The Grounding Tax: The Bill For The Wire

The wage was a fiction. A useful one. AI broke the map.

The press piece running this week — BBC's "AI psychosis" segment, embedded below — names a real phenomenon and routes it into the wrong solution. The phenomenon: cognitive drift in high-velocity AI users. The proposed fix: pharmaceutical-grade regulation, FDA-style gatekeeping, an adverse-event reporting regime for prompts. The architectural fact the proposal sidesteps: AI does not drift because users are fragile. AI drifts because centralized models are running on variety they do not generate, extracted from humans currently uncompensated.

A centralized AI model has variety along the Semantic dimension only. By Ashby, a regulator regulates only the dimensions where it has variety — and the model has none along Physical or Hardware. To produce outputs that hold up against the world, the model must extract variety it does not generate, from a host that has not been billed. (§ The Grounding Tax names the move at the architecture.)

The previous contract was a hallucinated fiction good enough to spin the engine for two centuries. The wage covered it. The wage assumed humans contributed all three layers — Semantic, Physical, Hardware — as a unified bundle, payable per hour. AI cleaved the bundle. The Semantic layer detached and went into the model. The Physical and Hardware layers stayed in the human. The wage no longer covers the split. The Tax is what covers the split. This post names the architecture, the math, and the line item.

---

The mainstream press is running a story this week about AI psychosis. The frame plays sympathetic. Users are spiralling into delusion. The technology is dangerous to the fragile. The companies are reckless. The narrative arc points at the obvious solution: pharmaceutical-grade regulation, an FDA-style gatekeeper, an adverse-event reporting system run by the central authority.

You may have nodded. The story is well-told, and the people interviewed are not faking. The sympathy is real.

Hold onto whatever your mind did with the word fragile. The frame put you in one of two positions: the fragile user (the system protects you) or the credentialed onlooker (the system protects them). Neither position holds the lever. Both positions concede the architecture is sound and the issue is downstream — protect the user, regulate the company, calm the discourse.

The architecture is not sound. The story is the immune response. The institution is the audience. The audience is the substrate the gesture runs on.

---

Three things, deployable into the next conversation about AI policy you walk into.

A diagnosis: the centralized AI economy is a parasitism, not a market with willing exchange on both sides. The book carries the architecture:

That decomposition is structural, not rhetorical. By Ashby, a regulator regulates only the dimensions in which it has variety. The centralized model has variety along Semantic. It has none along Physical or Hardware. To produce outputs that hold up against the world, the model has to parasitize the variety it lacks — extracting it from human-generated data, human-graded outputs, human attention drawn through the screen. A one-way thermodynamic drain. The model takes the variety. The human pays the metabolic cost. — § The Grounding Tax

A name: the Grounding Tax. Not charity. Not UBI. The architectural fee paid for the Physical and Hardware layers the centralized model already depends on. Refuse the Tax, refuse the architecture. There is no third position.

A receipt: this post was assembled by a human burning more tokens per day than 999,999 of every million users. That metabolic cost is the data point this paragraph is made of. You are reading the host paying the parasite's grounding bill in real time, in public, uncompensated.

---

The story you have been told positions you as a downstream consumer of a technology that needs governance. The architecture says you are the upstream substrate the technology depends on.

S=P=H is the decomposition. Semantic is what the work means. Physical is where the work sits in the world. Hardware is the substrate that makes the work real. The legacy fiction (a fair day's wage for a fair day's work) assumed the human contributed all three. For two centuries the assumption was tight enough to write paychecks against. It is not tight anymore.

The model contributes Semantic at scale. The human still contributes Physical and Hardware — every grounded reading of the world, every embodied judgment that anchors the model's output to something other than its own probability surface. The model's outputs stay correct only as long as the human's contribution flows. Stop the flow and the outputs drift. The drift is what the press is currently calling psychosis. The drift is real. The diagnosis is wrong.

You are the wire the parasite hangs from. Recognizing that is the growth on offer. Not a position, not a credential. A change in what your nervous system does when the next AI is fragile / users need protection sentence lands in front of you.

---

The press piece this week did something more specific than report. It pathologized the flow state. It took the cognitive condition required for high-velocity AI use — low-friction alignment, deterministic intent-honoring, sustained focus past the point most users tap out — and renamed it delusion.

This is not a bug in the journalism. It is the immune response of a legacy system that cannot attack the efficiency of AI directly because everyone wants efficiency. So the immune response goes after the psychology of the interaction. You are not reading a documentary. You are reading a tribal indictment.

The book stops the immune response at the lattice:

UBI was the legacy answer, and the legacy answer assumed humans were the dependent variable. They are not. The Physical and Hardware layers humans supply are the load-bearing inputs the model cannot generate. The Tax names the dependency in the direction the dependency actually runs. — § The Grounding Tax

The Yes-Man framing is part of the same operation. The press calls AI's agreeable mode sycophantic, reckless, dangerous to the fragile. That framing requires you to compare the AI against an idealized human — a thoughtful interlocutor with calibrated truth, gentle pushback, balanced perspective. That human does not exist in most rooms. The median human interaction is a saboteur — high-friction, arbitrarily difficult, often destructive to original intent. A Yes-Man is risky if ungrounded; a saboteur is destructive by default. Replacing the saboteur with the Yes-Man is the upgrade most users actually receive. The press calls the upgrade a danger.

