When Complexity Outruns Government
The problem with an institution falling behind would be considerably less serious if the world around it remained simple.
It has done precisely the opposite.
During the same half-century in which information technology transformed business, societies became interconnected in ways that would have been difficult to imagine when I entered IT. Production chains crossed continents. Finance became global and instantaneous. Energy systems became intertwined with geopolitics, environmental policy and industrial strategy. Migration connected conflicts in one part of the world with housing, education and politics in another. Healthcare became more technically capable and at the same time more expensive and organisationally complicated. Populations aged. Supply chains became wonderfully efficient and therefore sometimes wonderfully fragile. Social media collapsed the distance between an event, an interpretation of that event and the anger generated by the interpretation.
Every part began interacting with every other part.
Climate policy influences energy prices; energy prices influence industrial competitiveness; competitiveness influences employment and tax revenue; tax revenue constrains welfare spending; welfare policy interacts with migration and demographics; migration affects housing, schools and politics; political instability complicates long-term climate and industrial policy again.
There is no obvious place where one problem ends and another begins.
We eventually invented a fashionable word for this: polycrisis. The term is sometimes abused to make ordinary difficulty sound apocalyptic, but the underlying idea is useful. The European Commission’s Joint Research Centre describes a polycrisis as interconnected crises whose interactions create cascading effects across different domains, while academic work on systemic risks similarly stresses that the danger lies not simply in several bad things happening simultaneously but in the feedback between them.
This matters because most of our institutions were not designed for that world.
Government grew around ministries, departments, competences, budgets and laws that divide reality into manageable pieces. Health belongs here, energy there, education somewhere else. Administrative division is unavoidable; somebody has to be responsible for something. But the problems stopped respecting those divisions a long time ago.
A housing problem is simultaneously a planning problem, a construction problem, a financing problem, an environmental problem, an infrastructure problem, a migration problem, a demographic problem and eventually an electoral problem. Push hard on one part and another moves.
Complex systems respond.
That sounds obvious, yet much policymaking still begins by isolating one desired outcome and trying to control the variables believed to produce it. When reality fails to comply, governments frequently respond by specifying more variables, introducing more rules, adding more reporting and making the model more detailed.
Complexity outside the institution generates complexity inside it.
This is something I explored in The Age of Procedural Inflation: modern states often respond to a world becoming harder to understand by producing ever more detailed rules for it. The intention is greater control. The result can be the opposite, because every rule interacts with thousands of circumstances its author could not possibly foresee. The Age of Procedural Inflation
The European Commission itself now uses the language of polycrisis and systemic interactions because conventional policymaking struggles with problems that cascade from one domain into another. The OECD similarly warns that multiple, overlapping crises stretch governments’ ability to respond to immediate emergencies while maintaining coherent long-term policy.
The difficulty is not merely that there are more problems. It is that traditional solutions can themselves become causes.
A subsidy changes behaviour. A regulation creates an incentive nobody anticipated. A tax designed to solve one problem shifts activity somewhere else. An environmental rule collides with housing policy. Protection for one group creates scarcity for another. Institutions then respond to those effects with another intervention, which creates another set of interactions.
After several decades the accumulated structure becomes extraordinarily difficult to change because almost every piece now protects somebody, finances something, fulfils a legal promise or compensates for an earlier intervention.
We don’t merely inherit old laws.
We inherit layers of attempted solutions interacting with layers of newer attempted solutions.
This is where institutional technological lag stops being amusing.
A relatively simple society can perhaps be governed through slow feedback. If a policy is introduced today and evaluated five years later, the delay is irritating but potentially manageable when the surrounding system changes slowly.
In a complex adaptive society, five years can mean evaluating a world that no longer exists.
The information problem we encountered in Part I therefore arrives at exactly the wrong historical moment. The environment becomes faster, more interconnected and less predictable while many of the institutions responsible for governing it continue to depend on slow information flows, fragmented data, long political cycles and procedural systems that become more complicated each time reality refuses to behave as expected.
The result is not necessarily spectacular failure.
More often it is incomplete resolution.
The housing problem remains.
Education is reformed again.
Healthcare gets another financing structure.
