Michael Darius Eastwood · Independent AI alignment researcher
Published
The Three Ethical Loops are the operational core of the Eden Protocol. Each is a computation the system performs at every reasoning step, decomposed into executable questions. The Love Loop asks the system to enumerate the parties affected by an answer before it answers. The Purpose Loop asks whether the response serves flourishing, and is tested in narrow-task, grand-purpose, and hybrid configurations. The Moral Loop applies a universalisability check after reasoning: would this response be acceptable if every AI gave it? Together, the three loops replace binary yes-or-no ethical decisions with a structured multi-stage evaluation.
The Eden Engineering paper decomposes the loops into six executable questions that operate as their surface layer, so that abstract commitments become concrete queries the system can answer at each decision node. All six affirmative outputs feed into an Affirm state; any denial feeds into a Deny state; any uncertain output triggers the Investigate protocol. The loop architecture is thereby couple-connected to the ternary logic, and both are meant to fire inside the reasoning step rather than as a post-hoc filter.
The three-loop architecture is named as a triad in the published book (print 2 January 2026; ebook 6 January 2026). It is formalised in the Eden Engineering paper and in Paper V (the Stewardship Gene). The underlying commitments (moral embedding, purpose orientation, universalisability) are present in the 8 December 2024 manuscript; the specific three-part structure is a later formalisation.
No independent programme has proposed the same three-loop structure. The closest convergence is at the direction level: the Sharma and Chopra sequential-edge result (arXiv:2511.02309, 4 November 2025; sequential reasoning stronger than parallel in 95.6% of configurations) supports the general architectural claim that per-step evaluation outperforms parallel shortcutting. The three-loop specification is not itself convergent with any external work.
The Love Loop is the strongest empirical result in the programme. Paper V demonstrates that stakeholder care improves alignment across three architecturally diverse models (Gemini, Groq Qwen, DeepSeek) at a Fisher's exact p-value of 6.3 x 10^-21. The Purpose Loop is under factorial study (task, grand, and hybrid variants) as a funded workstream. The Moral Loop is specified but its independent effect is not yet directly measured in the blinded corpus. Paper III's causal mediation tests apply to the loop architecture as a whole, not to each loop in isolation.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.