open AI. Perhaps his most senior associate Grey Brockman, who states that to build better AGIs (artificial general intelligence, often referred to as AI, artificial intelligence, in this review) is the key answer for distributing economic benefits. Other main players are Ilya Sutskever, Dario Amadei and Mira Murato, all being more or less critical of Altman.

There are three informal groupings at Open AI, with different foci - “exploring research”, “safety” and “start-up”. Sutskever appears at the helm of the first, Amadei of the second, and Brockman of the third. But it was Atman, above all, who was behind the de-emphasis of “safety”. He pushed very hard on scaling!

Open AI’s launch of ChatGPT in late 2022 was indeed revolutionary. But it was perhaps released too quickly, pushed primarily by Altman. Others, including Amadei, had preferred to go somewhat more slowly, to be safer. The pressure for “quantity” eroded quality ever further in later releases.

Open AI was initially established as a non-profit organization, with a rather free-wheeling academic research environment. But gradually the firm shifted its focus to become more commercial, so as to deliver benefits, which, by the way, had invested more than 1 billion USD. Altman was indeed the lead figure to steer this gradual shift in focus. It was perhaps a natural consequence of this that Open AI started to significantly push back on transparency. Altman himself became paranoid about people leaking information. He did not want to see more emphasis on Open AI’s “unfiltered" product releases.

While much of part two of the book focuses on “marginal” issues, such as Kenya as a source of relatively cheap labor, or Chile as a prime location for data centers, there are also a few observations attributed to Altman that should be highlighted. He appears obsessed with “refounding” this company towards more and more commercialization, away from a more free-wheeling research organization. He wanted Open Ai to be the leading organization, seemingly to feel that “the winner takes all”!

This brings us to part three of the book. The lunch of Chat GPT-4, initially seemed to be a disappointment – the data quality was terrible. But the product was soon improved upon, and presented to key people at Microsoft, such as Bill Gates and Satya Nadella. Gates reportedly said that the GPT-4 demonstration was one of the worst demos he had seen in his life!

The friction between Altman and the board of Open AI was now already brewing. While Altman felt that the board should serve more as advisors to provide inputs for consideration, the board members felt that to “protect” Open AI’s non-profit status was essential. This led to leaning toward quality control.

Altman was virtuous when it came to testifying to Congress, as well as to briefing policymakers in general. As noted, he stressed the need for regulation to contain China. The domestic industry should remain relatively unregulated, to stimulate innovation, he felt.

Internal friction developed further at Open AI, in large part due to inconsistent messaging to internal staff members from Altman, even back-stabbing. This became increasingly toxic, particularly for Sutskever, but gradually also more and more for most of the board’s members.

In early December 2023, the board fired Altman. Reactions from many internal staff, from Microsoft, as well as from Brockman (one of the six board members) was very strong. So, after ten days, “the blip”, was undone and Altman was rehired. Then Sutskever left the board and eventually resigned from the company.

The trend towards faster releases was now promoted even more aggressively by Altman. Models were released more often and earlier than ever before. It was seen as central to get as much initial feedback from customers as possible. Friction between “boomers” such as Altman and Brockman, and “doomers” became more intense than ever. Even though Sutskever was no longer an employee of Open AI, his influence was still felt. There was a broadly shared perception that he had been “squeezed out” in an inappropriate and unfair way.

The release of GPT-5 was a disaster, with the product instantly being withdrawn from the market after only a few weeks. The lack of quality had become too apparent.

The Empire of AI is clearly an important book. But is it too unbalanced, perhaps with too much negative emphasis on Altman? In this reviewer’s opinion, this definitely does not seem to be the case. While many aspects of AI’s development might be positive, such as to be able to undertake faster, more accurate medical diagnoses, for instance, unregulated use of AI, even to substitute humans when it comes to critical decision-making, should definitely not to be called for. Hao points this out in a very clear way.

This reviewer is indeed impressed with the book. More than six years of hard work has gone into the research and writing. It is rare to get this high level of technical insight from an accomplished journalist, as is the case here. I strongly recommend this book.