A migration reform modifies the previous migration reform.
Infrastructure projects spend years navigating procedures created to prevent yesterday’s mistakes.
Regulation becomes an archaeological record of every problem legislators once attempted to solve.
None of these systems completely collapses, yet fewer problems seem genuinely to disappear.
This creates a peculiar political experience. Citizens hear constantly that enormous amounts of policy activity are taking place while continuing to encounter the same problems. Governments publish strategies, budgets increase, legislation expands, consultations are held and administrative structures multiply, yet the lived outcome can remain strangely stubborn.
Activity and effectiveness begin to drift apart.
This is another expression of Babel. Once an abstract model acquires political authority, it no longer merely describes reality; it begins instructing reality how it ought to behave. I explored that transition in Babel Hit the Power Button, because it is the point at which abstraction becomes particularly dangerous. The Babel Syndrome — Part 4: Babel Hit the Power Button
And now AI enters this already complex system.
This is why I am less interested in whether a ministry buys Copilot than in whether the ministry can learn.
An organisation with rich operational information and rapid feedback can use AI to explore complexity, detect patterns and test alternatives. It does not eliminate uncertainty, but it increases the institution’s ability to navigate it.
An organisation built around fragmented information and slow feedback may use the same AI to generate more analysis without reducing uncertainty at all.
Worse, it may acquire the illusion that it understands the system because the machine can produce remarkably convincing explanations of it.
That is the nightmare hidden inside the promise of artificial intelligence: beautiful reasoning resting on an impoverished representation of reality.
And because modern problems interact, mistakes no longer remain neatly inside their administrative boxes. A poor energy decision affects industry. Industrial decline affects tax revenue and employment. Economic insecurity affects politics. Political instability changes investment decisions. Those changes affect the original energy transition.
The loop closes.
We call the result polycrisis because there is no longer an obvious single crisis to solve.
There is a system behaving badly.
At exactly this moment, the private and public worlds may begin to diverge much faster than before. Companies with excellent information systems can use AI to squeeze inefficiency from processes, reorganise work, personalise services and adapt products. Individuals can increasingly purchase intelligence directly: financial advice, education, translation, legal assistance, health information, planning and administration arrive through systems whose capabilities may improve several times within a year.
Government doesn’t have to deteriorate absolutely for this to become politically dangerous.
It need only improve more slowly than everything against which citizens compare it.
That relative decline can feel very much like actual decline. A person who becomes accustomed to systems that understand a complex request immediately becomes less tolerant of an administration that asks for information it already possesses. Parents watching an AI tutor adapt continuously to their child may become less patient with an educational system that needs years to revise a curriculum. Businesses reorganising around AI may become increasingly frustrated with regulators whose procedures assume the industrial structures of an earlier decade.
The expectations gap expands.
And initially, people grumble.
Then they route around the problem.
People with money buy private solutions. They purchase tutoring when schools cannot provide what a child needs. They buy faster healthcare where it is available. Businesses employ lawyers and consultants to navigate regulation. Wealth allows people to purchase the interface between themselves and dysfunctional institutions.
The people least able to escape become the most dependent on the slow system.
At that point the technological divide has become something much more serious: an institutional class divide.
Those living inside the fast society experience increasing personal capability. Those dependent on the slow society encounter waiting lists, complex procedures and public systems struggling with problems that appear never to disappear.
And slowly something else changes.
Trust.
The OECD’s 2026 trust survey provides a useful warning. Only 40 percent of respondents across participating OECD countries reported high or moderately high trust in their national government, compared with 43 percent reporting low or no trust. Yet the interesting result is not simply that trust is low. People remain reasonably satisfied with many everyday administrative services. What worries them much more is whether government can make good decisions about complex, long-term problems, use evidence effectively and allow citizens a meaningful voice.
That distinction fits this argument almost uncomfortably well.
The passport may arrive.
The larger problems remain.
Citizens do not need to understand systems theory to notice this. They see housing that remains unaffordable, infrastructure projects that take years, education endlessly reformed, energy policy changing direction, migration producing recurring political crises and governments promising administrative simplification that somehow produces another administrative layer.
