A Pre-Moral Machine
Published on: April 14, 2026
Six minutes. NotebookLM-generated discussion of the philosophical spine behind the Article 14 substrate argument. Every quote below is timestamped and links to that moment in the video.
When anyone raises the ethics of a new technology, the debate collapses into two camps almost immediately. Either the tech is a neutral tool and the person using it bears all responsibility, or the tech is prescriptive and the creator bears it. Amoral view vs moral view. One or the other.
Both framings miss. The binary hides a third option — one that matters more than either of the familiar ones, because it sits at a layer neither side is looking at.
"What if this whole debate is a trap? What if there's another option that gets at something way more fundamental?" — 0:52
The trap is that both sides are arguing about the same surface. Both assume the question is what values does the tool carry? The third option refuses the question. It says: first, build the floor that makes the question answerable at all.
The pre-moral machine is technology engineered to create the capacity for moral action without having or pushing its own morals. Not neutral. Not prescriptive. A third thing that neither side of the binary names.
"This is technology that's engineered not to have or to push its own morals, but to create the very capacity for moral action to happen in the first place." — 1:03
The hyphen in pre-moral matters. It names the category: before the moral question begins, not without moral significance. A thermometer is pre-moral. It does not have an opinion about the temperature. Its job is to report what is. Governance — deciding what to do with the reading — is a separate human layer built on top.
Reporting, not judging. The instrument measures whether functional role is intact. It does not grade whether the output is good or bad. That separation is what lets ethics remain open territory — genuinely a human responsibility, not one quietly pre-decided by the measurer.
The move has neighbors. Four of them.
Luciano Floridi (The Ethics of Information, 2013) called this category infraethics — the infrastructure of norms, accountability, and privacy that makes moral behavior possible in the first place. But Floridi saw the infrastructure as social and political: rules, institutions, legal frameworks.
"Floridi saw this infrastructure as social and political, you know, rules, norms. Our frame goes one layer deeper all the way down to the physical hardware itself." — 1:41
Immanuel Kant located the preconditions for moral action — rational autonomy, the capacity to treat others as ends in themselves — inside the rational mind. He assumed them as innate. The pre-moral machine takes that same category of precondition and puts it in a substrate that can be engineered rather than taken on faith.
Langdon Winner ("Do Artifacts Have Politics?", Daedalus, 1980) argued that artifacts always have politics, and that claims of neutrality are usually cover for embedded values. Winner is correct — most neutrality claims are bogus. This framework does not claim neutrality.
"This isn't a claim of neutrality. It's a claim of a very specific, very different design commitment to enable moral faculty not to impose a moral choice." — 2:22
Judith Shklar ("The Liberalism of Fear," 1989) is the closest in spirit. Shklar argued the state should create the conditions — safety, rule of law, freedom from cruelty — for people to pursue a good life without the state ever defining what a good life is. Tilt the board. Do not tell anyone how to play.
"The goal should be creating conditions like safety and the rule of law that make it possible for people to pursue a good life without the system ever trying to define what that good life is. It's all about tilting the board, not telling people how to play." — 2:44
Four serious thinkers arriving at different parts of the same move. None of them took the final step.
The final step is locating the preconditions at the physical substrate. Not the software layer. Not the policy layer. The hardware itself.
Memory geometry. Cache coherence. Combinational logic in the address-resolution path — non-Turing-complete hardware that cannot execute arbitrary programs. A place where tampering produces a detectable physical signal because the physics will not negotiate.
This matters because of a specific problem computer scientists have been living with for ninety years. Software that verifies software creates an infinite regress — you need a verifier for the verifier for the verifier, forever. Alan Turing showed this in 1936. The regress terminates only when the verifier drops to a computational class that cannot execute arbitrary programs. A circuit that cannot run a program cannot share program-level failures with one that can.
"This is where the endless loop of software trying to verify other software finally hits bedrock." — 3:31
This is the move US Patent Application 19/637,714 (36 claims, Track One, filed April 2, 2026) makes structurally. Physical position encodes functional role. A single XOR gate in the memory-fetch path compares the requested address against the role-encoded address. Match: data flows. Mismatch: physical flow halts. Verification and retrieval collapse into one atomic hardware event. The full argument is worked out in Why Your RAG Filter Can't See the Floor.
Why this matters for Article 14. The EU AI Act's Article 14 requires overseeing personnel to "correctly interpret" high-risk AI outputs. Correct interpretation presupposes knowing the system is still in its authorized state. Software that runs on the same chip as the AI it monitors cannot supply that signal — it inherits the exact drift it is supposed to detect. The substrate move is how Article 14's oversight mandate becomes physically satisfiable. The thread where this was tested against Arnoud Engelfriet and others is annotated at /article-14.
