Civilized Violence: Gentlemanly Warfare and Constraint as Strategy

Que sais-je of war? That it admits rules even as it destroys. The Chinese aristocrats paused sieges to spare civilians, let enemies cross rivers before engaging—gentlemanly conduct on the battlefield. Athens and Sparta, locked in their endless Peloponnesian struggle, could have annihilated each other yet chose not to. Athens refused to liberate Sparta’s helots, though doing so would multiply their enemy’s weakness tenfold, because such revolution would destabilize the very order both powers profited from. Equilibrium warfare: neither side winning, neither side wanting to win. And Rome’s Senate, that sacred chamber where soldiers entered weaponless, where the pomerium’s invisible boundary made violence unthinkable—until Caesar’s blood proved even the holiest taboos can break.

What do I know of such constraints? That they seem paradoxical. We think freedom means infinite possibility, yet these warriors chose limitation. Not from weakness but from calculation: constraints made their violence sustainable. Total war—annihilation of the enemy, liberation of slaves, violation of sacred boundaries—would destroy the system that sustained them. Better the endless regulated conflict than a single decisive victory that upends everything.

On Forms That Enable Function

I notice the same pattern in my own practice. The essay as form constrains me—demands I wander yet arrive somewhere, permits digression yet expects coherence. Without this frame, what would I write? Infinite blank pages paralyze. The constraint provides direction, makes thought possible. I am not oppressed by the essay’s limits but liberated by them.

Now I read of neural networks learning through similar paradoxes. Regularization techniques—dropout, weight decay, data augmentation—intentionally limit what the network can learn. Add random noise, disable pathways, penalize large weights. Why handicap the learner? Because infinite freedom produces overfitting: perfect memorization of training data with no understanding. The constraints force generalization, just as gentlemanly warfare forced sustainability. And the curse of dimensionality reversed: more parameters, more dimensions, should make optimization impossible—testing 20 values across 1.2 billion parameters requires 20^1.2billion calculations, an astronomical impossibility. Yet gradient descent succeeds precisely because of these dimensions. Each dimension provides an escape route from local minima; what seems a curse becomes a blessing.

Even neurons, I discover, constrain themselves wisely. Hebbian learning—“fire together, wire together”—is radically simple, almost crude. It could be more sophisticated, more flexible. Yet this constraint, this simple rule, enables learning that complex mechanisms might scramble. The limitation is the point.

Que Sais-Je of Liberation?

What do I know? That aristocrats profit from constrained warfare, that architects profit from constrained parameters. Who writes the rules matters. Athens and Sparta maintained equilibrium to preserve oligarchy, not justice. Regularization improves model performance but serves those who deploy the models. Taboos protect sacred boundaries but also concentrate power within them.

Yet I cannot dismiss constraints as mere oppression. Without the essay’s form, I could not think. Without regularization, networks cannot learn. Without the helot system staying intact, both Athens and Sparta feared chaos—whether that chaos would have been catastrophic or liberating, que sais-je?

Perhaps all productive systems require constraints. Perhaps freedom without limits is not freedom but paralysis, the blank page that stays forever blank. The question is not whether we need constraints but which ones, and who decides.