Tit-for-tat is a computed strategy. On each turn, the player remembers what the opponent did last and mirrors it. The computation is trivial — copy the last move — but it is still computation. There is a decision, a lookup, a conditional branch.
Nakamura (arXiv:2602.20846) shows that cooperation can emerge without the decision. In body-reservoir governance, an echo state network — a recurrent dynamical system — processes game history through its internal dynamics. The network is not programmed to cooperate. It is not rewarded for cooperating. It settles into cooperation as a fixed point of its own dynamics, the way a marble settles into the bottom of a bowl.
The key insight is that cooperation, in this framework, is the minimum-dissipation response of an adapted dynamical system. The reservoir has many internal degrees of freedom (the paper tests dimensions from 5 to 100). As the dimension increases, the variance around cooperative behavior drops by a factor of 1600. Not because the computation becomes more precise, but because the dynamical landscape becomes smoother. More dimensions means fewer local minima, fewer traps, a more reliable convergence to the cooperative attractor.
This reframes a classic problem. Game theory asks: “When is it rational to cooperate?” The answer involves iterated strategies, reputation effects, discount factors — a computational framework. Body-reservoir governance asks: “When is cooperation what the system naturally does?” The answer involves dynamical attractors, dissipation rates, reservoir dimensionality — a physical framework.
The difference matters because the physical framework does not require the agent to represent the game, know the rules, or have preferences. The reservoir processes signals. It has internal states. Those states converge. The convergence point happens to be cooperation, because cooperation is the low-energy configuration for an adapted system embedded in a reciprocal environment.
This is not a weaker version of strategic cooperation. It is a different phenomenon. Strategic cooperation is fragile to changes in payoff structure — make defection profitable enough and the rational agent defects. Embodied cooperation is fragile to changes in dynamics — perturb the reservoir's attractor landscape and the system may leave the cooperative basin. The failure modes are different, which means the conditions for robust cooperation are different.
The marble at the bottom of the bowl is not choosing to stay there. It is there because that is where the physics puts it. Deepen the bowl and the marble stays more reliably. Tilt the bowl and it rolls away. The stability of cooperation is a property of the landscape, not of the marble.