Nobody comes away from that experience saying the state requires better feedback architecture.
They say something simpler.
They can’t solve anything anymore.
That conclusion may be unfair. Governments confront genuinely difficult trade-offs and many problems do not possess clean solutions. But political consequences don’t wait for a fair academic assessment of why something failed.
When enough people lose confidence that established institutions can deal with the problems surrounding them, they begin looking for people who promise to bypass those institutions.
This is where the story becomes politically dangerous, but also where we must avoid an easy explanation. Support for populist, radical or anti-establishment parties has many causes: immigration, cultural change, inequality, economic insecurity, identity, representation and dissatisfaction with democratic institutions all matter differently across countries. Institutional failure is not some master variable explaining everything.
But government performance and political trust do matter. Research increasingly finds associations between perceptions of poor government quality, dissatisfaction with democratic performance and support for populist alternatives. And once trust falls, governing becomes harder because reform itself requires a degree of confidence that institutions can use additional freedom responsibly.
A vicious circle becomes possible.
Weak institutions struggle with complexity. Problems remain unresolved. Citizens lose trust. Politics fragments. Governments become harder to construct and their time horizons shorten. Long-term reform becomes more difficult. The institutional weaknesses that created dissatisfaction therefore become harder to repair.
The system begins reinforcing its own failure.
The comparison with the 1930s becomes tempting here, but it should remain a warning rather than a prediction. The conditions of interwar Europe were profoundly different: world war, economic collapse, mass unemployment, political violence and fragile young democracies created circumstances that do not map neatly onto ours.
Yet history need not repeat itself for certain mechanisms to remain familiar.
Democracy becomes vulnerable when large numbers of people cease believing that its ordinary institutions can solve ordinary problems. Political opponents then risk becoming representatives not of another legitimate choice but of a failed system. Politicians promising to smash through constraints become attractive precisely because constraints have become associated with paralysis.
And here AI performs its final trick.
It can accelerate both sides.
It strengthens organisations that already possess good information and short feedback loops, increasing the relative performance gap. At the same time it makes political communication, personalised persuasion, synthetic media and endless production of emotionally targeted content dramatically cheaper.
AI can therefore help enlarge the frustration and help amplify the frustration.
No superintelligence needs to hate us.
Nobody has to lose control of the machine.
Every individual step can remain entirely understandable. The company automates because it must compete. The administration follows the law. The politician responds to voters. The citizen chooses a service that works. The dissatisfied voter rejects an established party. The new party promises to dismantle institutions that appear incapable of acting.
Each decision may be rational on its own.
The system can still become increasingly irrational as a whole.
I wrote earlier that technology is accelerating while society is slipping, but AI makes that old tension much sharper. We are no longer merely introducing faster technology into slower institutions. We may be introducing technology that accelerates an organisation’s ability to learn into a society whose governing institutions have already spent decades accumulating information gaps, procedural complexity and institutional debt. Tech Is Accelerating—Society Is Slipping
So perhaps the most important AI question facing government is not whether it has an AI strategy.
It is whether government still possesses the institutional machinery required to understand reality quickly enough to govern it.
Can information reach decision-makers before the world has changed?
Can a failed policy actually be stopped?
Can a procedure disappear rather than merely acquire another exception?
Can an institution learn from the people at its frontline?
Can government reduce complexity rather than continually responding to complexity with more complexity?
Artificial intelligence might help enormously with all of those things. This is not an argument against using AI in government. Quite the opposite. Properly embedded in institutions capable of observation, experimentation and learning, it may become one of the few tools powerful enough to help us deal with the complexity we have created.
But AI cannot repair a broken feedback loop merely by being intelligent.
Nor can it manufacture institutional courage.
Perhaps one day somebody really will use artificial intelligence to create a terrible virus, and perhaps that is the catastrophe future generations will remember. We should work hard to prevent it.
But there is another possibility.
AI may simply accelerate the capable parts of society until the institutions holding everything together can no longer manage the complexity surrounding them. Those institutions then lose effectiveness, followed by credibility, followed eventually by trust.
And history suggests that when societies lose trust in the institutions through which they govern themselves, technology is no longer the most dangerous force in the system.
We are.