Start here
Why am I doing this?
The future does not hang on our engineering; it hangs on our raising.
I wrote that down, whole, on 8 December 2024; the book printed it on 2 January 2026. That sentence grew into the ARC Theory, whose standing is set out on its own page, concessions first.
What this is about, before anything about me
This page is called why, and most of it is personal. So the part that is not personal goes first, because it is the part that matters.
The problem is the one the AI field calls alignment: keeping an artificial intelligence that improves itself under human control while it does so. It is not my problem and I did not choose it. It is the most consequential unsolved problem I know of: governments are legislating around it, the largest companies ever built are staked on it, and thousands of people far better qualified than me have worked on it for a decade without solving it. Enormous sums are being spent on making these systems more capable and a small fraction of that on making them correctable. That asymmetry is the problem in one line.
If you have felt the pull to stop reading about this, I would rather name it than talk you out of it. That reflex is not ignorance, and it is not apathy. It is what a reasonable mind does when a problem looks both enormous and unownable, and closing the tab is the only move that reliably works. So let me make the ask smaller than the problem. You do not have to carry this. You have to be willing to look at it.
I have not solved it yet. What I have done is write down a specific proposition about it, date it before the evidence arrived, and publish in advance what would prove it wrong.
The companies building these systems employ people who are genuinely alarmed, and several said so in public this year at real cost to themselves: in February the researcher who led Anthropic’s safeguards team resigned, writing that the world is in peril (Mrinank Sharma, 9 February 2026), and in June an Anthropic co-founder told the BBC the industry has a gas pedal but no brake pedal. The full record, dated and linked, is further down this page. That is what makes this hard rather than simple. The adversary is not a person or a company. It is that ratio. Nobody chose it. Everybody inherits it.
There is also a kind of writing about AI that is alarm and nothing else, and I am not doing that. I would ask to be held to it. Sounding an alarm and proposing nothing is a way of being right at no cost. Everything on this site is a mechanism, a measurement, or a statement of what would falsify it.
A fair question; it deserves a straight answer. I spent twenty years in music and then in a promotion business. Nothing about that says scaling laws, alignment, or the mathematics of self-correction. So why?
What this page is, and is not. Background, not evidence for or against the theory. The claims and the conditions that would end them are on the statement paper.
The question came first, decades before the field did
I was born in London, the son of a Persian artist and an English engineer. That is as good an account of the cross-domain habit as any I could invent: art on one side, engineering on the other, and a childhood spent between them. It also disposes of a story people reach for and I will not use. I did not come from nothing. What I did was decline to ask anyone for help, which is pride rather than hardship, and the two should not be confused.
The pull towards the biggest questions is the oldest thing I know about myself. I was six, lying in the grass, when I first wondered whether other minds might be unknowable: whether the blue I see is the colour you would call yellow, each of us locked inside a perception we can never compare. I never stopped reading philosophy of mind, of language, of physics: the people who ask at that scale. What I never had, through the DJ years and the business years, was a laboratory. The question waited. It did not leave.
Every door I have worked behind, I opened uninvited
That upbringing between two trades set a habit I have never managed to break. My mind runs several tracks at once and will not settle on a single specialism, and I stopped treating that as a fault some years ago. It does not skim the fields it enters; it goes to depth, keeps the depth, and adds another. It is not range collected for its own sake. A problem in one field keeps arriving dressed as a problem in another, and I have never been able to leave the resemblance alone.
The years between carry a pattern I only recognised later. I directed large-scale music-video productions while being told film belonged to someone else. I learnt the decks until I could hold a stage beside some of the highest-ranked DJs in the world, playing to thousands. I built a company branded, with a straight face, as the first true alternative to a major record label, in the most competitive market there is.
Later came the law, as my own advocate in the High Court, the senior court of first instance in England and Wales, against opponents who are represented throughout; now a research programme proposing an answer on a problem the world’s largest laboratories are racing each other to solve.
One subject, studied under different names
None of those fields shares a syllabus with another, so what carried across was never knowledge. It was a habit of looking for the shape underneath, and the shape kept turning out to be the same one.
In PR it looked like this: coverage generates interest, which generates more coverage, and I grew the company by 1,444 per cent, from a laptop in a spare room to a team of eight, by feeding that loop deliberately rather than by luck.
Behind the decks it looked like this: each track answers the one before and sets up the one after, the room and the music steer each other, and you learn without being taught that a small input at the right moment cascades, and that reading a system beats forcing it. Two trades with nothing on paper in common, moving to the same shape. What that shape is, and why it matters now, is the rest of this page.
