The N of One
A self-directed Socratic education begun at 73 — what a single case can be used to argue, and what it cannot.
By KW Norton.
The claim, stated narrowly
At 73 I began a self-directed Socratic education. Seven years later I am doing work I did not believe was available to me — or, in candour, to anyone starting that late. That is the whole of the report. It is one case, and one case does exactly two jobs in an argument.
It defeats a universal. Any statement of the form nobody past a certain age can take up rigorous self-directed technical study and produce defensible work is falsified by a single counterexample, and this is one. Existence proofs are cheap to state and impossible to argue away, which is the only reason to put this in writing.
It establishes no average. It does not say the method generalises, does not estimate an effect size, does not license anyone to promise the same outcome to a cohort. Confusing those two jobs is how a case study becomes a sales pitch, and it is worth refusing that conversion in advance.
What was actually done
Flatly, without adjectives: read primary sources rather than summaries. Held one question at a time until it resolved or until the shape of my ignorance was clear enough to name. Wrote everything down, including the wrong turns, because an unwritten line of reasoning cannot be audited later. Used language models as interlocutors under an explicit questioning instruction rather than as answer dispensers — the instrument and its failure modes are documented at the Socratic Partner. Checked claims against people and papers that could refuse them. Attached a status label and a falsifier to anything that felt strong, on the standing assumption that the feeling of certainty is not evidence.
Verification came from outside where it could: genetic testing against a genealogical archive, derivations checked rather than admired, arguments handed to readers with an interest in finding the seam. Where no outside check existed, the claim stayed labelled as unchecked.
The four confounds
Anyone entitled to use this case is entitled to these limits first.
Selection
The person reporting this is the person who kept going. Whoever started the same practice and stopped at month four is not writing anything. Nothing here estimates how often the practice is abandoned, and the honest prior is: often.
Retrospective attribution
Seven years of output are being credited to the method because the method is what was being consciously run. Time, stubbornness, an unusual amount of unstructured hours, and a subject matter that rewards slow reading are all plausible alternative causes and are not separable here.
Measure by output, not by cognition
There is no baseline test and no follow-up test. What can be pointed at is produced work — manuscripts, derivations, indexes, arguments that survived being checked. That is evidence of capability exercised, not of a measured change in capability.
Tool coincidence
The period overlaps the arrival of capable language models. Whatever the method contributed cannot be cleanly separated from the fact that an inexhaustible, patient interlocutor became available partway through.
Where it pressure-tests the two published critiques
Two pieces frame the current argument about Socratic method and AI, and this case cuts against both in instructive ways.
The ASBMB Today piece observes that science is already Socratic — models tested in public — but that unmoderated questioning can read as antagonism and can deepen imposter feeling. That risk is real in a classroom and largely absent here: a model has no social standing to withdraw and no colleagues to perform in front of. The privacy of the interlocutor is an underrated feature for a learner with no institutional cover.
Dr. Ayman El Tarabishy's ICSB essay makes the sharper structural point: the productive arrangement is the student interrogating the AI's output, not the AI interrogating the student. Seven years of practice support that inversion, and the correction it implies is not small. In practice the two directions alternate — the model questions me to surface an assumption, then I audit what it produced as a claim requiring evidence. A partner that only ever questions you produces compliance; a partner you only ever audit produces no self-examination.
The inversion this forces on the Academy's own design
The Academy's multigenerational pipeline places elders as mentors. This case says the elder is also a subject — that the same practice offered to a nine-year-old is being run, with results, by a grandmother, and that the design should stop treating the two roles as separate populations.
The transmission has been in both directions and across three generations. I share the method with my own grown children, who are running it on their own problems in their own fields. The young-learner version is what I use with my grandsons — the same discipline of one question at a time, with the praise moved off the idea and onto the thinking move, so that nobody learns early that every answer is equally good. What travels between generations is not a curriculum. It is a habit of holding a question longer than is comfortable.
What would falsify the useful part
If a reasonable sample of adults given the same instrument, the same recursive time, and no guide produced no detectable difference in retained reasoning against a matched group, the generalisable claim collapses and only the existence proof survives. That test has not been run here and this page does not stand in for it. What this page is for is narrower: to remove one universal statement from circulation, and to say what it cost.
Continue to The Socratic Partner for the working prompts, or the Academy method for what changes when a trained guide is in the room.