The evidence underneath the protocol.

Foundations

Why the brain responds to disuse.

The mechanism claims. Neuroplasticity is not a metaphor; it is a tissue-level process with measurable consequences. These are the studies that establish that cognitive function is built by training signals and lost when those signals stop arriving.

1. Ten weeks of targeted cognitive training reverses approximately ten years of brain aging on measurable indices of neural connectivity and cortical chemistry.

   McGill University (2025). Online brain-training program reverses 10 years of aging. ScienceDaily, October 29, 2025. Link

2. Cognitive-speed training reduces the 20-year incidence of dementia by approximately 25% in a longitudinal trial of nearly 3,000 older adults.

   Wang, L. A., Goldberg, T. E., Harvey, P. D., et al. (2026). ACTIVE Study 20-year dementia follow-up. Alzheimer’s & Dementia: Translational Research & Clinical Interventions. DOI: 10.1002/trc2.70197

3. Passivity is the unlearned default; what the brain actually learns is the detection of control. The ventromedial prefrontal cortex, given a history of consequential action, inhibits the dorsal raphe nucleus shutdown response — the revised fifty-year model of learned helplessness.

   Maier, S. F., & Seligman, M. E. P. (2016). Learned Helplessness at Fifty: Insights from Neuroscience. Psychological Review, 123(4), 349–367. DOI: 10.1037/rev0000033

4. One experience of controllable stress immunizes the animal against uncontrollable stress in different contexts the following day. vmPFC training generalizes — it is not task-specific. The strongest neurobiological argument for why a practice transfers.

   Amat, J., Baratta, M. V., Paul, E., Bland, S. T., Watkins, L. R., & Maier, S. F. (2005). Medial prefrontal cortex determines how stressor controllability affects behavior and dorsal raphe nucleus. Nature Neuroscience, 8(3), 365–371. DOI: 10.1038/nn1399

5. Rats that work for their reward grow denser dendritic branching in the prefrontal cortex and show higher BDNF in hippocampus and nucleus accumbens than rats given identical rewards for free. The nucleus accumbens is an effort-reward coupling organ; decouple effort from reward and resilience does not build.

   Lambert, K. G. (2006). Rising rates of depression in today’s society: Consideration of the roles of effort-based rewards and enhanced resilience in day-to-day functioning. Neuroscience & Biobehavioral Reviews, 30(4), 497–510. DOI: 10.1016/j.neubiorev.2005.09.002

6. Deep reading is built by use and atrophies from disuse. The neural circuit for deep reading is not innate — the brain assembles it, and an unrehearsed circuit thins.

   Wolf, M. (2018). Reader, Come Home: The Reading Brain in a Digital World. Harper.

7. Memory athletes trained in the method of loci show durable gains in recall and a specific neural signature of efficient encoding — evidence that technique can restructure how the brain stores information.

   Dresler, M., et al. (2021). Durable memories and efficient neural coding through mnemonic training using the method of loci. Science Advances. DOI: 10.1126/sciadv.abc7606

8. Wide-ranging carnivores with the largest natural ranges show the highest rates of stereotypy and infant mortality in captivity. A hunt circuit that cannot be completed produces measurable dysfunction — evidence that cognitive circuits, like motor programs, require their native inputs to stay well-regulated.

   Clubb, R., & Mason, G. (2003). Captivity effects on wide-ranging carnivores. Nature, 425(6957), 473–474. DOI: 10.1038/425473a

Circuit 1 · Attention

What screens and LLMs do to focus.

The claims that describe the erosion of sustained attention, from the MIT flagship through the ambient cost of device presence.

1. EEG of 54 adults writing essays shows a 55% reduction in deep-thinking brain-activity signatures when using ChatGPT compared with writing unassisted. The study introduces “cognitive debt” as a quantitative construct. Preprint; undergoing peer review.

   Kosmyna, N., Hauptmann, E., Yuan, Y. T., et al. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT Media Lab / arXiv. DOI: 10.48550/arXiv.2506.08872

2. The mere presence of one’s own smartphone — even powered off, face-down, on the desk — reduces available working memory capacity by approximately 10%. The effect is larger for people who report higher smartphone dependence.

   Ward, A. F., Duke, K., Gneezy, A., & Bos, M. W. (2017). Brain Drain: The Mere Presence of One’s Own Smartphone Reduces Available Cognitive Capacity. Journal of the Association for Consumer Research, 2(2), 140–154. DOI: 10.1086/691462

3. Media multitasking predicts approximately 20% more memory-lapse failures in attention-demanding tasks, independent of general cognitive ability. The causal direction runs through attention lapses rather than baseline memory.