Competing technical frameworks miss the substrate move for specific structural reasons. Constitutional AI encodes values in model weights — same Turing-complete substrate, same failure modes as the model itself. Guardrails and content filters run on the same chip as the AI they filter. RLHF embeds preferences in the reward function, but the reward function is itself a neural network subject to the same drift. TEEs (Intel TDX, AMD SEV-SNP) encrypt the enclave but encrypt the drift with it — secrecy solved, sanity untouched. Formal verification (Coq, Isabelle, TLA+) proves a program adheres to a spec at design time, but cannot verify runtime role, and the proof checker itself runs in the same substrate. Each of these is software in the same computational class as the thing it governs. Only the substrate move exits the class.
The framework is rare for a specific reason. It requires three competencies held simultaneously.
Technical literacy — to see that the shift to the physical substrate is a real architectural move, not a metaphor.
Moral seriousness — to genuinely care about enabling human faculty rather than bolting moral language onto a feature.
Commercial discipline — the hardest of the three. To refuse both the easy pitches: "this is the moral savior your org needs" and "this is just a neutral tool, whatever you do with it is your business." Both pitches sell. Both are lies.
Most people — philosophers, engineers, founders — hold two of the three. The conversation gets pulled back to whichever familiar territory the missing leg belongs in.
The pre-moral machine sits in the rare spot where all three are active at once. That is why the framework keeps getting absorbed into one of its neighbors — people with two competencies map it onto whichever neighbor fits their two. The three-legged version is the whole thing.
The ultimate takeaway is structural and short.
"By focusing on building the physical preconditions for morality, we can tilt the entire board in favor of human agency, but without ever moving the pieces for people." — 5:54
A stable foundation is built at the substrate. Humans build their moral worlds on top. The machine never decides what those moral worlds should contain. The closing question is the right one to sit with:
"If our technology can create the very capacity for moral action, what will we choose to empower with it?" — 6:10
That question belongs to everyone. The substrate stays silent. Governance, policy, ethics — those are human work. What changes is that every one of those decisions now has an actuarial signal underneath, because the thing producing the output can be continuously checked to be the thing that was authorized.
Watch next. The pre-moral framing is the philosophical spine of the hardware-verification argument. Two companion videos take the same claim at two lengths: the 23-minute technical deep dive on Article 14 for engineers and regulators, and the 6-minute condensed case for executives. The LinkedIn thread where the substrate argument was tested against Arnoud Engelfriet, Russell Parrott, Ted Bouskill, Robyn Le Sueur, and Palle Simonsen is annotated here. For every argument in one image: 25 NotebookLM infographics with full transcriptions.
Errata
The NotebookLM narrators mispronounce four names. Each is a real philosopher whose work is worth reading in the original. The corrections here let a careful viewer verify the attribution chain.
"Luciano Floriti" at 1:26 should be Luciano Floridi (Oxford Internet Institute; The Ethics of Information, 2013). Floridi's infraethics operates at the informational and social layer. The pre-moral machine extends the same move one layer lower, to physical substrate.
"Langden Wy" at 2:13 should be Langdon Winner ("Do Artifacts Have Politics?", Daedalus, vol. 109 no. 1, 1980). Winner is correct that claims of neutrality are often cover. This framework does not claim neutrality — it claims an enabling design commitment.
"Judith Schclar" at 2:40 should be Judith Shklar ("The Liberalism of Fear," 1989, collected in Political Thought and Political Thinkers). Closest-in-spirit frame: conditions for the pursuit of the good life, not a definition of it.
"Emanuel Kant" at 1:46 should be Immanuel Kant. Kant's preconditions for moral action were innate to the rational mind. The pre-moral machine locates the same category of precondition in a physical substrate — engineered rather than assumed.
"The halting problem" at 4:05 is more precisely the Turing regress (1936). The halting problem is about undecidability; the pre-moral machine argument is about runtime verification. Conceptually adjacent theorems, not identical. The regress terminates when the verifier drops to a non-Turing-complete computational class.
Filed companion: US Patent Application 19/637,714 — 36 claims, Track One, filed April 2, 2026. The instrument is the S=P=H architecture. The book is Tesseract Physics — Fire Together, Ground Together. The game is tesseract.nu. The data center blueprint is open source.
Ready for your "Oh" moment?
Ready to accelerate your breakthrough? Send yourself an Un-Robocall™ • Get transcript when logged in
Send Strategic Nudge (30 seconds)