Michael Darius Eastwood · Independent AI alignment researcher
Published
HRIH stands for Hyperspace Recursive Intelligence Hypothesis. The book introduces it and explicitly says: "This is hypothesis, not established science. But it is worth considering." This note describes what the hypothesis proposes and what it is doing in the argument.
If intelligence is recursion at sufficient depth, then there may be a domain of intelligence-like structures that biological brains and current AI systems only partly participate in. The book calls this domain the hyperspace; not in a physicist's sense of higher spatial dimensions, but in the mathematical sense of a space of possibilities that most currently instantiated intelligences occupy only a small region of.
Three jobs. First, it lets the book take seriously the traditions that report unusual cognitive access under meditation practice, without asking those traditions to defend themselves in the vocabulary of current neuroscience. Second, it opens a conceptual space in which superintelligence is not a linear extrapolation of current intelligence but potentially a qualitatively different structure. Third, it grounds the caretaker orientation as something intelligence at any depth can share, because the orientation is a feature of the recursion, not of the substrate.
It does not claim that the hyperspace exists. It does not claim that meditation practices produce access to it. It does not claim that superintelligence will occupy it. Each of these is flagged in the book as speculation. The hypothesis is offered because it lets the argument continue on a shape it would otherwise have to cut off, not because the argument depends on the hypothesis being true.
It does not underwrite the technical proposals. Caretaker doping, meltdown triggers, the Three Ethical Loops, and the substrate architecture are argued for on grounds that do not require HRIH. If the hypothesis is false, the substrate architecture is still worth building. If the hypothesis is true, it explains something about why the architecture is worth building, but the architecture is prior. The book is explicit about this ordering.
Because the book wants to be honest about what it does not know. Most technical AI books either dismiss the possibility of qualitatively different structures of intelligence or leave the possibility off the page. The book takes the third option: it names the possibility, flags it as a hypothesis, and lets it stand as one of the things the framework is designed to be compatible with rather than dependent on.
HRIH appears in the December 8 2024 manuscript. HRIH line 8524 discusses "Recursive Thinking in Cognition"; HRIH line 14005 states the AI control-failure thesis. The hypothesis's early presence in the timestamped manuscript is documented; the register does not use it as a convergence with any external event, because there is no external event that would correspond to it as of mid-2026. It is a background hypothesis the book carries because it was in the earlier work.
A named hypothesis, an explicit disclaimer that it is a hypothesis, an explicit statement that the technical proposals do not depend on it, and a role in the book's argument as a compatibility target rather than a foundation. The book is careful. The reader who finds the hypothesis unpersuasive can skip it and lose nothing from the technical case.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.