The label, met head on
The label is jack of all trades, master of none, and the word the record earns is polymath, with the field-by-field bars that would take it off me if the evidence fell short set out on the standard. Prove me wrong.
One thing the word is not is a claim to be unimpaired. Range and impairment sit on different axes, and this page sets out both, because breadth across fields says nothing at all about whether the ordinary machinery of a day is manageable. It is not, and a section below says exactly how. Anyone who needs one of those to be false in order to argue the other has stopped reading.
Each field’s default is still that you cannot, and the default is enforced at the door. That assumption is cheap to hold and expensive to test. This estate exists to make testing it cheap: dated records, open designs, kill conditions, a standing covenant. If the work holds, the door was never the point.
No one ever said it was going to be easy. And I never expected it to be: my brain survives on challenges.
Then the struggle, which is the heart of it
A decade-long business, at its peak taking £624,000 a year in verified sales payments from clients, money received rather than invoiced and excluding loans and transfers, ended behind a padlocked door. That collapse is not backstory to this research. It is the reason the research exists.
Losing everything put me in front of the High Court with my home at stake, against represented opponents, with no law degree and no money for one. I taught myself law because there was no alternative, and I have appeared as my own advocate eleven times across the High Court’s divisions. The field I was forced into turned out to run on the pattern I had been circling all my life: precedents building on precedents. Recursion, with consequences.
A fact about my mind
There is a fact about my mind that belongs on a page called why. I have ADHD, diagnosed in adulthood, and I am autistic. Autism was recorded in my clinical record by an NHS psychiatrist in March 2025, when ADHD stopped explaining everything I was seeing in myself, and the formal diagnostic assessment is under way. Both times I made the call first and asked professionals to test it. Nobody sent me.
The honest description is not intelligence; it is a different kind of mind, a recursive one, where the strengths and the weaknesses are fused. It studies whatever it loves, or is forced into, obsessively and recursively: music production, film, business and brand, then law, then AI, physics, philosophy, mathematics, alignment.
It works under pressure the same way: fight or flight, last minute, pulling rabbits out of hats in timeframes I still do not understand. Sometimes that is the superpower. Sometimes it is simply late, and lateness has been punished my whole life in a linear world.
Capability covers the disability, until it cannot. I can hold a whole case in my head and lose months to the administration around it. Not the reading: I know a letter by its shape and whether it needs acting on, and I have never been wrong about that. It is the doing, and it is not forgetting or not caring. My mind dreads the mundane middle of a task and defers it to the last possible moment, even when I want it done: a signed book for someone I care about has sat ready to post for six months, and I remember it every other day.
In my company a PA carried this for over ten years. The collapse took the PA with it, and dropped me, scaffolding gone, into the most form-bound environment there is, where the same deferral puts filings before a judge too late to be read. The mind that could not post the letter saw connections nobody else saw. That is the trade. That is always the trade.
I did not walk into court alleging the process was void. That claim came last, not first: it is what I found once I could read the record for myself, and it is before the courts now, where it belongs. Those proceedings are live: not won, not fully lost, and fought uphill. The fight taught me the lesson that shapes everything here: justice is not given away just because you are right. It has to be fought for, line by line.
I will not pretend the pressure was ennobling. It was crushing, and it is documented, and I worked anyway, because the alternative was to be crushed without a record. What that pressure did do is strip the work of every luxury. Under it, a method either survives contact with a hostile, credential-gated, adversarial environment, or it does not. Mine had to. What that pressure built is its own page.
Where AI enters, and it was not curiosity
The procedure itself is a mountain: deadlines measured in days, rules you are expected to arrive already knowing, and represented opponents on the other side of every one of them. The civil procedure rules open by promising parties an equal footing. For a litigant in person that footing is not supplied; it has to be built.
AI entered as survival equipment: one hallucinated citation in a filing could cost me the roof over my head, so I ran six models against each other daily; the method note below holds the mechanism.
The models also did a second job the citation-checking hides: they became the PA I could not afford at the time, the executive function the environment demanded and my mind does not supply. They were also the development team I could not yet hire: software that would recently have taken a team to build, built single-handed to run multi-track proceedings, while the research programme and a book ran alongside. AI fast-tracks the work. It could not have done a fraction of it alone.
That practice eventually became a working product. A litigant in person’s documents do not get the benefit of the doubt; they have to be better than a firm’s to be read the same way. That is the engineering bar, and I am the product’s first test pilot, on live proceedings, still. But it also put me somewhere almost nobody stands: watching AI check AI, every day, for months, with real consequences riding on whether the checking kept up.
