The sixteen terms the book runs on.

Core concepts

AI Detox

A deliberate, time-boxed interruption of AI-assisted cognitive work designed to reload the brain circuits that AI consumes — not abstinence, not recovery, not quitting.

The Anti-AI Brain names its 30-day program an AI detox because the target is cognitive tissue reloading, not abstinence. A detox in this book is neuroscience-first: you interrupt the delegations that shrink the circuit (drafting, recall, decision, attention) and reload the circuit with handwriting, deep reading, silence, strategic play, and Cognitive Ops. The book deliberately avoids the clinical "addiction" frame used by the Artificial Intelligence Addiction Scale (AIAS, 2025) and similar DSM-5-adjacent instruments: the reader is an operator recovering tissue, not a patient recovering from dependency. The result at Day 30 is not "AI-free" — it is AI-dosed. Cognitive sovereignty is the end-state; detox is the reloading phase.

Pharmakon

A substance that is medicine or poison depending on the dose. The classical Greek frame the book uses for AI.

From the Greek φάρμακον. Plato used it in the Phaedrus to describe writing itself — a technology that could either strengthen memory or replace it, depending on how it was practiced. The Anti-AI Brain argues that large language models are the pharmakon of this generation: the same tool that makes a disciplined operator sharper makes an undisciplined one thinner. The book is a dosage guide, not a prohibition.

Cognitive Debt

The accumulated loss of deep-thinking neural activation caused by chronically delegating cognitive work to AI.

Coined in Kosmyna et al., 2025 (MIT Media Lab). Across a sequence of essay-writing sessions with EEG monitoring, participants who used ChatGPT showed roughly 55% lower alpha-theta coupling in prefrontal deep-thinking regions relative to the unassisted group, and the reduction persisted even when the AI was taken away in later sessions. "Debt" because the signal compounds: the circuit that is not loaded one day is harder to load the next. The book uses the term throughout Part I to name the thing the protocol is designed to reverse.

Cognitive Partisan

A person who uses AI harder than most without losing the cognitive tissue AI consumes in them.

The book’s protagonist type and the name of the third reader tier in the Anti-AI Brain Score. A Cognitive Partisan is not an AI skeptic, an AI abstainer, or an AI maximalist — they are an AI operator who has deliberately preserved the circuits a passive user gives up. The stance is adversarial in the old sense: they treat the machine as a sparring partner whose outputs must be beaten into shape by a brain that can still hold, compare, and reject.

The Ten-Minute Wall

The point roughly ten minutes into an unassisted cognitive task when the urge to prompt an AI becomes physical.

A phenomenon reported by almost every reader who attempts Day 1 of the protocol. Ten minutes into a problem with no AI in reach, something that feels like withdrawal arrives: restlessness, a spike in device-checking, a conviction that the task is too hard. The book uses the Wall as a diagnostic — the day it stops arriving at ten minutes and starts arriving at thirty is the first measurable sign the attention circuit is reloading. Named to make the experience common knowledge for readers who otherwise assume the feeling is a failure of discipline.

Outsourced Mind

The terminal case of AI delegation — a reader whose self-reported cognition has merged with the model’s output to the point where the two are indistinguishable to them. Subject of Chapter 11.

A diagnostic category the book introduces in Chapter 11. Distinct from the Ghost tier of the Anti-AI Brain Score because the Outsourced Mind is often functional, even high-performing — the functionality simply runs through the model. Asked what they think about a topic, they open a chat window. Asked to recall a decision, they search the conversation history. The book argues that Outsourced Mind is not a ceiling: Recall-and-Teach (one of the Anti-AI Seven practices) is the discipline that discriminates what a reader actually owns from what they only accessed. Outsourced Mind is a book concept, not a quiz-returnable tier — most Outsourced-Mind readers land in the Ghost tier on the Score.

The Five-Layer Model

Flesh

Layer 1. The body the brain lives in — sleep, oxygen, load, light. The substrate every other layer rests on.

First of the five layers the protocol treats. The Flesh layer names the non-negotiable physiological inputs: eight hours of sleep in the dark, forty-five minutes of physical load, sunlight on the retina before noon, food that does not require a blood-sugar correction. Cognitive interventions stacked on a broken Flesh layer do not return. Every chapter of Part III that introduces a new circuit opens with a Flesh-layer check because none of the downstream practices transfer without it.

Archive

Layer 2. The memory store the brain builds through handwriting, spaced recall, and deep reading.

The layer where information becomes yours — not a pointer to a cloud document, not a bookmark, but an encoded trace. Sparrow et al. (2011) showed that people who expect to look something up later fail to encode it at all. The Archive layer is rebuilt through handwriting (van der Meer’s EEG work on the motor-sensory coupling of letter formation), deep reading on paper, the method of loci, and deliberate spaced retrieval. The protocol treats Archive as the precondition for Reasoning: you cannot reason about what you did not store.

