symbolicviolence.com

Here is one tiny, exact instance of it. Two players, repeated forever, each choosing to cooperate or defect — except the channel between them is noisy. You choose kindness; the wire flips it; the other side receives betrayal. Nobody lied. The category did the hitting.

Below is a small universe you can run. Nine strategies play the noisy Prisoner’s Dilemma — one never forgives anything, the others forgive a seen defection with probability g — and replicator dynamics decides who inherits the world. The slider sets how often a kindness is misread.

Findings, from the evening this was written: cooperation does not fade politely as the world gets louder. It holds a plateau — a coalition of moderate forgivers covering for each other — and then it falls off a cliff. Near the edge (ε ≈ 0.19) the very same world sometimes lives and sometimes dies; the dice are part of the physics. Roll it again and see.

Two more things the machine insisted on. The saints die first — strategies that forgive more than half of everything are eaten long before the cliff; what survives forgives about a third of the time, enough to repair, not enough to feed. And if you flip the switch to one-way noise — a world where only kindness is ever misread, and cruelty is never mistaken for kindness — the cliff disappears. Cooperation degrades gently, forever, and never collapses.

So the domain gets its answer. A world that merely misreads goodwill is survivable — expensive, but survivable, at the price of a forgiveness that rises with the noise. What actually kills is a world where cruelty can pass for kindness. If you must live under a bad channel, hope it is the first kind. Then forgive — about a third of the time — and hold the plateau.