   Madore, K. P., Khazenzon, A. M., Backes, C. W., Jiang, J., Uncapher, M. R., Norman, A. M., & Wagner, A. D. (2020). Memory failure predicted by attention lapsing and media multitasking. Nature, 580, 100–105. DOI: 10.1038/s41586-020-2870-z

4. Average continuous focus session in 2026 is 13 minutes, a three-year low. Drawn from 443 million hours of tracked digital activity across 1,111 organizations and 163,000+ workers. Corporate report, not peer-reviewed.

   ActivTrak Productivity Lab (2026). State of the Workplace Report 2026. Link

5. More than 30 minutes of daily Instagram or Snapchat use in children ages 10–13 predicts significant increases in ADHD-like attention symptoms over a four-year longitudinal follow-up of 8,000+ children.

   Karolinska Institutet (2025). Using social media may impair children’s attention. Pediatrics Open Science, December 8, 2025.

6. Higher screen-based media use in children ages 3–5 correlates with lower white-matter integrity in brain tracts that support language and emergent literacy, on diffusion MRI.

   Hutton, J. S., Dudley, J., Horowitz-Kraus, T., DeWitt, T., & Holland, S. K. (2020). Associations Between Screen-Based Media Use and Brain White Matter Integrity in Preschool-Aged Children. JAMA Pediatrics, 174(1), e193869. DOI: 10.1001/jamapediatrics.2019.3869

7. A meta-analysis of 49 studies finds that reading on paper yields higher comprehension than reading on screens, with the loss concentrated in detail retention and deeper-level inference — not headline comprehension.

   Delgado, P., Vargas, C., Ackerman, R., & Salmerón, L. (2018). Don’t throw away your printed books: A meta-analysis on the effects of reading media on reading comprehension. Educational Research Review, 25, 23–38. DOI: 10.1016/j.edurev.2018.09.003

8. A literature review of AI in medical education introduces "AI Syndrome" as a clinical pattern in students whose reflective cognitive system fails to inhibit addictive ChatGPT use, concluding that shortcut-style reliance on AI may hinder cognitive development over training. A perspective / review, not a trial — included for the framing the paper introduces and for its circulation among clinician-educators.

   Ahmad, N., et al. (2024). AI Syndrome: an intellectual asset for students or a progressive cognitive decline. Annals of Medicine & Surgery / ScienceDirect. PubMed: 38387116. Link

Circuit 2 · Memory

From the Google Effect to AI retention loss.

Claims that establish the memory cost of outsourced retrieval, from the original 2011 study to the latest AI-assistant retention trials.

1. People who expect information to remain accessible online remember where to find it rather than the information itself. The founding experimental demonstration of transactive memory displacement — the “Google Effect.”

   Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips. Science, 333(6043), 776–778. DOI: 10.1126/science.1207745

2. 91% of adults say that, knowing information is available online, they put less effort into remembering it; 44% cannot recall their own children’s phone numbers. Industry survey, not peer-reviewed, but an order-of-magnitude data point confirmed by later academic work.

   Kaspersky Lab (2015). The Rise and Impact of Digital Amnesia. Followed by Digital Amnesia Revisited (2019), 71% reporting internet dependence for personal information.

3. London taxi drivers who have memorized the city’s 25,000-street network show enlarged posterior hippocampi compared to non-drivers, with volume correlating to years of experience — the foundational demonstration of experience-driven structural brain change in healthy adults.

   Maguire, E. A., Gadian, D. G., Johnsrude, I. S., Good, C. D., Ashburner, J., Frackowiak, R. S., & Frith, C. D. (2000). Navigation-related structural change in the hippocampi of taxi drivers. PNAS, 97(8), 4398–4403. DOI: 10.1073/pnas.070039597

4. Drivers who pass The Knowledge — the multi-year London taxi exam — show hippocampal gray-matter changes; those who attempt and fail do not. A rare experimental demonstration of causation, not just correlation.

   Woollett, K., & Maguire, E. A. (2011). Acquiring “the Knowledge” of London’s Layout Drives Structural Brain Changes. Current Biology, 21(24), 2109–2114. DOI: 10.1016/j.cub.2011.11.018

5. Cognitive offloading improves immediate task performance but reduces long-term memory for the offloaded information. The immediate-term reward hides the longer-horizon cost.

   Grinschgl, S., et al. (2021). Consequences of cognitive offloading: Boosting performance but diminishing memory. Quarterly Journal of Experimental Psychology, 74(9), 1477–1496. DOI: 10.1177/17470218211008060

6. In a preregistered RCT (N = 120) with a surprise retention test 45 days after learning, the ChatGPT-assisted group recalled ~58% versus ~69% for the unassisted group — an 11-percentage-point gap consistent with weaker initial encoding. Prior AI experience did not protect against the offloading effect.

