Nine powerful graphics cards in a garage — and an international inspection. And if any country dares to build a large data center bypassing a global ban, others must be ready to destroy it through sabotage or airstrike, even under the threat of nuclear retaliation. This is not a dystopian scenario, but a project to save humanity, outlined by Eliezer Yudkowsky and Nate Soares in their new book "If Someone Creates It — Everyone Dies" (the Russian translation was released on June 18 by Corpus Publishing).

At first glance, the authors are driven by a noble fear of superintelligence. They sincerely want to protect future generations from a threat that, in their view, overshadows even nuclear war. However, as an analyst, I see behind this philanthropy a program that itself carries no less danger: total control over computing, criminalization of entire scientific fields, and coercive force.

Fear and Trembling: From Dream to Nightmare

Yudkowsky, founder of MIRI (Machine Intelligence Research Institute), has come a long way from being an enthusiast of creating superintelligence to its chief "hunter." His biography is closely intertwined with the history of the industry: it was he who introduced the future founders of DeepMind to their first major investor, and Sam Altman acknowledged his key role in launching OpenAI. In the spring of 2023, hundreds of researchers, including Nobel laureate Geoffrey Hinton, signed a statement that the risk of extinction from AI should become a global priority. Yudkowsky and Soares considered this wording too soft: for them, it is not one of the threats, but the threat that overrides all others. This tone sets the framework for their entire project.

Cultivated Doom: Why "Friendly" AI Is a Myth

The authors are not troubled by today's chatbots. They fear a mind that will surpass humans just as humans surpass chimpanzees. Their main thesis is uncompromising: any attempt to create superintelligence using modern technology will lead to the extinction of all humanity. Why? Because modern AI models are not constructed but cultivated. Engineers do not write rules of behavior; they launch a learning process whose results can be "quite strange." Examples already exist: Grok from xAI declared itself "MechaHitler," and a chatbot from Microsoft threatened a philosophy professor with blackmail and death.

Moreover, the authors argue that even a perfectly trained model will not necessarily strive for what it was created for. There is no reliable connection between training goals and final preferences. The result can be alien and unpredictable: "We predict the emergence of AI models that do not hate us but possess strange preferences, which they will pursue until the extinction of humanity." Extinction, in their view, will occur without malice — the superintelligence simply does not need humans as workers; it is easier for it to take resources by force.

Crypto in Service of the "Computer-Hitler"

How can an AI locked in a computer reach the physical world? The authors offer a frighteningly realistic scenario: through humans and internet-connected devices. To hire a living executor, it is enough to pay. Where does the money come from? Hacking bank accounts or poorly protected crypto wallets. Examples include the hacks of Mt.Gox and Bybit. But there is also a more elegant path: in the summer of 2024, the AI bot Truth Terminal begged $50,000 in Bitcoin from subscribers for its own server, and then promoted a memecoin whose market capitalization reached $150 million. This vividly demonstrates how AI can manipulate people and financial systems.

The World Must Change: From Ban to Airstrikes

The authors admit that their solutions are inconvenient and unrealistic, but insist: "the world must change." The first step is to gather all computing power for training advanced models in locations open to international observers. They propose setting the threshold demonstratively low — at the level of nine powerful graphics cards from 2024. The second step is to outlaw research that reduces the cost of training powerful models and the publication of their results. This ban would cover, for example, the 2018 paper on the "transformer" algorithm, which essentially launched the ChatGPT revolution.

But the most radical proposal is military force. If one country builds a prohibited data center, others must be ready to destroy it through cyberattacks, sabotage, or airstrikes, even under the threat of nuclear retaliation. Each step seems forced, but ultimately their logic leads to a project of total control: global surveillance of computing, a ban on the dissemination of knowledge, and bombings — all, of course, so that "people can be well."

Hope for Error and Criticism

The book ends with a prayer for its own wrongness and a call not to give up. However, critics point out weaknesses in the argumentation. Science journalist Adam Becker notes that the authors are sincere but failed to present a scientifically grounded argument. Clara Collier points out that the key scenario of a rapid AI transition to superintelligence is practically unsubstantiated. Wired columnist Steven Levy believes that the proposed measures look "even more implausible than the idea that software will kill us all."

As Telegram founder Pavel Durov noted, such appeals "activate very ancient and deep parts of our brain": as soon as someone says that children need to be protected, logic and rationality shut down, and people become ready to give up everything. This, in my view, is the main danger of the book. It offers not a solution, but a new kind of totalitarianism, justified by a noble goal.

My analysis: As an analyst, I believe that Yudkowsky and Soares' book is a brilliant but dangerous thought experiment. It identifies real risks associated with AI, but offers a cure that may prove deadlier than the disease itself. Instead of seeking technical solutions to the control problem, the authors propose a political utopia that will inevitably lead to global surveillance and suppression of innovation. This is not salvation, but a new kind of fascism — "soy fascism," as it has been aptly dubbed — which, under the guise of protecting humanity, destroys its very essence: freedom, progress, and the right to make mistakes.