AI-Proof: What Machines Can Never Be — Volume I: Awakening, Part II: The Machine at the Gate

Where we are: after Part I, the start of Part II. From the person to the machine: what it is, what it replaces, and how far to trust it.

← Previous: Chapter 4: Why This Matters Now · Series hub · Next: Chapter 5: When Intelligence Became Cheap →

A machine read an X-ray in four seconds and found what a radiologist had missed. Nobody was fired. Part II asks what the machine is, what it replaces, what it cannot, and how far to believe it, with a date on every claim.

In Brief: For readers who want to know what the machine at their work is, what it will replace and how far to believe it. Part II turns from the person to the machine. Its four chapters follow the falling price of trained thinking, the five forms replacement takes, the claims that a machine cannot do something, and the ways a language model fails. The print edition pictured a line the machine would not cross. This edition finds no fixed line, only dated claims and positions that people must keep full.

This edition keeps the gate and gives up the wall.

Opening: The Nodule Behind the Rib

The machine read the chest X-ray in four seconds. Dr Sanjay Rao, a radiologist in Mumbai, had read the same image the evening before, at the end of a long shift. He had missed a small nodule half hidden behind a rib, and the machine found it without effort.

Nobody was fired. The change came more quietly. Three of the hospital’s radiologists will retire in the next few years, and their posts will not be filled.

Part II turns from the person to the machine. Its title pictures the machine outside, at a gate it might one day pass. The four chapters find it already inside, doing tasks, while the jobs around those tasks change shape.

The question this part asks

What is the machine, what does it replace, what can it not replace, and how far should you believe it?

Part I asked what is yours. Part II asks the same question from the other side. Here every claim about what a machine can do carries a year. The Prologue compared this to an honest watch, which does not promise that water can never reach it. It states a depth.

The chapters

Chapter 5, “When Intelligence Became Cheap”, follows the falling price of trained thinking. It sets Adam Smith’s water and diamonds beside Nārada, who knew every science of his age and was still in grief. It keeps four words apart that panic runs together: cost, price, usefulness and wage. A person’s worth is none of them. What stays scarce is choosing what thinking is for and answering for what it does, which carry no date. Checking stays scarce too, for as long as it stays hard. Some of the gain goes to the owners of the machines. Whether anyone is paid for what stays scarce is still being decided.

Chapter 6, “What AI Will Replace”, names five forms of replacement. They run from the post that is never advertised to the reader who never clicks through. It follows Sanjay in Mumbai, Eleanor Marsh, a junior solicitor in London, and Rosario dela Cruz, a copywriter in Makati. Marx’s worker who “steps to the side” and the slow arrival of the electric dynamo help it read the confused years in between. Its conclusion is honest about loss. When work serves another, the better performer should do it. When the work was also forming the worker, its loss is real even if the machine does it better. None of the three has reached the other side.

Chapter 7, “What AI Cannot Replace”, tests the print edition’s strongest claims of the kind “a machine cannot”. It puts each against the evidence of 2026, from judgement under uncertainty and taste to skilled touch and the premium people pay for a person. Each gets a date, and some have already been matched. What cannot be replaced is a position, not a skill: the one whose life it is, the one who answers, the one with the right to decide. A position can be emptied while every name stays on the form. Rajan Subramanian, in a mill in Tiruppur, asks for a wash test rather than overruling the instruments, and ends the chapter by writing that rule down.

Chapter 8, “Understanding the Machine You’re Working With”, explains a language model in as few technical words as possible. Three facts matter, as of 2026. It stops learning when its training ends. Its account of its own reasoning is not reliable evidence of how it reached an answer. It writes fluent, specific text whether or not the knowledge is there. Believing a machine, the chapter argues, is a case of testimony, and the Nyāya account of the reliable speaker helps decide it. Believe it as far as you can check it, or as far as someone stands behind it. A grandmother’s recipe notebook in Tallinn carries the argument, and so does a study, cited by a machine for Kavitha Sharma’s lesson, that does not exist.

What joins them

The four chapters move from price to replacement, then to limits and to trust. Every claim about the machine has a date on it. What has none is the position: someone whose life it is, someone who explains and repairs, someone with the right to decide.

Part II closes on a finding that changes what Part I meant. Answering is not only a quality some people have. It is a relation that has to be kept full, and courts and tribunals have begun, case by case, to say who must fill it. Where nobody keeps it full, the person who should answer is simply missing.

How this edition reads the title

The print edition’s Part page promised to show what the machine “Cannot Touch”. This edition keeps the gate and gives up the wall. There is no line the machine cannot cross that holds for all time. There are claims with dates, and positions that people keep full or let go empty.

My reading of the title is this. A gate has a keeper. It is where people decide what to let through, and for what task. The Prologue allows a machine to brake a car or stop a stolen card, within limits people can justify and answer for. Read this way, the machine at the gate is a machine someone let in. The Prologue’s second never then applies: its makers and users “will never be free of what it does”.

What would prove this wrong?

Emerging. Part II rests on one checkable claim of mine. Replacement takes tasks before jobs, and arrives first as junior posts that are never offered, while experienced people mostly keep theirs. Chapter 4 reads the evidence that way as of 2026, and Chapter 6 states it as its conclusion.

The claim would be wrong if, in the jobs most exposed to these machines, experienced workers began losing their posts faster than young people were kept out. It would also be wrong if young people’s hiring into those jobs recovered while the machines kept improving. The place to look is hiring and employment data by age and occupation, which the chapters cite with their dates.

Ending: A Date on Every Recipe

Part II ends in a kitchen in Tallinn. Marten, who turned his grandmother’s handwritten recipes into an app, asks her which version of a dessert is right, his app’s or the one she has just made. She tells him both are, on their days.

That night he adds a date to every recipe in the app. Under the dessert he types a short note that sends the reader to her. It gives no method. It names the person who stands behind the recipe.

Words to keep

• position: a place no skill can fill: the one whose life it is, the one who answers, the one with the right to decide. The three are often held by different people.

• testimony: knowledge that comes from being told, rather than from seeing or working it out yourself.

• Nyāya: the Indian school of logic and of the ways of coming to know.

Named people in these chapters are composites, and their conversations and circumstances are constructed; the studies, laws and figures cited are real and sourced.

Revised from the print edition (2026)

• What the print edition promised: Part II’s own page promised to show “What Artificial Intelligence Actually Is, What It Changes, and What It Cannot Touch”; its contents page said “What AI Is, What It Replaces, and Where It Stops”. Its chapters drew that line with consciousness, and it closed on the machine being “unaware”: “That single fact is the foundation of everything you will build from here.”

• What I now hold: there is no fixed line where the machine stops. Every claim about what it can do carries a year. What it cannot replace is a position, not a skill, and a position stays full only while people and institutions keep it full. Believing a machine is a case of testimony. This overview is new in the online edition.

Next in the series: Chapter 5: When Intelligence Became Cheap.