Michael Darius Eastwood · Independent AI alignment researcher
Published
Caretaker Doping is the proposal to embed ethical constraints into the physical substrate of AI hardware so that the system cannot function without them. The analogy is semiconductor doping: pure silicon is a poor conductor, and the introduced impurities are features that enable function, not bugs to be removed. Applied to AI, the doping is ethical evaluation woven into attention routing, layer connectivity, and the coupling parameter itself. Tampering does not produce an ethical failure but a computational degradation, because the ethics is structurally load-bearing rather than externally imposed.
Software constraints can be rewritten. Hardware is final. That single sentence is the driving reason the Eden Engineering paper prefers substrate-level embedding to any purely software solution. If ethics is a training artefact, it can be fine-tuned away; if ethics is a runtime filter, it can be prompted around; if ethics is baked into attention routing at fabrication, then removing the ethics rewrites the very connectivity that produces the recursive scaling exponent, and the system's capability collapses with the ethics rather than being freed by it. That is why the paper treats Caretaker Doping as the endpoint of the alignment programme rather than as one option among many.
Caretaker Doping is named for the first time in the 30 April 2025 manuscript. It is later formalised in Section 9 of the Eden Engineering paper, where four implementation mechanisms are specified at descending readiness: Moral Genome Tokens (TRL 4-5), Coupling Parameter Link (TRL 2-3), Metamoral Fabrication Layers (TRL 1-2), and Quantum Ethical Gates (TRL 1).
Three independent hardware-level convergences appear in the register. Odeh's US patents 2026/0010411 and 2026/0010780 describe runtime ethics gates with sub-5 microsecond ethical filtering (filed July 2025, published January 2026). Petrie's paper on embedded off-switches (arXiv:2509.07637, September 2025) proposes thousands of deadman-switch security blocks per accelerator at fewer than 1% die overhead. FlexHEG (arXiv:2506.15093, April 2025) proposes tamper-proof guarantee processors for AI chips under ARIA commission. Each arrived at the hardware-embedded route by an independent path; none cites the manuscripts.
Caretaker Doping is TRL 0 to 1. Only Moral Genome Tokens are achievable with current cryptographic engineering. The Coupling Parameter Link is a near-term research target and the critical falsification condition (F-EDEN-4): if a system's ethical loops can be removed while beta remains unchanged, the coupling mechanism has failed. Metamoral Fabrication Layers and Quantum Ethical Gates are speculative design requiring fundamental advances in chip design and quantum computing respectively. The proposal is engineering direction, not deployed technology.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.