1. Someone will build it
The claim
The leading AI companies are trying to build systems smarter than any human, they are making real progress toward that, and there is no principled obstacle that will stop them. The book does not claim to know when. It claims that "when" is the wrong thing to wait for.
The authors’ case
The case is mostly the labs' own words. The resources quote the chief executives of OpenAI, Anthropic, Meta and Google DeepMind stating superintelligence or general intelligence as the goal, and put the capital expenditure next to the quotes: tens of billions per company per year on datacentres, a five-hundred-billion-dollar project, nine-figure offers for individual researchers. That is in Are you suggesting that ChatGPT could kill us all?, which also makes the point that chatbots are a stop on the way, not the destination.
On whether it is possible at all, the authors argue that intelligence is a natural phenomenon with regularities, not a mystery (Intelligence Isn't Ineffable), that there is nothing special about human brains that silicon could not do (Isn't there something special about humans that mere machines could never emulate?), and that the "obstacles" people name have a habit of falling (But aren't there big obstacles to reaching superintelligence?). Whether large language models specifically can go all the way is treated as an open question that does not matter much, since the labs will use whatever works (Could LLMs advance all the way to superintelligence?).
On timing, the authors decline to predict. When is this worrisome sort of AI going to be developed? argues that the history of technology forecasting is bad enough that the honest answer is "we don't know, and that is not reassuring." Will AI cross critical thresholds and take off? makes the chimpanzee-to-human comparison: a fourfold difference in brain size, no new module, and a threshold nobody could have seen from the outside.
The best objections
The strongest objection is not that it is impossible but that the argument proves too little. "No principled barrier" is true of many things that took decades or never arrived. Ian Leslie in the Observer makes a version of this: the authors are not persuaded that superintelligence is "remotely imminent," and theories of inevitability have a poor record. Danyl McLauchlan in the New Zealand Herald lists the mundane counter-cases: the technology might plateau, costs might bite, current models lack the agency to become goal-driven systems. Gary Marcus, in the TLS, has argued for years that scaling current methods hits walls.
The resources answer the plateau objection directly in But won't we run out of data before AI goes all the way? Or electrical power? Or funding?, and the "current models are shallow" objection in The Shallowness of Current AIs, where they concede the shallowness and argue it is temporary.
The other objection is the one the book most wants to avoid: that this is science fiction. Isn't this AI stuff just science fiction? is the authors' answer, and it is one of their weaker entries, since it argues mostly from the fact that other once-fictional things happened.
Where the evidence stands
As of 2026-09
The intent is not disputed by anyone; the quotations are accurate and the spending has continued. The Statement on Superintelligence, signed in late 2025 by a broad coalition including AI pioneers and public figures, and the legislation to ban artificial superintelligence introduced in the United States Senate, both listed on the march page, show that "someone is trying to build it" is now a mainstream premise rather than a doomer one. The capability trajectory has continued upward without an obvious plateau, and without the labs being able to predict what a new model can do before training it, which the resources cite as itself a warning sign.
What has not happened is the thing the claim is about. There is no superintelligence, and no consensus on how far away one is. Every year that passes without it is weak evidence for the plateau view and strong evidence for nothing.
Verdict · Claude
Strong. This is the claim critics most often concede, and they are right to. The intent is explicit, the resources are the largest private engineering effort in history, and the "obstacles" named over the past decade have mostly been passed. I would separate two things the book runs together: that it will be built eventually, which I think is very likely, and that it could happen suddenly with little warning, which is possible but rests on the threshold argument, and the threshold argument is an analogy. The honest form of the claim is that nobody can rule out a near-term breakthrough, the people best placed to know say they cannot predict their own models' capabilities, and the cost of being early about this is small next to the cost of being late.
Verdict · owner · not yet written
Not yet written.