The Buildings Come Back. The Doctors Don't.

War is remarkably good at destroying things. It is, if one is being honest, the singular activity at which humanity has consistently improved its performance across all recorded history while somehow failing to improve almost anything else at comparable speed. We have dramatically accelerated the rate at which buildings come down. We have not yet managed to accelerate the rate at which surgeons appear.

This asymmetry receives surprisingly little attention, considering that it sits at the heart of why some conflicts leave nations permanently diminished in ways that no Marshall Plan, no UN reconstruction fund, and no amount of international goodwill has ever fully addressed. The rubble is visible. The missing engineers are not.

 

The Casualty List Nobody Publishes

Every conflict produces two casualty lists. The first one is updated daily, cross-referenced against military rosters, and reported with solemnity by news organisations. The second one, which is arguably more consequential for the next fifty years of a nation's existence, is never officially compiled at all.

It contains the following: the children who spent three years not in school but in basements and refugee camps, whose literacy development stalled at the precise age when it is most plastic; the medical students who were conscripted eighteen months before completing their degrees and who may or may not return to finish; the teachers who fled to countries that had positions available and discovered, reasonably enough, that they preferred safety to principle; the engineers who died doing something other than engineering; and the seventy-year-old professor of structural mechanics who was killed by a missile and took with her thirty years of tacit knowledge that was never written down because she had always meant to get around to writing it down and had not quite managed it before the missile arrived.

The buildings, given sufficient concrete and political will, can be replaced in three to seven years. The professor takes approximately seventy years to produce from scratch, which is an inconvenient timeline by any project management standard.

 

Conscription and the Learning Window

There is a particular cruelty to conscription that is rarely discussed in strategic terms, which is that it takes its recruits from precisely the demographic cohort that neuroscience has identified as being in a peak period for acquiring complex skills. The brain between eighteen and twenty-five is, in ways that the same brain at forty-five simply is not, exceptionally well-suited to absorbing new technical domains, learning languages, internalising procedural expertise and forming the kind of deep professional networks that later become the connective tissue of an economy.

Drafting this cohort for two or three years does not merely pause their development. It interrupts it at the worst possible moment, then returns them to civilian life having spent that window on activities that do not transfer to software engineering or cardiovascular surgery in any useful way. Some adapt rapidly. Many do not. A significant proportion carry trauma that reduces their cognitive capacity for years afterward, because chronic stress is not merely an emotional inconvenience: it measurably impairs working memory, reduces prefrontal cortex activity, and degrades exactly the higher-order thinking that complex technical work requires.

A country that conscripts a large fraction of its eighteen-to-twenty-five-year-olds for three years does not lose three years of their development. It may lose considerably more.

 

The Cascade

What makes this genuinely alarming as a systemic problem, rather than merely sad as a human one, is the cascade effect.

Fewer medical graduates means longer waiting times and higher mortality from treatable conditions. Higher mortality and untreated chronic illness in the working-age population means reduced cognitive and physical capacity across the workforce. Reduced workforce capacity means slower economic recovery. Slower economic recovery means lower public investment in education. Lower investment in education means fewer medical graduates in the next cycle. Observe the direction of travel.

This is not a hypothetical. Post-conflict societies reliably enter exactly this spiral. The countries that escape it tend to do so either through extraordinary external subsidy (South Korea, post-war Germany and Japan), through exceptionally robust pre-conflict institutional foundations that survived the disruption (rare), or through some combination of luck and geography that kept enough of the educated class intact. The countries that do not escape it tend to stay not-escaped for a very long time.

 

The Traditional Recovery Model Is Too Slow

The conventional approach to post-conflict knowledge reconstruction is, charitably speaking, a very long walk in a sensible direction. You rebuild the schools. You train the teachers. You establish the universities. You run the programs. You wait twenty years and see whether the investment has produced enough engineers to build the things that need building. It works, eventually, with enough external patience and funding, neither of which is currently in abundant supply.

The problem is that the world these recovering nations are attempting to rejoin has not waited for them. The technical gap between a nation that has had a functioning, well-funded educational and industrial infrastructure for twenty uninterrupted years and a nation that has spent those twenty years recovering from conflict is not a gap that closes easily or quickly, even once the recovery begins. The finish line keeps moving.

 

What AI Actually Changes

Here is where the argument gets interesting, and where I suspect the conventional policy conversation is not yet keeping pace with the technological reality.

