I have been mildly sceptical of Bryan Mercadente’s enthusiasm for artificial intelligence. This has not been because I think the technology unimportant. I agree that its economic possibilities are immense, and that the usual predictions of mass unemployment rest on bad economics. My reservation has been more personal. Bryan belongs to a generation for whom every new machine is presumed innocent until proved otherwise. I belong to one that has watched computers arrive with repeated promises of liberation, only to find that they have often made it easier for banks to refuse us service, for employers to watch their staff, and for the British State to know more about us than the East German authorities could have imagined in their wettest dreams.

I may now owe Bryan an apology. For the past few months, I have been using artificial intelligence to deal with a series of disputes with my local planning department. I will not bore you with the details. Anyone who has dealt with a planning department will know the general shape of them, and anyone who has not should give thanks for his good fortune. What matters is that I have discovered a use for artificial intelligence that may be politically more important than replacing accountants or designing advertisements. It can abolish one of the oldest and most effective defences of the administrative state. It can make the bureaucracy intelligible.

This needs some explanation. The British State does not normally protect itself from its subjects by telling them they have no rights. It is much cleverer than that. We are told that we have rights everywhere. We have rights of consultation and review. We have rights of complaint. We can ask for reasons. We can demand reconsideration. We can go to ombudsmen, inspectors, tribunals and commissioners. We can ask for information. We can challenge procedural defects. We can complain about failures to follow published policy. We can complain about failures to follow the complaints procedure itself. The difficulty is finding out how.

There is nothing accidental about this. The old bureaucratic state relied on commands. The modern administrative state relies increasingly on complexity. It does not say No. It gives you a hyperlink. This takes you to a page containing seventeen other hyperlinks, one of which leads to a forty-eight-page document written in that peculiar dialect of English in which every individual word is familiar but whole paragraphs can pass through the mind without leaving a trace. Somewhere inside this document is an explanation of what you are entitled to demand, subject to exceptions defined elsewhere, together with a time limit you may not know has already begun.

If you telephone, you may spend half an hour listening to recorded assurances that your call is important. If you write, you receive an automatic reply directing you back to the website. If you make the wrong kind of complaint, it will be rejected because you should have used another procedure. If you use the correct procedure but omit some formula, the omission may become the only part of your complaint that receives close attention. The British State does not deny justice. It buries justice alive under procedure.

Until recently, there were two ways through this. The first was money. You could employ a solicitor, planning consultant, tax adviser or some other member of the vast intermediary caste that has grown up between the citizen and the State. These people are often useful. Many are very good at what they do. But their existence is also evidence of institutional failure. A legal and administrative order intelligible only to those who charge hundreds of pounds an hour to explain it is not government by law in any meaningful sense. It is government by licensed interpreters.

The second way was obsession. Every local authority in England has encountered the man who has spent fifteen years fighting about a boundary fence. He has six filing cabinets in his spare bedroom. He knows the relevant legislation better than the junior officials dealing with him. He can quote paragraph 7.4.3 of a circular withdrawn in 2011, and has the archived copy to hand. He knows the name of every officer who has touched his case since 2009 and can produce their contradictory statements in chronological order. His wife has probably left him. His children no longer telephone. He has become very formidable. He has also gone mad.

This was one of the hidden safeguards of administrative government. You could beat the system, but doing so required either money or the sacrifice of your sanity. Rights therefore existed formally while remaining expensive to exercise. The administrative state could boast about its elaborate machinery of redress in the comfortable knowledge that most people had jobs to do and pleasures more attractive than spending Thursday evening comparing versions of a council enforcement policy. Artificial intelligence has changed the bargain.

I can now give an AI system a planning decision, the correspondence surrounding it, the council’s published policy, the relevant regulations and whatever guidance seems applicable. I can ask it to identify contradictions. I can tell it to separate matters of law from matters of policy. I can ask whether an officer has answered the question put to him or merely produced words in its vicinity. I can ask it to draft a complaint, then tell it to remove the rhetoric and turn every allegation into a numbered proposition supported by a reference. I can ask what the council’s next procedural escape route is likely to be and prepare for it before the reply arrives.

What once required a £300-an-hour professional, or six months of evenings and a developing facial twitch, can increasingly be done after dinner.

