For most of history, organised intelligence has lived in castles.
Not always literal castles, although enough of it did. Temples, monasteries, universities, ministries, libraries, consultancies, laboratories, courts, corporations and software platforms have all performed a similar function. They gather knowledge, organise it, protect it, credential it and then decide who gets access.
That scarcity shaped civilisation.
If you wanted legal interpretation, you needed lawyers. If you wanted medical judgement, you needed doctors. If you wanted complex administration, you needed bureaucracy. If you wanted strategy, you needed generals, consultants, civil servants or the sort of person who says “operating model” without immediately apologising.
This was not necessarily a bad thing. Expertise matters. Institutions matter. Some knowledge is hard won, socially useful and dangerous when handled carelessly. I do not want aviation engineering replaced by a confident amateur with a chatbot and a laminated checklist.
But scarcity has always been one of power’s quiet foundations.
Who knows?
Who decides what counts as knowing?
Who gets to ask?
Who gets an answer?
Who is trusted when answers disagree?
AI matters because it interferes with that structure even if it never becomes conscious, never develops desires and never does anything cinematic enough to trouble a Marvel scriptwriter. It threatens the economics of access to organised cognition.
That may prove more disruptive than the usual stories about machine rebellion.
The current AI economy still largely assumes intelligence will remain expensive, centralised and infrastructural. The visible future is enormous: datacentres, specialised chips, energy demand, sovereign compute, foundation models, cloud contracts and a small number of firms positioning themselves as landlords of cognition.
It is not hard to see why Yanis Varoufakis’s techno-feudal argument keeps hovering around this conversation. The concern is not simply that large companies become very large. That story is hardly new. The concern is that access to the conditions of participation becomes mediated through private platforms. You do not own the land. You rent the field. You do not own the marketplace. You sell inside someone else’s walls. You do not own the infrastructure of intelligence. You subscribe to it.
In that version of the future, AI does not liberate cognition. It encloses it.
The terminology is modern. The shape is ancient.
Cloud landlords. Cognitive tenants.
But there is another possibility, and it is the one that makes the current investment story more fragile than it sometimes appears.
What if intelligence does not stay inside the castle?
Technologies often begin centralised because early versions are expensive, fragile and specialist. Then they shrink. They leak. They become ordinary. The mainframe gives way to the personal computer. Broadcast media gives way to messy participatory networks. Professional photography gives way to the smartphone. Publishing escapes the printing press. Music escapes the shelf. Information escapes the institution and promptly causes a thousand new problems, because liberation and wisdom are not the same thing.
The expensive phase can feel permanent while you are living inside it. Later, it looks transitional.
That may be where AI becomes interesting. Not because the giant datacentres are irrelevant. They clearly are not. For now, they are central to the economics and politics of AI. But it is at least possible that today’s industrial-scale AI infrastructure is not the final form of machine cognition. It may be the mainframe era.
Useful, powerful, costly and historically important.
But not the end state.
Smaller models are improving. Local systems are becoming more capable. Open-weight models change expectations. Hardware improves. Compression improves. Specialist systems get better at narrow tasks. People already run models on laptops, home servers and ordinary devices in ways that would have sounded absurd not long ago.
The direction of travel may not be simply bigger, bigger, bigger.
It may also be smaller, closer, cheaper, stranger.
And if capable intelligence becomes local enough, the politics changes.
Centralised AI is frightening, but at least it has an address. It can be contracted with, audited, regulated, blocked, fined or summoned before a committee. There is someone to write to. Someone to blame, even if they reply with a PDF, a dashboard and three pages of carefully managed responsibility.
Distributed intelligence is different.
If powerful cognitive tools live on ordinary devices, in homes, schools, small businesses, community groups, campaign organisations, council officers’ laptops and people’s pockets, then governance becomes much messier. The question is no longer only “what are the big platforms doing?” It becomes “what happens when advanced cognitive assistance becomes part of everyday human agency?”
That is not automatically utopian.
I distrust that sort of optimism almost as much as I distrust easy doom. Cheap cameras did not make everyone Cartier-Bresson. Cheap publishing did not produce universal wisdom. Infinite information did not make politics calmer, kinder or more truthful. In some ways, abundance made judgement harder because it buried signal under performance, noise and outrage.
AI may do the same to cognition.
Cheap answers will not create wise institutions. Cheap analysis will not create legitimacy. Cheap drafting will not create courage. Cheap strategy documents will not create shared purpose. Anyone who has spent time around governance knows that the difficult part is often not producing words. It is deciding what the words mean, who owns the consequences and whether anyone is prepared to act.
That is where this series keeps returning, almost annoyingly, to governance.
Not governance as decorative paperwork. Not governance as “we have a board and therefore we have oversight”. I mean governance as the living discipline of keeping judgement visible when systems are trying to make it disappear.
If AI makes cognition abundant, then the scarce thing may no longer be the ability to produce an answer. The scarce thing may be the ability to know whether the answer should matter.
That is a different economy.
Consultancies monetise organised cognition. Universities credential it. Professions defend it. Governments depend on it. Organisations rank people by it, often imperfectly. Entire careers are built on being able to interpret complexity and turn it into something actionable.
If cognitive assistance becomes cheap and widely distributed, many of those boundaries soften. Not vanish. Soften. Expertise still matters, but expertise will increasingly have to show its working. Authority will have to earn trust in environments where plausible alternative answers are always available.
That could be healthy.
It could also be exhausting.
There is a world where this democratises capability. A small charity gets analytical support it could never have afforded. A resident understands a consultation document that would otherwise exclude them. A junior officer can test assumptions before walking into a meeting. A small business gains access to expertise once locked behind professional fees. A community group can challenge a weak argument with evidence rather than instinct alone.
There is also a world where everyone arrives armed with confident nonsense.
Both futures are possible. More likely, they arrive together.
That is why the “intelligence escapes scarcity” argument needs to be handled carefully. It is not a claim that AI makes everyone equal. Technologies rarely do that by themselves. Power adapts. Gatekeepers return wearing new clothes. Scarcity migrates. When content becomes abundant, attention becomes scarce. When information becomes abundant, trust becomes scarce. When intelligence becomes abundant, legitimacy may become scarce.
This is where local government offers a useful antidote to the grander AI fantasies. Councils already live in the world where intelligence alone is not enough. You can have the analysis and still lack the money. You can have the evidence and still face competing duties. You can have the options appraisal and still need a political decision. You can have the perfect report and still meet the resident who experiences the outcome as loss.
AI may help with complexity. It will not abolish accountability.
Nor should we want it to.
The central question, then, is not whether AI becomes powerful. It is where that power lives.
If it remains concentrated in a few corporate infrastructures, the future bends towards dependence. If it diffuses into ordinary devices, the future bends towards disorder, opportunity and contested authority. If both happen at once, which seems entirely plausible, then we get something even more complicated: centralised platforms trying to enclose a capability that is also leaking into the street.
That feels more historically believable than either utopia or apocalypse.
The printing press did not abolish authority. It destabilised it. The internet did not abolish gatekeepers. It multiplied them, weakened some, created others and changed the terrain beneath everyone. AI may do something similar to expertise.
It may not end the castle.
But it may open too many side doors for the old map to remain useful.
And perhaps that is the real challenge for institutions. Not that intelligence disappears. Not that intelligence becomes divine. But that intelligence becomes common enough to stop functioning as the main source of authority.
When that happens, institutions will have to rediscover what makes them legitimate.
Not just what they know.
What they can be trusted to do with what they know.
Comments
Post a Comment