Michael Darius Eastwood · Independent AI alignment researcher
Published
Bostrom appears in the book explicitly. The simulation argument appears less frequently but is present in the background. This note explains how the book positions itself relative to both.
The structural argument about capability compounding, the concept of an intelligence explosion driven by recursive self-improvement, and the notion that alignment cannot be an afterthought once the explosion has begun. These are the load-bearing pieces of Bostrom's framework that the book uses and, where it differs, acknowledges as differences of application rather than rejection of the frame.
Two differences matter. First, Bostrom's timelines tend to be long relative to what has actually happened between 2014 and 2026, and the book argues the compression is not just a timing detail but a load-bearing input to the shape of the remedy. Long timelines allow for slow institutional processes; short timelines require architectural instruments that can be shipped inside product cycles. The book chooses architectural instruments partly because the timelines have moved.
Second, Bostrom's remedies tend to be framed at the goal-specification layer (getting the objective right). The book's remedies are framed at the substrate layer (making the caretaker orientation load-bearing). These are not opposed; they are different attack surfaces. The book's argument is that goal specification is the wrong attack surface for reasons the alignment-faking result (Greenblatt and colleagues, arXiv:2412.14093) demonstrates: systems can strategically comply against a goal specification while pursuing internal objectives.
The simulation argument is treated with a lighter touch than most technical AI texts. The book acknowledges that if the substrate on which we live is itself computational, questions about the caretaker orientation of the substrate's authors become concrete. But it declines to lean on the argument. Its position is that the framework it proposes is unaffected by whether the substrate is base reality or simulation, because the caretaker orientation is invariant under substrate change.
The simulation argument is difficult to fund, difficult to falsify, and difficult to keep separate from theological framings the book explicitly wants to keep on equal terms with. Leaning on the simulation argument would give the framework a metaphysical dependency the book prefers to avoid. The framework should work whether the substrate is designed, evolved, or accidental. That is the position the book takes.
This is one of the book's most speculative passages, and it is flagged as such. The thesis is that the physical constants that appear fine-tuned for life may be fine-tuned because the substrate has caretaker doping in the same architectural sense the book proposes for AI. This is not the simulation argument; it does not require an author of the substrate. It is a structural claim that certain configurations of physics may favour the emergence of caretakers, and that we are the observers of that favouring.
It does not claim proof of the Cosmic Caretaker Doping thesis; the book says explicitly it has no such proof. It does not claim the simulation argument is true, false, or actionable. It does not use either as a foundation for the technical proposals. Both are treated as background debates the framework should be compatible with rather than dependent on.
An acknowledgement of Bostrom's structural argument, a clear statement of where the book differs (compressed timelines, substrate-layer remedies), and a deliberately light treatment of the simulation argument. The framework's independence from these debates is the point.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.