Powder and Power: Gunpowder Empires and Violence Monopoly

The Economics of Centralized Violence

In The Origins of Totalitarianism, I traced how violence monopoly concentrates power in state hands, transforming political structures. The gunpowder empires—Ottoman, Safavid, Mughal, Qing—demonstrate this principle with brutal clarity. Before gunpowder, knights maintained independent armed forces, cavalry roamed borderlands, and feudal lords commanded distributed military capacity. Violence was plural.

Gunpowder changed the economics. Cannons require centralized manufacturing. Professional armies replace feudal levies. Artillery demands coordination, supply chains, bureaucratic administration. The technology itself favors concentration. What the four gunpowder empires reveal is not mere innovation but organizational capacity—the ability to mobilize resources, coordinate production, deploy force across vast territories. Technology did not determine this outcome; it enabled a choice that empires with existing infrastructure made fastest. This consolidation, not disruption, reinforced existing power centers rather than overturning them.

Notice the borderlands shift. Before gunpowder, frontier societies possessed energy, hunger, innovation—competitive advantages in mobility warfare. Empires had mass and depth but could not leverage their size. Gunpowder decisively shifted this balance. Borderlands energy without resources lost to empire organization with resources. The geopolitical lesson: certain technologies favor concentration, rendering distributed advantages obsolete.

Hierarchical Architectures and Centralized Judgment

Neural networks mirror this pattern with unsettling precision. Hierarchical feature learning builds abstraction from edges to faces—low-level features combined into high-level concepts through centralized decision-making at deeper layers. AlexNet learned face detectors without explicit programming, constructing compositional hierarchies that concentrate judgment in upper layers. This is not distributed intelligence; it is authority flowing upward.

Regularization techniques impose constraints centrally. Dropout, weight decay, data augmentation—all architectural decisions made at design time, controlling behavior throughout the network. The designers assume central coordination necessary for optimization, for preventing the chaos of overfitting. Yet this assumption deserves scrutiny. Gradient descent requires differentiable smoothness, constraining what structures become thinkable. Evolutionary methods handle discontinuous structures but pay dearly in efficiency, facing the curse of dimensionality. Both optimization approaches face fundamental limitations. Gradient descent cannot optimize fractal architectures; evolution explores structure space but converges slowly, potentially stuck in local optima.

The question becomes: does technology determine structure, or does structure reflect contingent choices masquerading as necessity?

The Fragility of Distributed Alternatives

Biological systems—fungi, slime molds—achieve coordination without centralization. Real evolution diversifies over billions of years into eyes, wings, brains, exploring possibility space far beyond our toy algorithms. These distributed systems suggest that centralization is not inevitable but contingent, a design choice favored by particular constraints.

Yet the gunpowder empires teach a darker lesson. Energy and innovation in borderlands could not compete with centralized mass organization once technology shifted. Violence monopoly enabled stability through concentration. The public realm—where speech and action occur among equals—shrinks when power consolidates. Totalitarianism arises when centralization supplants plurality.

Perhaps hierarchical architectures are totalitarian—excessive concentration masking design choices as optimization requirements. The challenge is not whether we can build centralized systems but whether we must. Can we design distributed alternatives that preserve plurality while achieving coordination? Or does the economics of certain technologies—gunpowder, neural hierarchies—inevitably favor the violence monopoly, the concentrated authority, the singular judgment flowing from centralized power?

The answer determines whether natality—the capacity to begin anew—remains possible in our architectural choices.