The Great-Man Theory Refutation

In the [YOU] on AI Field Guide

Every summit on AI safety, every regulatory hearing, every founder pledge operates on the assumption that the right people, correctly persuaded, can redirect the trajectory of the technology. Tolstoy’s refutation names this assumption directly: it is the same assumption that credited Napoleon with the burning of Moscow, and it is wrong for the same reason. The actual trajectory of AI is structural: economics that reward scale, competitive dynamics that punish unilateral restraint, the diffusion of capability across millions of actors none of whom alone constitute the steering force. Addressing the technology by persuading the prominent figures at its surface is addressing the label, not the chemistry.

The cycle’s companion volume, [YOU] on AI, asks what all of this means for the individual standing inside the transformation. The Great-Man Theory Refutation restores to that individual a form of agency that the great-man narrative takes away: you are not a spectator waiting for titans to decide. You are one of the millions whose summed choices constitute the system’s actual motion. The small choices are real forces. The responsibility is distributed, not absent.

Origin

The argument appears first in the battle scenes of War and Peace, where Kutuzov’s greatness consists in not issuing confident orders into chaos. Napoleon dictates dispositions that describe situations already dissolved; his orders arrive too late or cannot be carried out; his plan survives contact with the field only as a story told afterward and projected backward onto events that had already been decided by forces he did not see. The contrast is not between two commanders but between two theories of what command can do.

Tolstoy then makes the argument explicit in the second epilogue, dropping fiction entirely to write historical philosophy. Why did millions of Frenchmen march east to die in Russia? Napoleon’s ambition is insufficient: a single desire cannot move millions of bodies. Each soldier had his own reasons, fears, and calculations; the war was their resultant. Tolstoy’s term for the account that credits Napoleon is “superstition”—not a polite criticism but a precise one: the attribution of an outcome to a single cause when the actual cause is distributed and invisible.

The argument carries a structural corollary that Tolstoy presses: the more power a figure appears to hold, the less free he actually is. The emperor is the node through which the largest number of forces must pass, and so he can move only as those forces permit. The person who looks most like a sovereign agent is the most constrained. Applied to the founders and policymakers of the AI moment, each of whom insists he has no choice—that if he slows down a competitor will not—this is Tolstoy’s law of inverse freedom, recognized without attribution.

Key Ideas

The Battle No One Commands. Tolstoy’s foundational demonstration: a system of sufficient scale and complexity produces outcomes that no participant intended or controlled, and the commander’s belief that he is directing it is the primary cognitive hazard. Applied to large language models: the researchers who assemble a training process discover what capabilities have emerged rather than specifying them in advance; the gap between intention and result is exactly Tolstoy’s gap between the order and the battle.

The Inverse Freedom Law. Visible power correlates with actual constraint. The more forces that must pass through a node, the more its motion is determined by those forces rather than by its own will. The founder who says he cannot stop because competitors will not is not describing market reality alone; he is demonstrating Tolstoy’s law. This is why “steering” is the wrong metaphor: it assumes a wheel connected to a rudder, and Tolstoy’s analysis is that the rudder is not attached.

Agency at the Field Level. If distributed forces move the system, distributed actors have distributed agency. Tolstoy did not conclude that individuals are powerless; he concluded that power is in the field rather than at the summit. The researcher who chooses what to publish, the engineer who refuses to build, the user who declines to accept, the citizen who demands accountability—each is a real force, small in isolation and large in aggregate. This is where Pearl’s mathematical precision about what AI cannot do and Tolstoy’s historical philosophy about what leaders cannot control converge: both locate the honest locus of agency below the surface where it is advertised.

The Historian’s Superstition. The practice of attributing large outcomes to individual causes is, for Tolstoy, a cognitive error with a psychological explanation: a world of great authors is a legible world, one in which praise and blame have clear addresses. The superstition survives because it serves a need. Technology journalism, which organizes every transformative moment around the decisions of a few named individuals, is the contemporary form of the error.