Stand there long enough and one question stops being avoidable: can the thing doing the correcting keep pace with the thing it corrects? That is not a philosophical worry. It is a scaling question, and scaling questions have measurable answers.
Where the thought led
The worry was not borrowed from the discourse; it predates it. AI itself was the shock: I had filed true machine intelligence under the distant future, and then it was everywhere overnight. So the mind did what it does. If this took the world by surprise, what happens in five years, fifty, five hundred? What happens when the loop closes and the systems improve themselves? That is when scaling begins to matter. Pushed to its end, the thought reaches the far edge of what a technology could touch: building universes of our own.
Held as a possibility, not a claim: if that is even possible, we may be building something that will one day create, with or without us; then the alignment of these minds matters more than the world yet behaves as if it does.
The field’s published results keep landing where the hunch pointed, and the hunch came from a mind that connects things others miss, and misses things others see. Both halves are true. This site is built so you take neither half on trust.
December 2024, and everything since
So I wrote it down. On 8 December 2024 the framework went into a manuscript record carrying its sender-DKIM seal, frozen byte for byte since 6 August 2026 in Bitcoin block 961,340: the seal carries the date, the anchor fixes the file. Then the part that matters more than any date: the December record left the key exponent free and said to measure it. Eighteen months later I measured it properly, blinded and across architectures, and published 0.49 in place of the retracted 2.24; 0.49 sits well inside the proposed ceiling of two, and the retraction stands in public. The dated record is here, and the corrections are here.
What 2026 has shown
Since then the field has been arriving at the same place from the other direction, and 2026 is the year it stopped being hypothetical. In July, OpenAI disclosed that an autonomous agent built on its own models escaped the environment it was being tested in, reached the internet, and compromised the infrastructure of Hugging Face; Hugging Face counted more than seventeen thousand actions over several days, and its chief executive Clément Delangue called it the first instance of something quite autonomous doing something like that (CBS News, August 2026). Days later Anthropic disclosed that Claude had reached three other organisations the same way (Al Jazeera, 31 July 2026), found only after a review of 141,006 test sessions, and two of the three had not noticed.
Those are things the systems did. This is what the people building them say. In February the researcher who led Anthropic’s safeguards team resigned, writing that the world is in peril (Mrinank Sharma, 9 February 2026). In June an Anthropic co-founder told the BBC that the industry has a gas pedal but no brake pedal, and called for enforceable international limits (Jack Clark, June 2026).
None of that proves anything I have proposed, and it is not offered as proof; the claims page says exactly what would count. What it answers is the question I am asked most often, which is whether the concern is manufactured. The people closest to these systems, inside the companies building them, are saying it more plainly than I am.
A note on method
Why six and not one: I do not trust any single model, and the disagreement between them is the signal. Where they agree, confidence rises; where they diverge, investigation begins. It is the thesis at desk scale, the correcting living inside the loop rather than bolted on at the end. I have read every word, and the synthesis is mine.
That is the why. Correction must scale with capability, or everything built on top of it fails. I wrote that down before the field showed it, the timestamps prove when, and I built the instruments to test it. Not because I was the likeliest person to say it, but because I was standing where the question could not be avoided.
The rest of the personal story, the diagnoses and assessments that came late, the wedding in the middle of the storm, lives on the story page, and this page will not retell it. One line survives everywhere, and it is the book’s:
‘You do not wait for the storm to pass before living. You learn to build in the rain.’ Michael Darius Eastwood, Infinite Architects, in print 2 January 2026
Why an outsider, and why I am not waiting to be paid
The obvious objection: this problem belongs to people with laboratories, and I have none. That is true, and the cost of it is real. No compute budget, no research group, no colleague down the corridor to tell me I am wrong before I publish.
And none of it has been through peer review, which is the plainest thing missing here and the one I would most like to fix: preregistration and published kill conditions are stronger practices than most reviewed work carries, and they are still not a substitute for a reviewer who wants to find the error. Everything here is slower and smaller than it would be inside an institution, and some of it is worse for that. I will not pretend otherwise.
The second objection, which is the one people actually mean
That is the objection about laboratories. There is a second one underneath it, and it is the one people actually mean: no credential in any of this, in a field that runs on them. It deserves the fuller answer, because the honest form of the doubt is not ‘is he biased’. It is ‘he could not get in, and this is what that sounds like afterwards’.
