When Complexity Outruns Government
In Part I, I explored how fifty years of information technology created a growing distance between organisations that learned to connect themselves to operational reality and institutions that largely computerised their existing procedures. AI has arrived on top of those very different foundations, potentially accelerating the organisations already capable of learning while leaving others struggling to keep up. That would be worrying enough if the world around them had remained relatively simple. Unfortunately, it has done precisely the opposite.
During those same decades, society became an extraordinarily interconnected system. Global production chains, financial markets, energy networks, migration, healthcare, demographics, environmental policy and digital communication now interact in ways that make it increasingly difficult to isolate one problem from another. An energy decision affects industrial competitiveness, which influences employment and public finances, which in turn constrain social policy and shape politics. We have acquired a fashionable name for this entanglement: polycrisis. The term can be overused, but it captures something important. Problems no longer merely coexist; they increasingly reinforce one another, and solutions in one domain can produce unexpected difficulties elsewhere.
Yet most public institutions still divide reality into ministries, departments, competences and budgets, each responsible for its own carefully defined piece of the puzzle. When those pieces refuse to behave independently, the familiar response is to add coordination, procedures, legislation and reporting. Complexity outside the institution produces more complexity inside it, until the accumulated layers become so entangled that removing even an obviously obsolete rule threatens something else built upon it. This is the Babel Syndrome in another form: institutions increasingly govern through abstractions while losing the feedback from reality that would allow those abstractions to be corrected. Eventually, enormous administrative activity can coexist with a remarkable inability to resolve the problems it was created to address.
This would be troublesome in any period, but AI may make the mismatch much more visible. A private organisation with good information systems and short feedback loops can use AI to analyse problems, experiment and adapt more quickly. Public institutions may acquire exactly the same technology yet remain trapped inside structures where even recognising that a policy has failed takes years. Worse, AI can make those institutions spectacularly better at producing the reports, analyses and strategies that sustain the illusion of progress. We may end up with administrations that explain their failures more eloquently than ever while becoming no better at preventing them.
Meanwhile, citizens experience an entirely different technological world. Their bank responds instantly, their AI assistant solves complicated questions in seconds, businesses increasingly anticipate their needs, and services continuously improve. Government need not become objectively worse for this comparison to become painful. If everything surrounding a public institution improves much faster than the institution itself, it will increasingly feel incompetent. People with sufficient resources gradually purchase alternatives, whether in education, healthcare or professional assistance with bureaucratic obstacles, while others remain dependent on the slower system. What began as a technological divide becomes a social divide.
And this is where the real danger emerges. Citizens rarely interpret persistent institutional failure as a problem of information architecture or broken feedback loops. They see governments that announce ambitious programmes while familiar problems remain unresolved, politicians who promise reforms that never quite materialise, and public administrations that appear increasingly disconnected from everyday reality. Frustration gradually becomes distrust, and distrust eventually reaches beyond individual politicians to the institutions themselves. The OECD’s research on trust underlines how strongly confidence in government is connected to perceptions of competence, responsiveness and the ability to handle complex problems.
Of course, ineffective administration does not by itself explain the rise of populist or extremist politics. Economic insecurity, migration, cultural tensions and many other forces contribute. But persistent frustration with institutions that seem incapable of solving problems creates fertile ground for movements promising to sweep away the established order. As traditional parties lose support, political fragmentation can make stable government and long-term reform even harder. The resulting paralysis reinforces the dissatisfaction that produced it. There are uncomfortable echoes of the 1930s in this mechanism, even if the historical circumstances are profoundly different. Democracy becomes vulnerable when enough people stop believing that its institutions are capable of governing.
AI adds a final twist. While it accelerates the private world against which public performance is judged, it also makes political persuasion, manufactured outrage and the endless production of emotionally charged content remarkably cheap. The same technology that may widen the institutional gap can help amplify the frustration it generates. No malevolent superintelligence is required, and no individual decision need be particularly irrational. Companies improve their efficiency, administrations follow their procedures, politicians respond to dissatisfied voters and citizens democratically reject parties they no longer trust. Yet together these entirely understandable choices can push society into a self-reinforcing cycle of declining institutional effectiveness and political instability.
None of this is inevitable, and AI itself could become an extraordinary instrument for repairing the damage. But that requires something more fundamental than purchasing new software or announcing another digital strategy. Governments need to rediscover how to observe reality, learn from failure, remove unnecessary complexity and change course while change is still useful. After decades of institutional inertia, that may prove a greater challenge than developing the technology itself.
Perhaps one day AI really will help somebody create a devastating virus. We should certainly take that danger seriously. But there is another catastrophe worth considering, one that unfolds without an explosion or a machine turning against humanity. AI may simply accelerate parts of society beyond the ability of its governing institutions to follow, until those institutions lose effectiveness, credibility and finally public trust. And when people stop believing in the institutions holding their society together, the machines may no longer be the greatest danger.
We may do the rest ourselves.