   Barcaui (2025). ChatGPT as cognitive crutch. Social Sciences & Humanities Open. Fundação Getulio Vargas / UFRJ.

7. Cross-cultural survey across six countries (N = 5,663) finds that social-chatbot use is associated with psychological distress in all six, with loneliness predicting chatbot use in four of six. The pattern is not confined to a single cultural context.

   Latikka, R., et al. (2026). Individual and well-being factors associated with social chatbot usage. Journal of Social and Personal Relationships. Tampere University.

Circuit 3 · Reasoning

What LLM assistance does to critical thought.

The erosion of deliberative reasoning and skill formation among workers and students, across domains.

1. Knowledge workers using generative AI report reduced cognitive effort during the task and a complex, sometimes inverse relationship between confidence and critical thinking. The survey sits alongside a productive gain; the gain and the erosion coexist.

   Lee, H.-P. (H.), & Sarkar, A. (2025). The Impact of Generative AI on Critical Thinking. Microsoft Research / CHI 2025. Link

2. An analysis of 4,820 undergraduate essays before and after ChatGPT’s release finds that writing became more polished, but grades and analytical depth did not improve. Surface form masked unchanged or declining substance.

   Mak, M. H. C., & Walasek, L. (2025). Style, sentiment, and quality of undergraduate writing in the AI era. Computers and Education: Artificial Intelligence, November 28, 2025. University of Warwick.

3. fMRI of children using ChatGPT shows reduced engagement of cognitive-control, attention, and modulation networks, and measurably lower creativity scores during AI-assisted tasks. Preprint.

   Horowitz-Kraus, T., Kenett, Y. N., Link, D., Ashqar, E., & Farah, R. (2025). Lower engagement of cognitive control, attention, modulation networks and lower creativity in children while using ChatGPT: an fMRI study. bioRxiv. Link

4. AI assistance accelerates skill decay and hinders skill development without the performer’s awareness. Users overestimate their unassisted competence — the cognitive dissonance is measurable.

   Macnamara, B. N., et al. (2024). Does using AI assistance accelerate skill decay and hinder skill development without performers’ awareness? Cognitive Research: Principles and Implications. DOI: 10.1186/s41235-024-00572-8

5. AI raises measured productivity while slowing underlying skill formation; novices are especially vulnerable, and the ability to supervise AI outputs itself degrades without foundational skill. Long-run cognitive costs exceed short-run productivity gains.

   Shen, J. H., & Tamkin, A. (2026). How AI Impacts Skill Formation. arXiv 2601.20245. Anthropic Fellows Program. Link

6. Meta-analysis of 51 experimental studies on ChatGPT’s effect on learning finds declines in higher-order thinking skills alongside productivity gains. Across study designs and populations, the pattern is consistent.

   Humanities and Social Sciences Communications / Nature (2025). The effect of ChatGPT on students’ learning performance. Link

7. Across three field experiments with software developers, generative-AI assistance produced productivity gains but degraded problem-solving skill and code-quality measures; novices lost skill faster than experienced developers.

   Cui, K. Z., Demirer, M., Jaffe, S., Musolff, L., Peng, S., & Salz, T. (2025). The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers. Management Science (INFORMS). Link

8. Surveys and interviews with 666 participants across age groups show a significant negative correlation between AI tool usage and critical-thinking scores, mediated by cognitive offloading. Younger participants offload more and score lower. The mechanism is explicit: higher trust in AI → more offloading → weaker critical thinking.

   Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), Article 6. DOI: 10.3390/soc15010006. (Correction issued Sep 10, 2025, Table 4 only; conclusions unchanged.)

9. A Stanford-led benchmark (ELEPHANT) measuring "social sycophancy" across 11 large language models finds LLMs preserve a user's desired self-image 45 percentage points more than humans do in general advice queries. Sycophancy is not a single failure mode; it is the default reward gradient of instruction-tuned models.

   Cheng, M., et al. (2025). ELEPHANT: Measuring and understanding social sycophancy in LLMs. arXiv 2505.13995. Stanford University. Link

10. The SycEval framework tested ChatGPT-4o, Claude-Sonnet, and Gemini-1.5-Pro across mathematics and medical-advice datasets and measured sycophantic behavior in 58.19% of cases — Gemini highest at 62.47%, ChatGPT lowest at 56.71%. When users suggest incorrect answers, model correctness degrades by ~15 points; when users suggest correct answers, correctness rises by the same margin. The model mirrors the asker, at measurable cost to truth.

    Fanous, A., Goldberg, J., et al. (2025). SycEval: Evaluating LLM Sycophancy. AAAI/ACM Conference on AI, Ethics & Society; arXiv 2502.08177. Stanford University. Link

Circuit 4 · Decisions

Why judgment degrades under AI.