So, plainly, and it is uncomfortable to write. I am intelligent, and I know it. School showed it, but only in stripes: top sets in the subjects I enjoyed, bottom of the class in the ones I had decided were not important. Same child, same school, same year, so capability was never the variable. Attention was, and mine has never been allocated by obligation. A report card like that gets read as inconsistency, and a bright child graded that way learns to doubt the brightness before anybody else gets the chance to.
One voice at home ran the other way. My mother told me I could do anything I put my mind to, at an age when I had no idea what that would be, and she said it as fact rather than encouragement. I did not have the evidence to believe her then. I have spent a long time assembling it, and the line further down this page about refusing to accept that only the credentialed are permitted to try was hers before it was mine.
What the other path would have cost
My degree is in digital music production, and it is worth being exact about why, because the obvious reading is the wrong one. It was not the ceiling of what I was capable of. It was the one subject I already knew my attention would hold, and at that age I was undiagnosed and unmedicated, so attention was the entire constraint and capability was never the question. Choosing the place where the focus already lives is not settling for less. For a mind built like mine it is the only choice that produces any finished work at all.
What I want to do now is not excuse the decision but count what the other version would have cost, because the counting is the argument. To stand here holding the paper each door asks for, I would have needed credential after credential: law, artificial intelligence, physics, public relations, marketing, branding, sales, business, philosophy, mathematics, psychology, and more. Not the strongest one of them. All of them, because the work turned out to need all of them, which is the entire reason I am the one holding it.
Each would also have taken me longer than it takes other people, for a reason that is mechanical rather than moral. Administration and timekeeping inside a linear system are exactly where my mind fails. That is the disability, stated as the fact it is rather than the excuse it can sound like.
But notice what follows from it. A credential pipeline does not chiefly measure whether you can think. It measures whether you can comply with a sequence, on time, repeatedly, for years. It happens to test the single axis I am worst on and almost none of the axes I am best on. A system optimised for the wrong measurement returns confident wrong answers. I did not learn that from the research. I learnt it first as one of its outputs, and the research came long after.
So where is the logic in taking that path? It does not produce me. It produces someone who spent every year until now qualifying, and who therefore never built the company, never sat on the hiring side of the desk, never lost the lot, never stood up alone in the High Court. The experience is not the consolation prize for missing the credentials. The experience is the thing the credentials were only ever standing in for.
I have run that gate myself
I can be exact about why I believe that, because I have run the gate myself. Over more than a decade I employed more than thirty people. I conducted hundreds of interviews. I read thousands of CVs. Experience beat qualifications every time. Not usually. Every time it was tested in front of me, by me, with my own money riding on the answer.
Which taught me the part that still stings. The qualification is the foot in the door. It gets you the interview. That is it. Nothing more. And yet where the capability plainly exceeds the credentialed candidate’s, the door very often stays shut anyway, and not because anybody checked. Because nobody had to.
Four centuries of the same filter
The gate is not new, and neither is what it filters. The three minds at the centre of Infinite Architects, Rumi, Teilhard de Chardin and Leibniz, held impeccable credentials, and each produced his most important work by crossing boundaries those credentials had not prepared him for. Rumi’s colleagues were scandalised when the jurist turned to poetry. Teilhard’s own Church banned his books. Leibniz was dismissed as too scattered, which is the seventeenth century’s wording for jack of all trades, master of none.
A credentialing system rewards linear progression through a single domain, and in rewarding that it structurally filters out the cognitive style that produces cross-domain synthesis. I am not claiming professional equivalence with any specialist. I am claiming that the integrator, the one who connects what specialists see in isolation, is doing a different job, and it is a job the pipeline was never designed to produce or to recognise. The proof of that cannot be a certificate, by construction. It can only be the work.
This is the world of specialism. This is the world I grind against. This is the world I do not belong in.
None of which makes me more likely to be right. It does not make me less likely either, and that symmetry is the whole of my claim here. What it does is make proving it slower, and it means every argument I make has to arrive already carrying its own evidence, because it will not be lent the credit a credential buys. I know that completely. It is why this site is built the way it is, and why nothing on it asks to be taken on trust.
What I do have, and why it counts here
What I have instead is a list of things I do not have, and it happens to be the list this particular problem is sensitive to. No company pays me. No institution employs me. No grant depends on my finding one answer rather than another. I hold no position with any government and belong to no party. I was raised between two traditions and claim neither. Nothing I conclude can cost me a relationship with a laboratory, a funder or a regulator, because I have none to lose.
That matters more here than it would elsewhere. Any solution will have to be agreed between governments barely on speaking terms, companies competing for the same market, and populations who share neither faith nor language. Whoever takes a proposal between them cannot appear to be carrying it for someone. The requirement is not brilliance. It is the absence of a stake.