Lens

Layer 3. The attention filter — what you let in, what you reject, what you hold.

The Lens layer governs what reaches your Archive in the first place. Ward et al. (2017) measured a roughly 10% drop in working memory capacity from the mere presence of a smartphone on the desk; Madore et al. (2020) linked media multitasking to a 20% increase in memory-lapse failures. The Lens layer protocol is about phone location, notification architecture, tab discipline, and the conscious decision to read long-form on paper. When the Lens is fouled, nothing downstream works.

Loop

Layer 4. The reasoning cycle — hold, compare, steelman, reject, iterate.

The Loop is what most people mean when they say "thinking": the iterated cycle of holding a hypothesis, testing it against evidence, steelmanning the opposing view, rejecting or updating, and looping again. Lee & Sarkar (2025) found that heavy AI use correlates with a shortening of this loop — users report "less cognitive effort" and also less confidence in the result. The protocol reinstalls the Loop through Socratic prompting (self-administered), structured disagreement, and the Ten-Minute Wall.

Voice

Layer 5. The output layer — the sentence, the decision, the shipped artifact that is unambiguously yours.

The top layer of the model. Voice is what emerges when a functional Flesh, Archive, Lens, and Loop produce an output the world can read — a paragraph, a judgment, a design, a line of code. The book’s argument is that AI degrades Voice from the top down: you notice the loss at the sentence level first, then at the judgment level, then at the idea level, until what is "yours" is only the prompt. The protocol rebuilds Voice by insisting that the draft is written before the model touches it, not after.

The protocol

The Anti-AI Seven

The seven daily disciplines that structure the 30-day protocol: Deep Reading, Handwriting, Movement, Strategic Play, Silence, Cognitive Ops, Recall-and-Teach.

The seven practices the reader installs, one per circuit, across the 30-day protocol. Deep Reading: one paper book, sixty minutes, no device in reach. Handwriting: twenty minutes a day, longhand. Movement: forty-five minutes of physical load. Strategic Play: Go, chess, or equivalent, against a human or the clock. Silence: thirty minutes with no phone, no input, no task. Cognitive Ops: Claude or GPT as sparring partner — Socratic, steelman, falsification modes only. Recall-and-Teach: explain a concept without looking it up. The set is not aspirational — each of the seven is calibrated to a specific circuit from the neuroscience chapters.

The four reader tiers

Ghost

Tier 1 (score 0–7) on the Anti-AI Brain Score. A reader whose attention, memory, and judgment circuits have been handing work to AI long enough that the transfer is well underway.

The lowest tier on the diagnostic quiz. The MIT Media Lab EEG study found that 83% of heavy AI-assisted writers could not recall a sentence from the essay the machine wrote for them — most Ghost-tier readers are in that cohort. The fog is real, the circuits are thin, but the brain rebuilds fast once the protocol reloads it. The first week will feel wrong — that is the signal. The protocol was written with Ghost-tier readers as the primary audience; Part II of the book is the neuroscientific argument that the condition is reversible, not terminal.

Partisan

Tier 2 (score 8–14). A reader who sees the pull of AI delegation and loses to it about half the time. The largest cohort on the chart, also the one with the sharpest view.

The second-lowest tier, and the most consequential. A Partisan still has most of the equipment — they probably caught themselves this month reaching for a prompt window before finishing a thought, then finished the thought once or twice, felt a small spike of sharpness, and drifted back to the easier path. Microsoft Research’s 2025 CHI paper on 319 knowledge workers used the word "atrophied" to describe higher-AI-trust cohorts. Partisan tier is the stage where the transfer is cheap to reverse and expensive to ignore. The book’s Chapter 3 — the Cognitive Partisan workflow — is calibrated here.

Operator

Tier 3 (score 15–21). A reader who runs the machine rather than being run by it. Sharper than most; has not surrendered the circuits AI degrades fastest.

The second-highest tier. An Operator feels the pull of AI delegation and chooses against it more often than their peers. They draft before they paste, hold hard decisions for an hour before asking, and still read on paper. The weakness Operators share is that awareness is not structure: the practices they do intuitively lose consistency under deadline pressure. A protocol gives them a schedule; a schedule survives the Tuesday afternoon when they are tired. The 30-day program in the book is designed for Operator-tier readers who want to lock in the habits they already have.

Sovereign

Tier 4 (score 22–30). Rare category. A reader who spars with the machine and does not hand it the pen — the habits are already installed.

The top tier. A Sovereign reads on paper, writes by hand, holds hard decisions, and verifies AI outputs against primary sources. They use the machine as a sparring partner, not a ghostwriter. They score where they score because the habits are installed. The question for Sovereign-tier readers is not recovery — it is maintenance and transmission. The book’s Part IV covers the year-long arc of staying Sovereign and teaching the protocol to the 80% of people around them who are still at Partisan tier or in transfer.