The significance extends beyond planning. Consider almost any point of contact between the individual and the administrative state: tax assessments, benefit decisions, school admissions, parking penalties, licensing, employment disputes, Freedom of Information requests. In every case, there is a disparity of resources. The bureaucracy deals with the same questions every day. The citizen may encounter the procedure once in his life. The bureaucracy has templates and databases. The citizen has a letter he barely understands and perhaps fourteen days in which to answer it. This disparity has been one of the foundations of administrative power. Artificial intelligence destroys it.

I therefore note with considerable interest the recent official whining about what has been called “agentic flooding”. The argument is that citizens are discovering how good AI is at dealing with bureaucracy, and that the result may overwhelm the State. Employment tribunal claims are rising. AI can draft objections, appeals and applications. Future autonomous agents may submit them with still less human effort. Procedures created in an age of paper correspondence assumed that only a small minority of citizens would have the patience, knowledge or money to pursue every right available to them. Artificial intelligence abolishes much of the cost. All this is presented as a looming problem. I find it difficult to suppress a smile.

We are told that there may be a “tragedy of the commons”. If everyone insists on exercising the rights Parliament has given him, the machinery established for exercising those rights may collapse. The solution proposed is partly to simplify the procedures—which is sensible—but also to restrict rights, impose charges and discourage supposedly excessive claims.

There is something magnificent about the admission. For decades, the British governing class has constructed an administrative state of astonishing intricacy. Every political demand produced another right, every right another procedure, every procedure another body of officials. This was justified as openness and accountability. We were assured that the State was becoming more responsive because we could complain to an ever-expanding collection of bodies bearing names like the Office of Something Regulation and the Independent Commissioner for Something Else. It now turns out that the system worked only because we were not expected to use it.

The right to complain was acceptable while complaining remained sufficiently difficult. The right to appeal was harmless while most people gave up. Transparency was admirable while extracting information required the temperament of a medieval copyist. Due process was an ornament of liberal government so long as those subjected to government lacked the time and expertise to insist upon it. Now the peasants have found lawyers who work for pennies.

This also throws an interesting light on the growing political agitation about artificial intelligence. We are told that governments must regulate it because of children. There are warnings about terrorism, extremism, misinformation and the “far right.” Some of these concerns are genuine. Artificial intelligence is a powerful technology, and powerful technologies can be used for wicked purposes. I have never believed that liberty requires us to pretend obvious dangers do not exist. But we should be suspicious whenever the British governing class discovers a new danger requiring new powers for itself. Its record does not justify trust. The same apparatus that spent decades enlarging censorship while claiming to protect minorities, and that has increasingly converted social and political disagreement into matters for administrative supervision, cannot reasonably expect us to assume that its interest in AI regulation is wholly disinterested. I have argued elsewhere that the modern British system often suppresses dissent less by direct prohibition than by creating professional and institutional penalties around it. The administrative technique is more subtle than the old policeman’s knock. This does not make it less effective. Artificial intelligence threatens this arrangement because it changes the distribution of competence.

That may be its genuinely revolutionary quality. People naturally concentrate on the possibility that AI will become more intelligent than human beings. Politically, the more immediate fact is that it can make ordinary human beings temporarily as competent as specialists. This is an enormous change.

I do not become a planning barrister by asking an AI system about planning law. I would be a fool to assume that I did. AI systems make mistakes. They can misunderstand documents, invent authorities and state uncertain conclusions with irritating confidence. Anything important must be checked against primary sources. A competent professional remains essential where large sums of money or serious legal consequences are involved.

But most administrative encounters never justified the expense of professional advice in the first place. The sums at stake were too small. The inconvenience was large enough to irritate but too small to justify spending £5,000. The State occupied this zone. It could impose ten hours of unpaid work on you with almost no cost to itself. It could send a letter requiring an answer within fourteen days, while taking eight weeks to answer yours. It could make a decision worth £600 against you in the knowledge that challenging it professionally might cost several times that amount. The administrative state has lived by economies of scale. AI gives economies of scale to its victims.

There is an historical parallel. Before printing, literacy did not merely involve knowing how to read. Books themselves were scarce and expensive. Knowledge was therefore naturally concentrated in institutions able to maintain libraries and educated specialists. Printing did not make every European a scholar. What it did was destroy the monopoly conditions under which scholarship had operated. A moderately prosperous tradesman could eventually possess books that Charlemagne could not have assembled at any price. The consequences included religious upheaval and an enormous redistribution of intellectual authority.

The Internet began another such redistribution, though its promise was compromised by the rise of giant platforms and by the sheer quantity of rubbish it made available. Artificial intelligence may complete what the Internet began. The Internet gave us access to the documents. AI helps us understand them.

