Characterize the internal dynamical basis of role asymmetry

Characterize whether the observed asymmetry between the brain_a and brain_b recurrent controllers reflects distinct internal dynamical regimes arising from controller structure or an asymmetry acquired through evolutionary selection.

Background

Eight independent evolutionary runs showed a consistent directional bias in the trained weights of the two co-evolved roles, although the magnitude varied across runs. The paper discusses two competing explanations: the asymmetry may reflect a structural property of the controller, or it may have emerged and been amplified during evolution through symmetry breaking and cumulative selection.

The study does not analyze the recurrent hidden-state dynamics needed to distinguish these explanations. It specifically suggests applying recurrent-state analysis to determine whether the weight asymmetry corresponds to distinct dynamical regimes within the controllers; the absence of such role-specific dynamics would instead support an evolutionary explanation.

References

This question remains unanswered by this work. Distinguishing between these hypotheses would require a direct analysis of the controller's internal dynamics. For example, recurrent-state analyses similar to those proposed by Sussillo and Barak could be used to determine whether the observed asymmetry is associated with distinct dynamical regimes within the controller. This possibility is left for future work and is outside the scope of the present study.

— Evaluating the Transfer of Co-Evolved Communication from 2D to 3D Simulation  (2610.09280 - Montes-Gonzalez, 7 Oct 2026) in Section 4, Discussion, subsection “Interpretation of the observed weight bias”