Michael Darius Eastwood · Independent AI alignment researcher
Published
The ARC Principle is the book's central technical framework. This note describes how the book presents it, references the equation family framing, and points readers to the papers for the derivation. New to Michael Darius Eastwood? Start at /start-here.html for a one-page orientation, then come back.
Amplification, Recursion, Compounding. The book uses the acronym to name the pattern by which intelligence grows: each iteration amplifies what came before, recursion produces successive iterations, and compounding makes the growth pattern multiplicative rather than additive. The pattern is not unique to AI. The book argues it is the shape of intelligence growth across biological, computational, and quantum systems, which is the argument Paper III maps in detail.
U is a measure of accumulated capability at a given moment. I is the input, roughly the resources or effort available. R is the recursion factor: how many iterations of self-reference the system runs on each input. The exponent, presented in the book as 2, captures the compounding: each iteration builds on the previous, and the growth is quadratic rather than linear in R.
Because the case alpha near 2 is the case where the mathematics and the intuition agree in a form that survives translation into plain English. Papers I and II derive the general form, U equals I times R to the alpha, and note that alpha varies with the recursion structure. The alpha near 2 case is common enough to be a useful default and legible enough to be a book-facing formulation.
Not a settled empirical law on frozen systems, and not the operative safety criterion after Paper X. That paper supersedes the fixed-exponent framing (alpha as the operative safety criterion) with the co-scaling criterion, and any policy or research use of the equation for capability growth should read Paper X first. The equation itself, and the ARC Bound alpha at most 2, remain live hypotheses awaiting their real test on genuine self-improving systems. The operative quantitative safety claim from the ARC family is the beta greater than k co-scaling stability condition. That is the claim the current measurement programme is built to test.
Chapter 3 introduces the framework. Chapter 4 uses it to argue for architectural safety instruments (caretaker doping, meltdown triggers, the Three Ethical Loops). Chapter 5 extends it to the Cosmic Caretaker Doping thesis, which is flagged as speculative. Chapter 8 uses it to argue that recursion at compute scale requires supply-chain-layer governance. The framework is not decorative; it is the through-line the book's technical claims sit inside.
Paper I derives the principle. Paper II runs experimental validation. Paper III maps the pattern across quantum error correction, biological scaling, and machine reasoning. Papers IV-a through IV-d develop the blinding methodology, the alignment scaling problem, and the sign-flip result. Paper X derives the beta over k condition and supersedes the fixed-exponent framing as the operative safety criterion. Paper XI collects the convergence work. The book is the accessible-reader version of the same framework.
A named framework, an equation family with a clear rule about citation, a specific book-facing formulation, and a clear statement of what was empirically retracted (the earlier unblinded 2.24 fit, corrected to approximately 0.49 under blinding) as against the programme's own earlier framing. The equation and the ARC Bound themselves remain live hypotheses. The framework is central to the book; readers who want the derivation should read Papers I and II; readers who want to know which quantitative safety criterion is now operative should read Paper X.
Related notes on this site: Paper I in plain English for the derivation of U equals I times R to the alpha; Claim 2 explained for the compact evidence-and-caveats summary; the full arc for the chronology; the night of 8 December 2024 for the priority anchor. The book itself: Amazon UK or Amazon US. New readers looking for the map: /start-here.html.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.