Michael Darius Eastwood · Independent AI alignment researcher
Published
The claim is neither that the ideas in Infinite Architects are all uniquely original, nor that no similar-sounding phrase exists anywhere in the literature. It is more specific. The book coins and defines 37 named concepts, each with a distinct role in the argument, and to the best of a prior-art search across the standard databases none of them appears with both the same label and the same technical definition before the book. The concepts include the Three Ethical Loops (Purpose, Stakeholders, Universalisability), the Chokepoint (the compute-supply pinch that turns AI policy into industrial policy), Caretaker Doping (an unbuilt hardware token that binds value alignment at the substrate), the HARI Treaty (a proposed governance framework), and roughly thirty more. The book is where the ideas are laid out; the claim is that most of them have no equivalent in the prior literature.
The primary evidence is the published book itself, whose ISBN and publication dates fix the priority moment. The secondary evidence, and the more interesting kind, is what has happened since. The Caretaker Doping direction, defined in the book as a hardware-level moral genome token, was independently converged upon by hardware-security work: two Odeh patents filed in the United States (US 2026/0010411 and US 2026/0010780) describe runtime ethics gates and sub-five-microsecond ethical filtering at the chip level, and Petrie's arXiv preprint (2509.07637, September 2025) describes thousands of embedded off-switches per accelerator with less than one per cent die overhead. None of this work cites the book, and the book does not claim to have caused it; the point is that the direction is converged upon rather than isolated.
Coining a term is not the same as proving it works. Several of the 37 concepts are speculative by design, most notably Caretaker Doping, which the book explicitly rates at technology readiness level 0 to 1 (no hardware implementation exists). The claim being made here is a lexical and definitional one, not an engineering claim. A prior-art audit conducted by someone other than the author is what turns the claim from an author's assertion into a verifiable fact. Until that audit is on the record, the number 37 is defended by the search the author has already done and by the fact that the book, which contains the definitions verbatim, is publicly indexed.
The falsification contract asks an independent researcher, not affiliated with the programme, to search at least four databases (arXiv, Google Scholar, Semantic Scholar, PhilPapers) for each of the 37 concepts by name and by semantic paraphrase. If thirty or more of the 37 have no prior published equivalent with matching label and definition, the claim is confirmed. If twenty to twenty-nine come through clean, the claim moves to partial and the specific concepts that flunk are identified by name on the dashboard. If fewer than twenty survive the audit, the claim is refuted and the coinage count is publicly reduced to whatever the audit supports.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.