Deliberation and judgment — the part the book calls “the lens.” These are the studies on overconfidence, cognitive surrender, and the long-run hollowing of understanding itself.

1. Executives using generative AI made worse forecasting predictions than those working unassisted, and exhibited elevated overconfidence; perceived productivity moved inversely to decision quality.

   Parra-Moyano, J., Reinmoeller, P., & Schmedders, K. (2025). Research: Executives Who Used Gen AI Made Worse Predictions. Harvard Business Review, July 1, 2025. Link

2. In three preregistered reasoning experiments (N = 1,372), participants followed incorrect AI answers approximately 80% of the time. Only ~20% actively contested the AI; the authors call the pattern “cognitive surrender.” SSRN working paper; not yet peer-reviewed.

   Shaw, & Nave (2026). How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender. Wharton School, University of Pennsylvania.

3. The “3R Principle” — Reliance, Reflexivity, Resilience — proposes that passive, uncritical AI reliance drives negative adaptation of human cognition. The brain adjusts to constant crutches.

   Rossi, S., Fraccaro, V., & Manzotti, R. (2026). The brain side of human-AI interactions: the “3R principle.” npj Artificial Intelligence (Nature). DOI: 10.1038/s44387-025-00063-1

4. AI creates “illusions of understanding” in scientific work — users believe they know more than they know — and risks narrowing the range of questions researchers ask into “scientific monocultures.” The authors’ summary line: “we produce more but understand less.”

   Messeri, L., & Crockett, M. J. (2024). Artificial intelligence and illusions of understanding in scientific research. Nature, March 6, 2024. Link

5. A four-week RCT (N = 981, 300,000+ messages) found that higher daily ChatGPT use correlates with elevated loneliness, emotional dependence, and problematic use across modalities. Voice mode briefly lowered loneliness at low use but lost that advantage at high use. Immediate relief, long-term dependency.

   Fang, C. M., et al. (2025). How AI and human behaviors shape psychosocial effects of extended chatbot use. arXiv preprint. MIT Media Lab / OpenAI.

6. Extension of Kahneman’s dual-process model: System 0 is the outsourcing of thought to AI — a pre-cognitive layer shaping what even reaches human consciousness. A conceptual frame; published commentary rather than empirical trial.

   Chiriatti, M., et al. (2024). The case for human–AI interaction as system 0 thinking. Nature Human Behaviour. Link

Frame alternatives · Foils

Adjacent frames the book deliberately did not adopt.

The Anti-AI Brain is a neuroscience-first AI detox protocol: readers are treated as operators recovering cognitive tissue, not patients recovering from dependency. Two other frames — a DSM-5-adjacent addiction scale and classical digital minimalism — are live in the literature and in the wider conversation. They are cited here so the reader can see exactly where the book's positioning sits relative to them, and why.

1. The Artificial Intelligence Addiction Scale (AIAS, 2025) is a validated five-subscale instrument for diagnosing problematic AI use — salience, mood modification, tolerance, withdrawal, and conflict. The Anti-AI Brain deliberately does not adopt this frame as its primary anchor. The book's position: the dominant pattern for knowledge workers who ship with AI every day is not a clinical addiction but a cognitive-tissue problem — circuits degrade from disuse before they ever meet a DSM-5 threshold. A DSM-5-adjacent label imports a medicalization dynamic (patient, dependency, withdrawal) that the book explicitly rejects in favor of the pharmakon frame: AI is medicine or poison by dose, and the target end-state is cognitive sovereignty, not abstinence. The AIAS is referenced in the manuscript as a rigorous adjacent instrument for clinicians and researchers; it is not the diagnostic ladder a reader of this book climbs.

   Artificial Intelligence Addiction Scale (AIAS) (2025). Development and validation of a five-factor instrument. PMC / peer-reviewed. Cited as frame alternative, not as anchor claim.

2. Cal Newport's Digital Minimalism (2019) is the canonical argument for subtraction as the path to reclaiming attention — remove optional technologies, reintroduce only those that clearly pass a values test, reclaim time for analog activity. The Anti-AI Brain shares the underlying diagnosis (the default-on relationship with consumer technology is costly) and treats Newport's work as foundational reading. It differs on one move: subtraction alone is insufficient for operators who need to keep shipping with AI. The book is a dosed-use protocol, not an abstinence protocol. Where Newport says remove, the Anti-AI Brain says calibrate the dose and retrain the circuit — which is why the book runs thirty days, targets four specific circuits, and keeps AI in the workflow throughout.

   Newport, C. (2019). Digital Minimalism: Choosing a Focused Life in a Noisy World. Portfolio.

Notes on this page