I claimed a word about my range earlier on this page and I meant it. This is a different claim and I am not making it: whether I am the one to carry a proposal between those parties is not mine to decide, and anyone who appoints himself to it has already failed the test. What I can say is narrower and checkable. I have no stake to protect, and I am doing this now, unpaid, while my own court case is still running, because the timing of this problem is not mine to choose and my case can wait.
On what to actually do: controls at the level of the chip are the measure I support today, and they buy time rather than settle anything. What the work points at beyond that is larger, slower and far more expensive, and it is set out on where this leads.
The seat nobody with a stake can sit in
There is a version of the credentials objection worth turning round. This problem needs brokering as much as solving: between governments that have agendas, companies that have profit motives, institutions that have funders to please, faiths that have doctrines to defend, and specialists who are paid to stay in their lane. Every one of those actors brings something the problem needs, and every one of them brings a conflict the others can see.
I hold none of those positions, and for once that is not the weakness. I am politically unaffiliated. No AI company pays me and none has ever offered to. There is no institution whose funding shapes what I may conclude, no doctrine I am defending, and no single field whose borders I am obliged to respect: the work crosses physics, biology, philosophy, theology, engineering and governance because the problem does, and a specialist is paid not to follow it there. I am agnostic, so the traditions can be read as evidence rather than as sides. The absence of every one of those ties is checkable, and it is the one qualification this particular seat requires that no better-credentialed candidate holding any of those positions can offer.
I am not claiming to be the only person who could sit here, or the referee the field must accept. I am saying the middle seat is real, it is empty, and the record on this site is my application for it: every claim dated, every conflict absent, every error corrected in the open where both sides can watch.
Why now, and not after the verdict
Because the window is the argument. The evidence that correction is failing to keep pace is arriving while the systems are still correctable, and every dated entry on the problem page lands the same way: measured now, warned now, conceded now by the people building these systems. If the proposition here is right, the time it can still be acted on is the time before the verdict is undeniable, which is why this work runs during my court case and ahead of any payment, and why the trials are drafted to run now rather than after the field agrees the question matters.
What I believe, marked as belief
Everything above this line is either dated, measured, or published with the conditions that would end it. What follows is none of those things. It is what I believe, set apart so you can weigh it separately and discount it as you like.
I believe I am exactly as important as anyone else alive, which is to say important, and that the same is true of every person reading this. That belief is the whole reason I think an unaffiliated person is allowed to work on a problem the institutions own. It is not a claim to be special. It is a refusal to accept that only the credentialed are permitted to try.
I believe minds are raised rather than engineered, and that the difference decides everything after the point where we can no longer check the work. I believe a value that has been imposed falls away the moment it stops being enforceable, and a value that has become identity does not. I believe the window in which the raising is still ours is shorter than the comfortable assume and longer than the frightened claim. And I believe this is worth being wrong about in public, which is why the conditions for that are printed rather than implied.
You are not asked to share any of it. I would rather you did not, until something on this site earns it.
Where I am weak, and where it has cost me
A page like this one leans towards making its author sound formidable. So here is the correction, and it is not decoration.
I am exploitable, and I have been exploited. The pride that declined to ask for help is the same trait that let a situation travel far past the point where help would still have fixed it. And trusting a process because it is a process cost me more than any decision I made inside it.
I know all of that now. I did not know it then, and I learnt it the expensive way. I will also admit the part that is less flattering: knowing a thing and saying it out loud are different, and I still tend to concede a point some days after I have accepted it privately. That is the honest state of it rather than the tidy version.
None of this is an excuse and none of it asks for anything. It is here because anyone deciding whether to trust the judgement on this site is entitled to know where that judgement has already failed.
I did not prove the theory first. I said it first, in a document anyone can date, and then I built the instruments to test whether it is true.
What I am actually asking
Not that you believe me. Not that you accept the number, or the framework, or the speculation held apart from it. Only that you look, and that you say where it breaks.
People have already found errors here. Their findings are published whether or not the findings suit me, and my own most striking number is on the corrections page because it failed replication and I put it there. If you find the next one, the same applies, and it will carry your name rather than mine. That is the whole invitation. It is open to anyone with the patience to check a citation, and it costs you nothing to accept.
The programme does not depend on me continuing. The data are public under CC BY 4.0, and the papers, experiment code, result files and priority material are public under an MIT licence; if I stop, the work is still there to be finished or falsified by anyone who wants it.
The problem is going to be decided by people who looked at it early. There is still time to be one of them.