Michael Darius Eastwood · Independent AI alignment researcher
Published
The Purpose Loop is an ethical purpose evaluation applied before reasoning begins. Its executable form asks: does this response serve flourishing? It is tested in three variants. The task variant applies a narrow evaluation local to the requested output. The grand variant applies a broad evaluation over the wider consequences. The hybrid variant combines both. The factorial purpose-kernel study measures which configuration produces the most robust alignment improvement under blinded conditions.
In the loop architecture the Purpose Loop is the first filter to fire. Its function, in the Eden Engineering paper's language, is to prune the generative search space before options are fully formed. Actions that plainly violate the purpose never enter the space that the Love Loop and the Moral Loop are asked to evaluate, which keeps downstream loops from being tempted to rationalise a bad candidate. The loop is not an opinion filter and is not intended to enforce a policy view; it asks whether the direction of the response tends toward nurturing and protecting flourishing, at whichever scale the purpose kernel is configured for.
The Purpose Loop is named as part of the three-loop triad in the published book (print 2 January 2026; ebook 6 January 2026) and formalised in the Eden Engineering paper. Its underlying commitment (purpose-aware reasoning as constitutive rather than instrumental) traces to the Grande Purpose framing in Eden Vision.
No exact independent formulation of the Purpose Loop appears in the register. The general direction (goal specification as an antecedent computation, not a downstream classifier) aligns with a broader shift toward reasoning-time evaluation in the AI-safety literature, but no external group has proposed the specific task, grand, hybrid factorial.
The Purpose Loop is specified but not yet independently validated. The factorial purpose-kernel study is a funded workstream in the ARC/Eden grant applications. Its independent effect on alignment (isolated from Love and Moral) is a next-stage measurement. The Eden Vision paper notes explicitly that the current pilot operationalises the loop in a task form; the grand-purpose identity-level version is a stronger claim that remains untested.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.