The Human Factor AI Cannot Predict- The One Thing Artificial Intelligence Still Doesn’t Understand
Military planners have always wanted certainty. They want to know how many tanks the enemy has, how many soldiers can be mobilized, how much fuel is in reserve, and how long a siege can be sustained before supplies run dry. For decades, this kind of quantifiable data has been the backbone of strategic forecasting. Today, artificial intelligence promises to take that forecasting to an entirely new level, crunching satellite imagery, troop movements, economic indicators, and historical patterns in seconds. Yet for all its computational power, AI keeps stumbling over the same obstacle: the human being.
Wars are not won or lost purely on logistics. They are shaped by pride, fear, grievance, ideology, and the psychological state of the people making decisions. These are not variables that fit neatly into a spreadsheet, and that is precisely why so many AI-assisted predictions about conflicts have missed the mark in recent years.
The Seduction of the Numbers
It is easy to understand why militaries and intelligence agencies have leaned so heavily on data-driven models. Numbers feel objective. A model that says “Country A has three times the armored divisions of Country B” appears to offer a clean, almost mathematical answer to the question of who would win a conflict. Defense analysts have built careers on these comparisons, and AI systems have only made the process faster and more granular.
But history is full of examples where the side with fewer resources refused to behave the way the numbers predicted. Smaller forces have repeatedly outlasted larger ones, not because of superior equipment, but because of a willingness to endure losses that a purely rational actor, as defined by a model, supposedly would not accept. When a population believes it is fighting for its homeland, its identity, or its survival, the cost-benefit calculations that underpin most predictive models simply stop applying in the way they’re supposed to.
National Pride as a Strategic Multiplier
National pride is one of the hardest factors to quantify, yet it consistently changes the outcome of conflicts. A country that feels humiliated, whether by a previous defeat, a territorial loss, or perceived disrespect on the world stage, often behaves differently than its raw capabilities would suggest. Pride can drive a nation to fight long after a model would predict capitulation, and it can also drive escalation that a purely capability-based assessment would never anticipate.
AI systems trained on historical data can certainly identify patterns of nationalist rhetoric, increased media coverage of historical grievances, or shifts in public sentiment. What they struggle with is translating that sentiment into a forecast of behavior. A surge in nationalist messaging might precede a willingness to negotiate, just as easily as it might precede a refusal to back down. The same emotional signal can point in opposite directions depending on context that is deeply rooted in culture, history, and leadership psychology, things that don’t always show up cleanly in the data an algorithm is trained on.
Ideology Doesn’t Compute
Ideology presents a similar challenge. Whether it’s religious conviction, a particular political worldview, or a deeply held belief in a cause, ideology can make actors behave in ways that appear irrational from the outside but are entirely consistent from the inside. A leadership group that views a conflict as existential, tied to the survival of a belief system rather than just territory or resources, may continue fighting even when every conventional indicator suggests the situation is hopeless.
This is where AI models often default to assumptions of rational actor behavior, because that’s the framework most strategic theory has historically been built on. But ideological commitment can override what would otherwise be a rational retreat. A model might calculate that continuing a conflict carries an unacceptable cost in lives, infrastructure, and economic damage, and conclude that a ceasefire is the likely outcome. Yet if the people in charge view surrender as a betrayal of their core beliefs, no amount of cost makes that option viable to them. The model isn’t wrong about the costs; it’s wrong about what those costs mean to the people bearing them.
Fear Changes the Math
Fear is another variable that resists quantification, and it cuts in multiple directions. Fear of losing power can push a leader toward aggressive action they might otherwise avoid, simply because the alternative, being removed from office, exiled, or worse, feels more threatening than the risks of military escalation. Fear of appearing weak in front of domestic audiences or international rivals can cause leaders to double down on commitments that no longer make strategic sense.
At the same time, fear among a civilian population can either fracture a society’s will to fight or, in some cases, harden it. AI models can track indicators like refugee flows, economic disruption, or casualty rates, all things that theoretically should correlate with declining morale. But fear doesn’t always produce the expected response. Sometimes it produces resignation and a desire for any resolution, no matter how unfavorable. Other times, it produces a defiant unity that makes a population more willing to sacrifice, not less. Predicting which way fear will tip a society requires an understanding of culture, history, and collective memory that goes far beyond pattern recognition in datasets.
Leadership Decisions: The Black Box Within the Black Box
Perhaps the most unpredictable element of all is the individual psychology of the people making final decisions. Leaders are shaped by their personal histories, their relationships with advisors, their health, their egos, and their sense of legacy. A leader who feels they have nothing left to lose may take risks that no model would assign a meaningful probability to. A leader surrounded by advisors too afraid to deliver bad news may receive a distorted picture of the battlefield, leading to decisions that look baffling from the outside but make sense once you understand the information environment that leader was actually operating in.
AI can analyze a leader’s past statements, public appearances, and even biometric data if it’s available, looking for signs of stress or instability. But it cannot get inside that leader’s head. It cannot know what conversation happened behind closed doors the night before a major decision, or what personal grievance from decades ago is quietly influencing how a leader interprets a current threat. These are the kinds of details that sometimes only emerge in memoirs and historical records years after the fact, long after the predictions that mattered have already proven wrong.
Why This Matters Beyond the Battlefield
The limitations of AI in predicting human behavior during conflict aren’t just an academic curiosity. As militaries and governments increasingly integrate AI into decision-making processes, from threat assessments to resource allocation, there’s a real risk of over-relying on outputs that look authoritative but are built on assumptions that don’t hold up when emotions take over.
A model that consistently underestimates the role of pride, ideology, fear, and individual leadership psychology will consistently be surprised by outcomes that, in hindsight, seem almost inevitable to anyone who understood the human context. This isn’t a call to abandon data-driven analysis. The numbers still matter, and AI’s ability to process vast amounts of information remains genuinely valuable. But it’s a reminder that those numbers need to be interpreted alongside a deep understanding of the people involved, their history, their culture, and the emotional forces driving their decisions.
The Limits of the Machine
Artificial intelligence excels at finding patterns in data that humans might miss. It can spot trends, flag anomalies, and process information at a scale no human team could match. What it cannot do, at least not yet, is fully grasp what it feels like to believe your nation’s survival is at stake, or to carry the weight of a decision that could define how history remembers you.
Until AI can account for these deeply human elements, and it’s worth questioning whether it ever truly can, the most sophisticated models will continue to be surprised by the same thing that has surprised military strategists for centuries: people don’t always do what the numbers say they should. Sometimes they do something else entirely, driven by feelings that no algorithm has yet learned to measure. Is This the Most Dangerous Time Since 1945? | Maya