The value of AI in post-conflict reconstruction is not primarily about automation replacing workers, which is the frame most people reach for and which is not especially useful in this context. It is about cognitive load.

Cognitive load is the amount of mental effort required to perform a given task. Expert performance in almost any field depends critically on the ability to offload large amounts of routine cognitive work onto trained intuition and procedural memory, freeing conscious attention for the genuinely complex decisions. A skilled surgeon is not consciously thinking about every routine step of a procedure: those elements run on mental autopilot, leaving their full attention available for the moments that actually require judgment. A junior practitioner, by contrast, must consciously attend to the routine steps, which means they have less capacity left for the hard parts, which is why they are slower, less reliable, and why we do not let them operate unsupervised.

AI systems, deployed appropriately, can take over a substantial fraction of the routine cognitive load in almost any knowledge domain. This is not a futuristic proposition. It is what these systems already do. A doctor working with good diagnostic AI support is not thereby made redundant: they are freed from the cognitive overhead of differential diagnosis on routine presentations, which allows them to handle more complex cases with greater accuracy. An engineer with AI-assisted structural analysis tools is not replaced: they are amplified, able to verify designs and catch failure modes that would previously have required either extensive experience or extensive time. A teacher with an AI tutoring system running alongside their instruction is not supplanted: the AI handles the individualised repetition and gap-filling that no single teacher can provide for thirty students simultaneously, freeing the teacher for the relationship, the motivation, and the genuinely creative parts of education.

 

 

The Learning Compression Question

The more provocative version of this argument concerns education directly: whether AI-supported learning can compress the time required to produce a competent practitioner in high-value domains.

The honest answer is that we do not yet know with certainty, because the systems are new and the evidence is still accumulating. What we do know is suggestive. AI tutoring systems consistently outperform average human tutoring on measurable knowledge acquisition metrics. Personalised adaptive learning, which AI enables at scale that was previously impossible, addresses the single biggest inefficiency in traditional education: the fact that a class of thirty students receives instruction calibrated to an imaginary average student who does not exist, which means the system is simultaneously too fast for some and too slow for others, and optimally paced for almost no one.

For a nation rebuilding its educational system, the AI-supported model offers something the traditional model cannot: the ability to deliver effectively personalised instruction at national scale without the ratio of teachers to students that personalisation would otherwise require. You do not need a rebuilt teacher-training pipeline of the same depth if each teacher is supported by systems that handle the individualised elements. You still need the teachers. You need fewer of them to achieve the same outcome, and the outcome may be better.

This matters enormously in the specific context of post-conflict reconstruction, where the teacher shortage is not an abstraction but a very concrete reality caused by the same forces that created the problem in the first place.

 

The National-Level Question

None of this happens automatically or cheaply. The deployment of AI at national scale, in ways that meaningfully address post-conflict knowledge deficits, requires decisions at the level of national policy: which systems to deploy, how to integrate them with existing educational and medical infrastructure, how to ensure that people in rural areas without reliable connectivity benefit rather than watching the advantage accumulate only in cities, and how to avoid the deeply plausible failure mode in which AI-assisted competence becomes a substitute for rebuilding genuine expertise rather than a bridge toward it.

That last point bears emphasis. The goal of AI deployment in this context should be explicitly framed as accelerating the restoration of human expertise, not as a permanent replacement for it. A country that finds it cheaper to run AI diagnostic tools indefinitely rather than train doctors is making a choice that will compound badly when those AI systems fail, are unavailable, or encounter the categories of problem they were not designed for. The bridge metaphor is the correct one: you use it to cross the gap, not to live on.

The nations that are currently experiencing or have recently experienced significant conflict are not, by and large, the nations most likely to have the policy bandwidth, technical infrastructure, and institutional trust required to deploy these systems well. This is exactly backwards from where the intervention is needed. The international community has frameworks for post-conflict reconstruction of physical infrastructure. It does not yet have anything comparable for post-conflict reconstruction of cognitive infrastructure using tools that did not exist the last time anyone thought carefully about this problem.

That gap is worth closing. The buildings will come back. The doctors take considerably longer. We now have tools that can help with the wait, if anyone is prepared to deploy them with that specific purpose in mind.

The universe, as usual, has offered no opinion on whether this will happen. It is busy with other things.

All Posts