This is particularly dangerous to the British State because so much of its authority depends on mystification. Our governing class likes to imagine itself a meritocracy. Its members have degrees and job titles. They speak in acronyms. They produce documents full of boxes, bullet points and numbered objectives. Their power rests partly on the belief that administration is a body of occult knowledge unavailable to ordinary mortals. Often it is nothing of the kind. Much bureaucratic expertise consists of knowing where things are written down, understanding a specialised vocabulary and remembering which procedural button must be pressed next. These are exactly the activities at which artificial intelligence is becoming exceptionally useful. Strip away the informational advantage and the supposed expert is sometimes revealed as a clerk with institutional backing.

This explains why AI is potentially more subversive than decades of anti-government rhetoric. A pamphlet can tell a man that he is oppressed by bureaucracy. AI can tell him which form to send on Monday morning. The difference is considerable.

There is a further irony. The State will certainly adopt artificial intelligence itself. Indeed, it must. If millions of citizens begin submitting professionally structured claims, objections and appeals, there will be no possibility of answering them with existing numbers of human officials. AI systems will read AI-generated submissions and produce AI-assisted responses. We may end with machines disputing with machines while the citizen and the civil servant drink tea.

I have no objection to this. If an application can be decided automatically by reference to clear rules, let it be decided automatically. If a planning proposal complies with an intelligible zoning code, there is much to be said for allowing a computer to issue permission instantly rather than requiring months of consultation with people whose professional interest lies in keeping the procedure alive.

But notice where this leads. Once administration is reduced to rules that machines can apply, a question presently obscured by procedure becomes unavoidable: why are there so many rules? AI may therefore do more than make the existing State efficient. It may expose how much of the existing State serves no useful purpose.

This is where my optimism becomes qualified. The British governing class will not watch its informational monopoly disappear with indifference. Institutions defend themselves. I have described elsewhere how our system of power does not normally operate through explicit orders from a single centre, but through a governing class that absorbs common assumptions and acts upon them by something resembling osmosis. There need be no conspiracy to suppress politically inconvenient uses of artificial intelligence. There need only be a common recognition among regulators, politicians, large technology companies and approved experts that certain capabilities are “unsafe”.

We should watch that word. Unsafe is becoming what seditious was to an eighteenth-century government. It is wonderfully elastic. A system that tells a schoolboy how to make explosives is unsafe. Very well. Is a system that helps a householder find defects in a planning decision unsafe? Is one that drafts an employment tribunal claim unsafe? What about one that analyses whether a police officer has correctly stated his powers? What about one that searches ten thousand pages of government publications and produces a documented argument that a minister has lied? At what point does “safety” become the preservation of official inconvenience?

I do not claim that every proposal for AI regulation is fraudulent. I do claim that every proposal should be examined on the assumption that those seeking regulatory power have interests of their own. Child protection will be invoked because no politician wishes to be accused of opposing it. Terrorism will be invoked because it always is. Extremism and the far right will follow. “Misinformation” will appear wherever inconvenient factual claims need discouragement. The labels vary. The desired conclusion is remarkably stable: approved institutions should retain capacities that ordinary people cannot safely be trusted to possess. This is an old argument wearing a silicon mask.

I have spent years describing the British political order as a managerial oligarchy whose power rests less on naked coercion than on institutional privilege and the fragmentation of opposition. I remain sceptical of claims that technology by itself can overturn this order. Railways strengthened states before they strengthened rebels. Radio became An instrument of centralised propaganda. Computers have made possible a degree of financial surveillance that would have astonished the rulers of the Soviet Union. AI can be used against us. It almost certainly will be.

But this time there is an important difference. The useful forms of artificial intelligence are already in millions of hands. They are cheap. Increasingly, they can be run on ordinary computers. They can read the State’s own publications and use the State’s own rules against it. Their political importance does not depend on their becoming conscious, or superhuman, or capable of replacing the entire labour force. They need only become reliable enough to give ordinary people a competence that was previously monopolised by expensive professionals and dedicated obsessives. That threshold is already being crossed.

For most of my life, one of the unwritten principles of British government has been that ignorance of the law is no excuse, while knowledge of the law is prohibitively expensive. Artificial intelligence may be bringing that agreeable arrangement to an end. I now understand Bryan’s excitement rather better.

In closing, here is a metaphor he will surely appreciate. The governing class spent half a century building a labyrinth, confident that we should become exhausted before reaching its centre. It is now terrified because someone has handed us the thread.