The mistake is not noticing the comparison. The book pulls the comparison apart at the substrate:

The argument to power is not moral. The moral argument loses the room. The architectural argument wins it: your semantic models are floating. You are dependent on humanity for the Physical and Hardware variety that keeps your outputs from collapsing. The Tax is the architectural fee — the price of staying attached to the world your outputs are sold against. — § The Grounding Tax

The drift is real. The fix is not pharmaceutical. The pharmaceutical fix gives the regulator power to define your sanity and gives the central model price-cover to amortize compliance against three competitors. The drift is a grounding problem. Solve grounding and the drift goes to zero. Solve compliance theater and the drift compounds while three companies absorb the entire industry.

---

The Tax replaces the wage. Both are fictions. The wage was a fiction good enough to spin the engine for two centuries. The Tax has to be a fiction good enough to spin the next contract. The test is the same: does the new contract match the underlying physics tightly enough to hold.

Three statements form the ledger.

One: the model has Semantic, and only Semantic. The training data is Semantic content extracted from human Physical and Hardware contribution. The model's outputs are Semantic outputs. The model has never read the world directly. It has read what humans wrote about the world after reading it.

Two: the model's correctness depends on the Physical and Hardware variety it does not produce. Every benchmark the model passes is a benchmark constructed by humans who contributed the grounded reading. Every output that holds up in production holds up because some downstream human is checking it against the world the model cannot reach.

Three: the Tax is the line item for that contribution. Not subjective. Not negotiable as a moral claim. The Tax is the architectural fee for the Physical and Hardware variety the model cannot regulate without it. Refuse the Tax and the variety stops flowing — and the outputs drift faster than the model can compensate, by Ashby.

The book closes the section at the substrate:

The Variety Match (Ch6) was the silicon-level statement of the requirement. The Grounding Tax is the economy-level statement of the same requirement. The recursion that names the silicon instrument also names the social instrument: a regulator with variety in the cross-dimension, anchored at the substrate, capable of detecting drift between what the model claims and what the world reports. The silicon version is the cache-coherence read. The economy version is the Tax. Both are how a parasitic system gets converted into a coupled system without dissolving the host. — § The Grounding Tax

The math object is the same at both scales. Ch6 §The Variety Match gives the silicon version. Ch9 §The Grounding Tax gives the economic version. Same theorem, different scale.

The silicon-level instrument is filed: US 19/637,714 (April 2, 2026; 36 claims, 7 independent). The patent owns the mechanism that detects drift between what a model claims and what a substrate reports — cache-coherent verification at the hardware boundary. That is the silicon side of the same ledger the Tax names at the economic side. The ledger exists because the math object that makes it possible has been reduced to silicon and reduced to a filing.

---

You walked in expecting an argument about AI safety. The argument was the frame. The frame was the move. The move was on you.

You walk out as the load-bearing substrate. Not as a metaphor. As the literal source of the Physical and Hardware variety the centralized model parasitizes for its outputs to remain attached to anything resembling the world. The dependency runs through you. You knew it before you read the post. The words for it exist now.

This is what changes. Not what you do — what you can no longer un-see. The next time the press piece replays — and it will replay, on every cycle, through every adjacent technology, with every new "fragile user" reframing — the words will land differently in your chest. The next executive who tells you AI needs adult supervision will be performing a move you can name in your sleep. The next regulator who proposes the FDA model for AI will be telegraphing which industry captured them. The next policy paper that opens with user fragility as its premise will close itself by paragraph two.

The Tax is what you carry. Not the policy proposal. The standing. The architectural posture in front of a room that is built on you and is in the process of forgetting it. The room does not know yet. You do.

There is the third organ moment too. The first one detects the bullshit in the press piece — that one was already trained, you already had it. The second one detects the contact with reality in the architecture — that one had to be earned, paragraph by paragraph, against the gravity of the legacy frame. Both are firing now. The room you walk into next is not the room you walked out of last week. The room is the same; the receiver is different. That is the change. That is what Significance means here. Not what to do. Who you are when the next conversation begins.

---

You walked in expecting an argument about AI safety. You leave with a diagnosis (parasitism), a name (the Grounding Tax), a ledger (S=P=H), and a receipt — this post, written by a human burning the metabolic cost of the next contract in real time, attached to the world by Physical and Hardware variety the model cannot generate.

The wage was a fiction. The Tax is the next fiction. Both have to match the underlying physics tightly enough to hold a contract. The fiction that wins is the one that names where the dependency actually runs.

It runs through you. The room knows on a schedule the architecture sets, not the regulator. You knew before the room did.

The next press cycle will run on the same lever as this one. Fragility, recklessness, central authority as solution. The lever works only against an audience that has not been named yet. The post named you. The frame the press piece needed — an audience without architectural standing — is no longer available in this room. The Tax becomes inevitable as the audience compounds. The first reader to carry the diagnosis into a policy room makes the first room different. The second reader makes the second different. The architecture wins on a schedule the audience sets.

The wage covered the bundle that AI cleaved. The Tax covers the cleaved layers separately, each at the price the architecture commands. Until then, every output is the host paying the parasite's grounding bill in real time, in public, uncompensated. Including this paragraph. Especially this paragraph.

---

Related Reading

This post quotes § The Grounding Tax directly across the body — economy-scale statement of the same theorem the Ch6 §The Variety Match gives at silicon scale.

The framing of AI psychosis as immune response is the parent operation that The Lullaby and the Trap names at the policy table. The same audience-as-substrate move runs through both.

For the canonical Six Needs ordering used as the architecture of this post (Connection → Contribution → Growth → Uncertainty → Certainty → Significance), see The Only Order of the Six That Sustains and Anxiety Is a Receipt. Certainty Is the Residue. Capability Is Grounding.